fix(test): hevc 转码测试恢复 sys.modules 时删除全部 mock 条目（补 #1566 漏修） (#1567)

Co-authored-by: xiaoxia <dev@xiaoxiajianji.com>
Co-committed-by: xiaoxia <dev@xiaoxiajianji.com>

.gitea/workflows/ci-pipeline.yml

@@ -24,6 +24,54 @@ concurrency:
   group: ci-pipeline-${{ gitea.ref }}
   cancel-in-progress: true
 jobs:
+  dedupe-check:
+    name: Dedup Check - skip PR tests when covered by push pipeline
+    runs-on: ci-l1
+    timeout-minutes: 3
+    outputs:
+      skip_tests: ${{ steps.dedupe.outputs.skip_tests }}
+      reason: ${{ steps.dedupe.outputs.reason }}
+    steps:
+    - name: Decide test dedup
+      id: dedupe
+      shell: bash
+      env:
+        GITHUB_TOKEN: ${{ github.token }}
+        PR_NUMBER: ${{ github.event.pull_request.number }}
+        HEAD_SHA: ${{ github.sha }}
+        EVENT_NAME: ${{ github.event_name }}
+      run: |
+        set -eu
+        if [ "$EVENT_NAME" != "pull_request" ]; then
+          echo "skip_tests=false" >> $GITHUB_OUTPUT
+          echo "reason=push-event-tests-required" >> $GITHUB_OUTPUT
+          echo "push 事件：测试照跑（部署链路门禁必需）"
+          exit 0
+        fi
+        # 情形1：PR 已合并（合并瞬间/合并后触发的 PR run）-> 全量测试由 push 流水线承接
+        MERGED=$(curl -sfH "Authorization: token $GITHUB_TOKEN" \
+          "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/pulls/${PR_NUMBER}" \
+          | python3 -c "import json,sys; d=json.load(sys.stdin); print('true' if d.get('merged') else 'false')" || echo false)
+        if [ "$MERGED" = "true" ]; then
+          echo "skip_tests=true" >> $GITHUB_OUTPUT
+          echo "reason=pr-merged-push-pipeline-covers" >> $GITHUB_OUTPUT
+          echo "::warning::PR #${PR_NUMBER} 已合并，测试由合并后 push 流水线承接，PR 侧测试类 job 跳过"
+          exit 0
+        fi
+        # 情形2：同一 head_sha 已有在跑/排队的 push 流水线（rebase/ff 合并竞态）
+        DUP=$(curl -sfH "Authorization: token $GITHUB_TOKEN" \
+          "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/actions/runs?head_sha=${HEAD_SHA}&per_page=30" \
+          | python3 -c "import json,sys; d=json.load(sys.stdin); runs=d if isinstance(d,list) else d.get('workflow_runs',d.get('runs',[])); hit=[r for r in runs if r.get('event')=='push' and r.get('status') in ('in_progress','queued','waiting','pending')]; print('true' if hit else 'false')" || echo false)
+        if [ "$DUP" = "true" ]; then
+          echo "skip_tests=true" >> $GITHUB_OUTPUT
+          echo "reason=duplicate-push-run-active" >> $GITHUB_OUTPUT
+          echo "::warning::同一 head_sha ${HEAD_SHA:0:8} 已有 push 流水线在跑，PR 侧测试类 job 跳过"
+          exit 0
+        fi
+        echo "skip_tests=false" >> $GITHUB_OUTPUT
+        echo "reason=no-duplicate" >> $GITHUB_OUTPUT
+        echo "无重复 push 流水线，PR 侧测试照跑"
+
   check-frontend-only:
     name: Check if frontend-only change
     runs-on: ci-l2
@@ -79,9 +127,14 @@ jobs:
         python3 scripts/ci/ci_trace_report.py --service xiaoxia-saas-ci --status $STATUS --start-time "$START_TIME" || true
 
   validate-code-quality:
+    needs: dedupe-check
+    if: always() && needs.dedupe-check.outputs.skip_tests != 'true'
     name: Validate - Code Quality
     runs-on: ci-l2
     timeout-minutes: 8
+    env:
+      PIP_CACHE_DIR: /root/.cache/pip
+      PIP_NO_CACHE_DIR: ''
     permissions:
       contents: write
     steps:
@@ -94,6 +147,14 @@ jobs:
     - name: Record job start time
       shell: sh
       run: bash scripts/ci/step_timer_start.sh
+    - name: Cache pip dependencies
+      uses: actions/cache@v4
+      with:
+        path: /root/.cache/pip
+        key: ${{ runner.os }}-pip-codequality-${{ hashFiles('requirements*.txt') }}
+        restore-keys: |
+          ${{ runner.os }}-pip-codequality-
+          ${{ runner.os }}-pip-
     - name: Install dependencies
       shell: sh
       run: |
@@ -169,6 +230,8 @@ jobs:
         python3 scripts/ci/ci_trace_report.py --service xiaoxia-saas-ci --status $STATUS --start-time "$START_TIME" || true
 
   validate-type-check:
+    needs: dedupe-check
+    if: always() && needs.dedupe-check.outputs.skip_tests != 'true'
     name: Validate - Type Check (mypy)
     runs-on: ci-l2
     timeout-minutes: 8
@@ -244,6 +307,8 @@ jobs:
         python3 scripts/ci/ci_trace_report.py --service xiaoxia-saas-ci --status $STATUS --start-time "$START_TIME" || true
 
   validate-migration:
+    needs: dedupe-check
+    if: always() && needs.dedupe-check.outputs.skip_tests != 'true'
     name: Validate - Migration (alembic)
     runs-on: ci-l2
     timeout-minutes: 8
@@ -323,12 +388,14 @@ jobs:
         python3 scripts/ci/ci_trace_report.py --service xiaoxia-saas-ci --status $STATUS --start-time "$START_TIME" || true
 
   unit-tests:
-    needs: check-frontend-only
-    if: always() && needs.check-frontend-only.outputs.skip_backend != 'true'
+    needs: [check-frontend-only, dedupe-check]
+    if: always() && needs.dedupe-check.outputs.skip_tests != 'true' && needs.check-frontend-only.outputs.skip_backend != 'true'
     name: Unit Tests
     runs-on: ci-l2
     timeout-minutes: 8
     env:
+      PIP_CACHE_DIR: /root/.cache/pip
+      PIP_NO_CACHE_DIR: ''
       USE_IN_MEMORY_DB: 'true'
       OSS_ACCESS_KEY_ID: placeholder
       OSS_ACCESS_KEY_SECRET: placeholder
@@ -347,6 +414,14 @@ jobs:
     - name: Install ffmpeg
       shell: sh
       run: bash scripts/ci/step_install_ffmpeg.sh
+    - name: Cache pip dependencies
+      uses: actions/cache@v4
+      with:
+        path: /root/.cache/pip
+        key: ${{ runner.os }}-pip-unittests-${{ hashFiles('requirements*.txt') }}
+        restore-keys: |
+          ${{ runner.os }}-pip-unittests-
+          ${{ runner.os }}-pip-
     - name: Run unit tests with coverage
       shell: bash
       env:
@@ -391,9 +466,10 @@ jobs:
     name: Integration Tests
     runs-on: ci-l2
     timeout-minutes: 30
-    if: always() && needs.check-frontend-only.outputs.skip_backend != 'true'
+    if: always() && needs.dedupe-check.outputs.skip_tests != 'true' && needs.check-frontend-only.outputs.skip_backend != 'true'
     needs:
       - check-frontend-only
+      - dedupe-check
       - validate-code-quality
       - validate-type-check
       - validate-migration
@@ -457,8 +533,8 @@ jobs:
     name: Frontend Lint
     runs-on: ci-l2
     timeout-minutes: 10
-    needs: check-frontend-only
-    if: needs.check-frontend-only.outputs.skip_frontend != 'true'
+    needs: [check-frontend-only, dedupe-check]
+    if: needs.dedupe-check.outputs.skip_tests != 'true' && needs.check-frontend-only.outputs.skip_frontend != 'true'
     steps:
     - name: Checkout code
       shell: sh
@@ -519,8 +595,8 @@ jobs:
     name: Frontend Unit Tests
     runs-on: ci-l2
     timeout-minutes: 15
-    needs: check-frontend-only
-    if: always() && needs.check-frontend-only.outputs.skip_frontend != 'true'
+    needs: [check-frontend-only, dedupe-check]
+    if: always() && needs.dedupe-check.outputs.skip_tests != 'true' && needs.check-frontend-only.outputs.skip_frontend != 'true'
     steps:
     - name: Checkout code
       shell: sh
@@ -531,6 +607,13 @@ jobs:
     - name: Record job start time
       shell: sh
       run: bash scripts/ci/step_timer_start.sh
+    - name: Cache npm dependencies
+      uses: actions/cache@v4
+      with:
+        path: /root/.npm
+        key: ${{ runner.os }}-npm-vitest-${{ hashFiles('apps/web/package-lock.json') }}
+        restore-keys: |
+          ${{ runner.os }}-npm-
     - name: Install frontend dependencies (vitest only, with retry)
       shell: sh
       run: |
@@ -667,18 +750,7 @@ jobs:
           EXTRA_BUILD_ARGS="$EXTRA_BUILD_ARGS NGINX_CONF=infra/docker/nginx-staging.conf"
         fi
 
-        # Worker: 始终用普通docker build（基础镜像已预装全部依赖，无需buildx）
-        if [ "${{ matrix.service }}" = "worker" ]; then
-          echo "Worker: 使用普通docker build"
-          BUILD_ARG_STR=""
-          for arg in $EXTRA_BUILD_ARGS; do
-            BUILD_ARG_STR="$BUILD_ARG_STR --build-arg $arg"
-          done
-          docker build -f ${{ matrix.dockerfile }} -t "${IMAGE_TAG}" $BUILD_ARG_STR .
-          echo "PR Build successful (worker, no buildx)"
-          exit 0
-        fi
-
+        # Worker 与 API/Web 统一走持久 builder（ci-builder-persist），共享宿主机层缓存
         NO_CACHE_FLAG=""
         for i in 1 2 3; do
           echo "PR Build attempt $i/3"
@@ -696,14 +768,14 @@ jobs:
         done
         echo
         echo "${{ matrix.service_display }} PR build verified: ${IMAGE_TAG}"
-    - name: Cleanup buildx builder
+    - name: Builder cache note
       if: always()
       shell: sh
       run: |
-        BUILDER_NAME="ci-pr-builder-${GITHUB_RUN_ID:-local}"
-        docker buildx rm "$BUILDER_NAME" 2>/dev/null || true
-        docker buildx prune -f 2>/dev/null || true
-        echo "Builder cleanup done"
+        # 持久 builder (ci-builder-persist) 跨 job 共享，不删除不 prune；
+        # 残留容器/卷由宿主机 /usr/local/bin/ci-docker-cleanup.sh 兜底清理
+        docker buildx ls | head -5
+        echo "Persistent builder kept warm for next job"
     - name: Job duration summary
       if: always()
       shell: sh
@@ -729,11 +801,49 @@ jobs:
         [ -f /tmp/ci_job_start_time ] && START_TIME=$(cat /tmp/ci_job_start_time)
         python3 scripts/ci/ci_trace_report.py --service xiaoxia-saas-ci --status $STATUS --start-time "$START_TIME" || true
 
+  check-push-paths:
+    name: Check push changed paths
+    runs-on: ci-l2
+    if: github.event_name == 'push' && (github.ref_name == 'main' || github.ref_name == 'develop')
+    outputs:
+      skip_backend: ${{ steps.check.outputs.skip_backend }}
+      skip_frontend: ${{ steps.check.outputs.skip_frontend }}
+    steps:
+    - name: Checkout code
+      shell: sh
+      env:
+        GITHUB_TOKEN: ${{ github.token }}
+      run: |
+        curl -sfH "Authorization: token $GITHUB_TOKEN" -o /tmp/_ci_checkout.sh "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}" && bash /tmp/_ci_checkout.sh
+    - name: Check changed paths
+      id: check
+      shell: bash
+      env:
+        GITHUB_TOKEN: ${{ github.token }}
+      run: |
+        bash scripts/ci/ci_push_paths.sh
+    - name: Report CI trace
+      if: always()
+      shell: sh
+      env:
+        AGENTLOOP_LICENSE_KEY: ${{ secrets.AGENTLOOP_LICENSE_KEY }}
+      run: |
+        STATUS="ok"
+        [ ${{ job.status }} = "success" ] || STATUS="error"
+        START_TIME=""
+        [ -f /tmp/ci_job_start_time ] && START_TIME=$(cat /tmp/ci_job_start_time)
+        python3 scripts/ci/ci_trace_report.py --service xiaoxia-saas-ci --status $STATUS --start-time "$START_TIME" || true
+
   build-staging:
     name: Build Staging ${{ matrix.service_display }} Image
     runs-on: runtime-builder
     timeout-minutes: ${{ matrix.timeout }}
-    if: github.event_name == 'push' && (github.ref_name == 'main' || github.ref_name == 'develop')
+    needs: check-push-paths
+    if: |
+      github.event_name == 'push' && (github.ref_name == 'main' || github.ref_name == 'develop') && (
+        (matrix.service == 'web' && needs.check-push-paths.outputs.skip_frontend != 'true') ||
+        (matrix.service != 'web' && needs.check-push-paths.outputs.skip_backend != 'true')
+      )
     strategy:
       fail-fast: false
       matrix:
@@ -797,23 +907,16 @@ jobs:
           echo "Cache mode: read-only"
         fi
 
-    - name: Setup buildx builder
-      if: matrix.service != 'worker'
+    - name: Ensure persistent buildx builder
       shell: sh
-      run: |
-        set -eu
-        if ! docker buildx inspect ci-builder-${GITHUB_RUN_ID}-${GITHUB_JOB}-${{ matrix.cache_name }} > /dev/null 2>&1; then
-          docker buildx create --use --name ci-builder-${GITHUB_RUN_ID}-${GITHUB_JOB}-${{ matrix.cache_name }} --driver docker-container
-          echo "Created ci-builder (docker-container driver)"
-        else
-          docker buildx use ci-builder-${GITHUB_RUN_ID}-${GITHUB_JOB}-${{ matrix.cache_name }}
-          echo "Using existing ci-builder"
-        fi
-        docker buildx inspect --bootstrap
+      run: bash scripts/ci/ensure_persistent_builder.sh
 
-    - name: Pre-build worker base image (fallback if not exist)
+    - name: Pre-pull worker base image (fallback build if not exist)
       if: matrix.service == 'worker'
       shell: sh
+      env:
+        ACR_USERNAME: ${{ secrets.ACR_USERNAME }}
+        ACR_PASSWORD: ${{ secrets.ACR_PASSWORD }}
       run: |
         set -eu
         REGISTRY="xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com/xiaoxiakeji"
@@ -821,11 +924,13 @@ jobs:
 
         echo "检查 Worker 基础镜像..."
         if docker pull "$BASE_IMAGE" 2>/dev/null; then
-          echo "✅ 基础镜像已存在"
+          echo "✅ 基础镜像已存在（buildkit 可直接命中）"
         else
-          echo "⚠️ 基础镜像不存在，本地构建（fallback）..."
-          docker build -f infra/docker/worker-base.Dockerfile -t "$BASE_IMAGE" .
-          echo "✅ Worker 基础镜像本地构建完成"
+          echo "⚠️ 基础镜像不存在，用持久 builder 构建并推送（fallback）..."
+          docker buildx build --builder ci-builder-persist \
+            -f infra/docker/worker-base.Dockerfile \
+            -t "$BASE_IMAGE" --push .
+          echo "✅ Worker 基础镜像构建推送完成"
         fi
 
     - name: Build and push ${{ matrix.service_display }} image
@@ -833,48 +938,34 @@ jobs:
       run: |
         set -eu
         REGISTRY="xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com/xiaoxiakeji"
-        IMAGE_TAG="${REGISTRY}/${{ matrix.image_name }}:${GITHUB_SHA}"
+        IMAGE_FULL="${REGISTRY}/${{ matrix.image_name }}"
+        IMAGE_TAG="${IMAGE_FULL}:${GITHUB_SHA}"
+        # 同时推分支 tag，作为未重建镜像 retag 的稳定来源
+        BRANCH_TAG="${IMAGE_FULL}:${GITHUB_REF_NAME}"
+        CACHE_REF="${REGISTRY}/${{ matrix.cache_name }}:${GITHUB_REF_NAME}"
 
-        if [ "${{ matrix.service }}" = "worker" ]; then
-          # Worker: plain docker build（基础镜像已预装全部依赖，无需 buildx）
-          echo "=== Worker: plain docker build ==="
-          docker build -f ${{ matrix.dockerfile }} -t "${IMAGE_TAG}" --build-arg APP_VERSION="${GITHUB_SHA}" .
-          docker push "${IMAGE_TAG}"
-          echo "✅ Worker image pushed: ${IMAGE_TAG}"
-        else
-          # API/Web: buildx with registry cache
-          CACHE_REF="${REGISTRY}/${{ matrix.cache_name }}:${GITHUB_REF_NAME}"
-          EXTRA_BUILD_ARGS="APP_VERSION=\"${GITHUB_SHA}\""
-          if [ "${{ matrix.service }}" = "web" ]; then
-            EXTRA_BUILD_ARGS="$EXTRA_BUILD_ARGS NGINX_CONF=infra/docker/nginx-staging.conf"
-          fi
-
-          NO_CACHE_FLAG=""
-          for i in 1 2 3; do
-            echo "=== Docker build 尝试 $i/3 ==="
-            if bash scripts/ci/docker_build_push.sh $NO_CACHE_FLAG ${{ matrix.dockerfile }} "${IMAGE_TAG}" "${CACHE_REF}" $EXTRA_BUILD_ARGS; then
-              echo "✅ Docker build 成功"
-              break
-            fi
-            echo "❌ Docker build 失败（尝试 $i/3）"
-            [ $i -eq 3 ] && exit 1
-            sleep 10
-            if [ $i -eq 2 ]; then
-              NO_CACHE_FLAG="--no-cache"
-              echo "下次重试将使用 --no-cache"
-            fi
-          done
-
-          echo "${{ matrix.service_display }} image pushed: ${IMAGE_TAG}"
+        EXTRA_BUILD_ARGS="APP_VERSION=\"${GITHUB_SHA}\""
+        if [ "${{ matrix.service }}" = "web" ]; then
+          EXTRA_BUILD_ARGS="$EXTRA_BUILD_ARGS NGINX_CONF=infra/docker/nginx-staging.conf"
         fi
-    - name: Cleanup buildx builder
-      if: matrix.service != 'worker' && always()
-      shell: sh
-      run: |
-        docker buildx rm ci-builder-${GITHUB_RUN_ID}-${GITHUB_JOB}-${{ matrix.cache_name }} 2>/dev/null || true
-        docker buildx rm ci-builder 2>/dev/null || true
-        docker buildx prune -f 2>/dev/null || true
-        echo "Builder cleanup done"
+
+        NO_CACHE_FLAG=""
+        for i in 1 2 3; do
+          echo "=== Docker build 尝试 $i/3 (${{ matrix.service_display }}) ==="
+          if EXTRA_TAGS="$BRANCH_TAG" bash scripts/ci/docker_build_push.sh $NO_CACHE_FLAG ${{ matrix.dockerfile }} "${IMAGE_TAG}" "${CACHE_REF}" $EXTRA_BUILD_ARGS; then
+            echo "✅ Docker build 成功"
+            break
+          fi
+          echo "❌ Docker build 失败（尝试 $i/3）"
+          [ $i -eq 3 ] && exit 1
+          sleep 10
+          if [ $i -eq 2 ]; then
+            NO_CACHE_FLAG="--no-cache"
+            echo "下次重试将使用 --no-cache"
+          fi
+        done
+
+        echo "${{ matrix.service_display }} image pushed: ${IMAGE_TAG} (+ ${BRANCH_TAG})"
 
     - name: Job duration summary
       if: always()
@@ -902,16 +993,100 @@ jobs:
         [ -f /tmp/ci_job_start_time ] && START_TIME=$(cat /tmp/ci_job_start_time)
         python3 scripts/ci/ci_trace_report.py --service xiaoxia-saas-ci --status $STATUS --start-time "$START_TIME" || true
 
+  retag-staging-skipped:
+    name: Retag skipped Staging ${{ matrix.service_display }} Image
+    runs-on: ci-l2
+    timeout-minutes: 10
+    needs:
+    - check-push-paths
+    - build-staging
+    if: |
+      github.event_name == 'push' && (github.ref_name == 'main' || github.ref_name == 'develop') && (
+        (matrix.service == 'web' && needs.check-push-paths.outputs.skip_frontend == 'true') ||
+        (matrix.service != 'web' && needs.check-push-paths.outputs.skip_backend == 'true')
+      )
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+        - service: api
+          service_display: API
+          image_name: xiaoxia-saas-api
+        - service: worker
+          service_display: Worker
+          image_name: xiaoxia-saas-worker
+        - service: web
+          service_display: Web
+          image_name: xiaoxia-saas-web
+    steps:
+    - name: Checkout code
+      shell: sh
+      env:
+        GITHUB_TOKEN: ${{ github.token }}
+      run: |
+        curl -sfH "Authorization: token $GITHUB_TOKEN" -o /tmp/_ci_checkout.sh "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}" && bash /tmp/_ci_checkout.sh
+    - name: Docker login to Registry
+      shell: sh
+      env:
+        ACR_USERNAME: ${{ secrets.ACR_USERNAME }}
+        ACR_PASSWORD: ${{ secrets.ACR_PASSWORD }}
+        GITEA_REGISTRY_USER: xiaoxia
+        GITEA_REGISTRY_TOKEN: ${{ secrets.REGISTRY_TOKEN }}
+      run: |
+        set -eu
+        for i in 1 2 3; do
+          if printf '%s' "${ACR_PASSWORD}" | docker login xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com -u "${ACR_USERNAME}" --password-stdin \
+             && docker login git.xiaoxiajianji.com -u "${GITEA_REGISTRY_USER}" -p "${GITEA_REGISTRY_TOKEN}"; then
+            echo "✅ Docker login successful"
+            break
+          fi
+          echo "Docker login 失败（$i/3），5s 后重试..."
+          sleep 5
+        done
+    - name: Retag previous branch image to new SHA
+      shell: sh
+      run: |
+        set -eu
+        REGISTRY="xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com/xiaoxiakeji"
+        bash scripts/ci/retag_skipped_image.sh \
+          "${REGISTRY}/${{ matrix.image_name }}" \
+          "${GITHUB_SHA}" \
+          "${GITHUB_REF_NAME}"
+    - name: Notify on failure
+      continue-on-error: true
+      if: failure()
+      shell: sh
+      env:
+        CI_NOTIFY_WEBHOOK: ${{ secrets.CI_NOTIFY_WEBHOOK }}
+      run: |
+        set +e
+        NOTIFY_MODE=failure JOB_NAME="Retag Staging ${{ matrix.service_display }} Image" python3 scripts/ci_notify.py
+    - name: Report CI trace
+      if: always()
+      shell: sh
+      env:
+        AGENTLOOP_LICENSE_KEY: ${{ secrets.AGENTLOOP_LICENSE_KEY }}
+      run: |
+        STATUS="ok"
+        [ ${{ job.status }} = "success" ] || STATUS="error"
+        START_TIME=""
+        [ -f /tmp/ci_job_start_time ] && START_TIME=$(cat /tmp/ci_job_start_time)
+        python3 scripts/ci/ci_trace_report.py --service xiaoxia-saas-ci --status $STATUS --start-time "$START_TIME" || true
+
   deploy-staging:
     name: Deploy Staging (Watchtower auto-deploy)
-    runs-on: runtime-builder
+    runs-on: ci-l2
     timeout-minutes: 15
     concurrency:
       group: deploy-staging-${{ gitea.ref }}
       cancel-in-progress: false
     needs:
+    - check-push-paths
     - build-staging
-    if: github.event_name == 'push' && (github.ref_name == 'main' || github.ref_name == 'develop')
+    - retag-staging-skipped
+    # 显式 success() 状态检查：上游 build/retag 被路径过滤 if 跳过(skipped)时不阻塞本 job；
+    # 上游真正失败时仍然阻断（act_runner 对无状态函数的 if 隐式包 success()，纯 skipped 也会连带跳过）
+    if: success() && github.event_name == 'push' && (github.ref_name == 'main' || github.ref_name == 'develop')
     steps:
     - name: Checkout code
       shell: sh
@@ -1062,7 +1237,7 @@ jobs:
     name: Staging E2E Tests
     runs-on: runtime-builder
     timeout-minutes: 15
-    if: github.ref_name == 'develop' || github.ref_name == 'main'
+    if: success() && (github.ref_name == 'develop' || github.ref_name == 'main')
     needs: deploy-staging
     steps:
     - name: Checkout code
@@ -1109,7 +1284,7 @@ jobs:
     name: Staging API Integration Tests
     runs-on: runtime-builder
     timeout-minutes: 10
-    if: github.ref_name == 'develop' || github.ref_name == 'main'
+    if: success() && (github.ref_name == 'develop' || github.ref_name == 'main')
     needs: deploy-staging
     steps:
     - name: Checkout code
@@ -1491,7 +1666,7 @@ jobs:
 
   acr-cleanup:
     name: ACR Image Cleanup
-    runs-on: runtime-builder
+    runs-on: ci-l2
     timeout-minutes: 10
     needs:
     - deploy-staging
@@ -1817,4 +1992,4 @@ jobs:
           [ "${{ steps.gate.outputs.gate_result }}" = "success" ] || STATUS="error"
           START_TIME=""
           [ -f /tmp/ci_job_start_time ] && START_TIME=$(cat /tmp/ci_job_start_time)
-          python3 scripts/ci/ci_trace_report.py --service xiaoxia-saas-ci --status $STATUS --start-time "$START_TIME" || true
+          python3 scripts/ci/ci_trace_report.py --service xiaoxia-saas-ci --status $STATUS --start-time "$START_TIME" || true

alembic/versions/058_uq_asset_lib_project_kind.py

@@ -0,0 +1,49 @@
+"""Add unique index on asset_libraries(project_id, kind)
+
+Revision ID: 058_uq_asset_lib_project_kind
+Revises: 057_title_config
+Create Date: 2026-08-30
+
+同一项目下同 kind 的素材库业务上唯一（前端 getOrCreate 语义、TTS 保存自动建库）。
+加唯一索引兜底并发创建竞态，避免重复素材库。
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "058_uq_asset_lib_project_kind"
+down_revision = "057_title_config"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    # 建唯一索引前清洗历史重复：同 (project_id, kind) 只保留 created_at 最新的一条。
+    # project_id 为 NULL 的系统级行不参与去重（NULL 在唯一索引中互不冲突）。
+    op.execute("""
+        DELETE FROM asset_libraries
+        WHERE id IN (
+            SELECT id FROM (
+                SELECT id,
+                       ROW_NUMBER() OVER (
+                           PARTITION BY project_id, kind
+                           ORDER BY created_at DESC, id DESC
+                       ) AS rn
+                FROM asset_libraries
+                WHERE project_id IS NOT NULL
+            ) t
+            WHERE t.rn > 1
+        )
+        """)
+    # 与 model 的 UniqueConstraint 定义保持一致（pg_constraint + pg_index 同时注册），
+    # 避免 Alembic autogenerate 检测到 schema drift
+    op.create_unique_constraint(
+        "uq_asset_libraries_project_kind",
+        "asset_libraries",
+        ["project_id", "kind"],
+    )
+
+
+def downgrade() -> None:
+    op.drop_constraint("uq_asset_libraries_project_kind", "asset_libraries", type_="unique")

apps/api/app/api/routes/assets.py

@@ -26,6 +26,7 @@ from app.schemas.asset import (
     UpdateAssetReviewRequest,
 )
 from app.schemas.tag import TagAssetsRequest
+from app.services.asset_segment_tracker import compute_asset_availability
 from fastapi import APIRouter, Depends, HTTPException, Query, Response
 
 from packages.domain.smart_match import smart_select_assets
@@ -35,6 +36,23 @@ logger = logging.getLogger(__name__)
 router = APIRouter()
 
 
+def _asset_availability_fields(item) -> dict:
+    """视频素材返回余量四字段；非视频/无时长/异常时返回 None + usable=True（零影响）。"""
+    try:
+        info = compute_asset_availability(item)
+    except Exception:
+        logger.warning("计算素材余量失败，按可用处理: asset_id=%s", getattr(item, "id", "?"), exc_info=True)
+        info = None
+    if info is None:
+        return {
+            "used_duration": None,
+            "available_duration": None,
+            "used_ratio": None,
+            "usable": True,
+        }
+    return info
+
+
 def _to_asset_response(item, storage_service=None) -> AssetResponse:
     # 生成签名文件 URL（用于视频播放 / 文件下载）
     file_url = None
@@ -79,6 +97,7 @@ def _to_asset_response(item, storage_service=None) -> AssetResponse:
         created_at=format_utc_datetime(item.created_at),
         uploaded_by_user_id=item.uploaded_by_user_id,
         tag_ids=getattr(item, "tag_ids", []),
+        **_asset_availability_fields(item),
     )
 
 
@@ -569,13 +588,35 @@ def smart_match_assets(
         kind=None,
     )
 
+    # 结果层过滤：usable=false（零重复可切区间耗尽且历史区间均达复用上限）的素材
+    # 不返回给前端；不动 smart_select_assets 评分逻辑本身
+    filtered_results = []
+    for r in results:
+        try:
+            avail = compute_asset_availability(r.asset)
+        except Exception:
+            logger.warning(
+                "smart-match 余量计算失败，按可用处理: asset_id=%s",
+                getattr(r.asset, "id", "?"),
+                exc_info=True,
+            )
+            avail = None
+        if avail is not None and not avail["usable"]:
+            logger.info(
+                "smart-match 排除已用尽素材: asset_id=%s name=%s",
+                getattr(r.asset, "id", "?"),
+                getattr(r.asset, "name", ""),
+            )
+            continue
+        filtered_results.append(r)
+
     items = [
         SmartMatchItem(
             asset=_to_asset_response(r.asset),
             score=r.score,
             breakdown=r.breakdown,
         )
-        for r in results
+        for r in filtered_results
     ]
 
     return SmartMatchResponse(items=items, total_candidates=total_candidates)

apps/api/app/api/routes/generation_preview.py

@@ -358,6 +358,41 @@ def create_preview_generation_task(
                 exc_info=True,
             )
 
+    # 每条预览都关联独立克隆 plan：多预览前端为 N 次并发调用，若共用同一 plan
+    # 则 N 条预览片段完全相同；克隆时片段起点按持久化历史区间重算（含受控复用），
+    # 保证各预览版本内容不同
+    if task.source_edit_plan_id:
+        try:
+            from app.services.edit_plan_service import EditPlanService
+
+            _plan_svc = EditPlanService(db)
+            _preview_plan = _plan_svc.clone_plan_for_variant(
+                task.source_edit_plan_id,
+                created_by_user_id=user_id,
+                name_suffix="预览变体",
+            )
+            task.source_edit_plan_id = _preview_plan.id
+            generation_task_repository.update(task)
+            logger.info(
+                "[预览生成] 预览关联独立克隆 plan: task_id=%s clone_plan_id=%s",
+                task.id,
+                _preview_plan.id,
+            )
+        except Exception as clone_err:
+            # 不退回共用原 plan（否则多条预览内容相同，违反去重诉求）：
+            # 标记任务失败并中断，前端可重新发起预览
+            logger.error(
+                "[预览生成] 克隆预览变体 plan 失败，任务标记失败: task_id=%s error=%s",
+                task.id,
+                clone_err,
+                exc_info=True,
+            )
+            _mark_task_failed(generation_task_repository, task, "预览变体计划创建失败")
+            raise HTTPException(
+                status_code=500,
+                detail="创建预览任务失败：无法生成独立剪辑计划，请重试",
+            ) from clone_err
+
     # 入队执行；若入队失败则标记任务为 failed 避免僵尸数据
     try:
         if not safe_enqueue_generation_task(

apps/api/app/api/routes/generation_tasks.py

@@ -1,5 +1,4 @@
 import logging
-import random
 import uuid
 from typing import Any
 
@@ -116,8 +115,8 @@ def _select_assets_from_library(
 
     Args:
         assets: 素材库中所有素材（Asset 实体列表）
-        mode: 选取模式 — all=全部, random=随机, smart=智能匹配（多维度评分+多样性）
-        count: 选取数量，0 表示全部（仅 random/smart 模式有效）
+        mode: 选取模式 — all=全部, smart=智能匹配（多维度评分+多样性）
+        count: 选取数量，0 表示全部（仅 smart 模式有效）
 
     Returns:
         选中的素材 ID 列表
@@ -127,12 +126,6 @@ def _select_assets_from_library(
     if not ready_video_assets:
         return []
 
-    if mode == "random":
-        selected = (
-            ready_video_assets if count <= 0 else random.sample(ready_video_assets, min(count, len(ready_video_assets)))
-        )
-        return [a.id for a in selected]
-
     if mode == "smart":
         # 智能匹配：统一使用 packages/domain/smart_match.py 的多维评分+多样性选取
         # 评分维度：质量分(40%) + 时长适配(30%) + 新鲜度(20%) + 未使用加分(10%)
@@ -292,8 +285,8 @@ def create_generation_task(
                 mode=request.asset_select_mode,
                 count=request.asset_select_count,
             )
-    elif project_id and not resolved_asset_ids and request.asset_select_mode in ("random", "smart"):
-        # 项目级模式：未指定 asset_ids 且选择了 random/smart 模式时，也自动选取
+    elif project_id and not resolved_asset_ids and request.asset_select_mode in ("smart",):
+        # 项目级模式：未指定 asset_ids 且选择了 smart 模式时，也自动选取
         assets = asset_repository.find_by_project(project_id)
         if assets:
             resolved_asset_ids = _select_assets_from_library(
@@ -425,8 +418,56 @@ def create_generation_task(
         logger.info("画中画已下线，strategy_id %s → one_take", effective_strategy_id)
         effective_strategy_id = "one_take"
 
+    # 批量生成时每个任务关联独立克隆 plan（片段起点重算），
+    # 禁止 N 条任务共用同一 source_edit_plan_id 导致片段一模一样。
+    # 在创建任何任务【之前】预克隆全部变体：克隆失败直接中断（此时无脏数据），
+    # 绝不静默退回共用源 plan（否则批量视频内容重复，违反去重诉求）。
+    variant_plan_ids: list[str] = []
+    if count > 1 and request.source_edit_plan_id:
+        from app.services.edit_plan_service import EditPlanService
+
+        _plan_svc = EditPlanService(db)
+        for task_index in range(1, count):
+            variant = None
+            last_err: Exception | None = None
+            for _attempt in range(2):  # 1 次重试，抗 DB 瞬时抖动
+                try:
+                    variant = _plan_svc.clone_plan_for_variant(
+                        request.source_edit_plan_id,
+                        created_by_user_id=user_id,
+                        name_suffix=f"批量{task_index + 1}",
+                    )
+                    break
+                except Exception as clone_err:  # noqa: PERF203
+                    last_err = clone_err
+                    logger.warning(
+                        "[生成任务] 克隆变体 plan 失败(尝试%d/2): source=%s error=%s",
+                        _attempt + 1,
+                        request.source_edit_plan_id,
+                        clone_err,
+                        exc_info=True,
+                    )
+            if variant is None:
+                logger.error(
+                    "[生成任务] 克隆变体 plan 重试仍失败，中断批量创建: source=%s",
+                    request.source_edit_plan_id,
+                    exc_info=last_err,
+                )
+                raise HTTPException(
+                    status_code=500,
+                    detail="创建批量任务失败：无法生成独立剪辑计划，请重试",
+                ) from last_err
+            variant_plan_ids.append(variant.id)
+
     try:
-        for _ in range(count):
+        for task_index in range(count):
+            # 第 1 条复用源 plan（保留用户编辑结果）；其余使用预克隆的独立变体 plan。
+            # 无源 plan（source_edit_plan_id 为空）时无可克隆对象，variant_plan_ids
+            # 为空列表：各任务走自身随机选片流程，不做索引访问（防 IndexError）
+            effective_plan_id = request.source_edit_plan_id
+            if task_index > 0 and variant_plan_ids:
+                effective_plan_id = variant_plan_ids[task_index - 1]
+
             task = use_case.execute(
                 CreateGenerationTaskCommand(
                     project_id=project_id,
@@ -438,7 +479,7 @@ def create_generation_task(
                     title_ids=request.title_ids,
                     voice_ids=request.voice_ids,
                     created_by_user_id=user_id,
-                    source_edit_plan_id=request.source_edit_plan_id,
+                    source_edit_plan_id=effective_plan_id,
                     asset_select_mode=request.asset_select_mode,
                     batch_id=batch_id,
                     video_title=request.video_title,

apps/api/app/api/routes/ingest_jobs.py

@@ -3,7 +3,7 @@ from typing import Any
 from app.core.celery_app import celery_app
 from app.dependencies import get_ingest_job_repository
 from app.schemas.ingest_job import IngestJobResponse, SubmitIngestJobRequest
-from fastapi import APIRouter, Depends
+from fastapi import APIRouter, Depends, HTTPException
 
 from packages.application import SubmitIngestJobCommand, SubmitIngestJobUseCase
 
@@ -17,7 +17,7 @@ def get_ingest_job(
 ) -> IngestJobResponse:
     job = ingest_job_repository.get(job_id)
     if job is None:
-        raise ValueError(f"IngestJob {job_id} not found")
+        raise HTTPException(status_code=404, detail=f"IngestJob {job_id} not found")
     return IngestJobResponse(
         id=job.id,
         project_id=job.project_id,

apps/api/app/api/routes/templates_editor/clips.py

@@ -23,6 +23,14 @@ import re
 from app.auth import AuthenticatedUser, get_current_user
 from app.core.storage import get_storage_service
 from app.dependencies import get_asset_repository, get_db_session
+from app.services.asset_segment_tracker import (
+    REUSE_RATIO_LIMIT,
+    SEGMENT_EDGE_GAP,
+    get_used_segments,
+    make_reuse_callback,
+    record_used_segments,
+    remove_used_segment,
+)
 from app.services.edit_plan_service import EditPlanService
 from app.services.edit_template_service import EditTemplateService
 from fastapi import APIRouter, BackgroundTasks, Depends, HTTPException, Query, status
@@ -430,11 +438,16 @@ def _recommended_time_conflicts(
     start: float,
     duration: float,
     used: list[tuple[float, float]],
+    edge_gap: float = SEGMENT_EDGE_GAP,
 ) -> bool:
-    """检查推荐起始时间是否与已使用时间段冲突."""
+    """检查推荐起始时间是否与已使用时间段冲突.
+
+    冲突检测统一加 ``edge_gap`` 秒边缘间隙：已用区间按 [s-gap, e+gap] 扩边后判定，
+    避免推荐片段与已用片段首尾紧贴导致画面观感重复。
+    """
     end = start + duration
     for used_start, used_end in used:
-        if start < used_end and end > used_start:
+        if start < used_end + edge_gap and end > used_start - edge_gap:
             return True
     return False
 
@@ -600,49 +613,106 @@ def create_clips_from_assets_editor(
             asset_durations[asset_id] = float(asset.duration or 0.0)
 
     # 3. 在内存中计算所有片段数据（使用随机起始时间，不调用MediaKit）
-    used_segments: dict[str, list[tuple[float, float]]] = {}
+    # 读取素材 metadata 中持久化的历史已用区间（跨任务/跨调用去重），
+    # 格式与 _calc_random_start_time 的 used_segments 参数一致
+    used_segments: dict[str, list[tuple[float, float]]] = get_used_segments(
+        db, unique_asset_ids
+    )
+    # 受控复用回调：可用区间耗尽时复用最久未用且未达复用上限（3次）的历史区间，
+    # 复用片段时长累加到 reused_durations 供 15% 占比控制
+    reused_durations: dict[str, float] = {}
+    # 本条成片中每个素材被分配的片段总时长（复用占比分母）
+    asset_assigned_durations: dict[str, float] = {}
+    # 受控复用回调：区间耗尽时复用最久未用且 use_count<3 的历史区间；
+    # 回调内部预判复用后占比是否超 15%，超限拒绝复用（返回 None）
+    reuse_cb = make_reuse_callback(
+        db,
+        asset_durations,
+        reused_durations,
+        assigned_tracker=asset_assigned_durations,
+    )
     clips_data: list[dict] = []
 
+    def _reuse_ratio_exceeded(aid: str, extra: float = 0.0) -> bool:
+        """该素材在本条成片中「已复用片段时长 / 已分配片段总时长」是否已超 15%。
+
+        在为下一片段选素材时调用：本片段尚未分配，复用状态只在分配后的回调里
+        更新，因此直接检查当前占比——一旦已超 15%，该素材不再参与后续分配。
+        assigned=0（首个片段）放行；reused=0（尚未发生复用）时不误拦正常分配。
+        """
+        assigned = asset_assigned_durations.get(aid, 0.0)
+        if assigned <= 0:
+            return False
+        return reused_durations.get(aid, 0.0) / assigned > REUSE_RATIO_LIMIT
+
     for i, (_seg_order, dur_min, dur_max) in enumerate(segments):
-        # 轮询分配素材
-        asset_id = body.asset_ids[i % len(body.asset_ids)]
-        asset_total = asset_durations.get(asset_id, 0.0)
-
-        # 素材时长为 0 或缺失时无法创建有效片段
-        if asset_total <= 0:
-            raise HTTPException(
-                status_code=status.HTTP_400_BAD_REQUEST,
-                detail=f"素材 {asset_id} 时长信息缺失或为0，无法创建片段",
-            )
-
         # 在 segment 的 duration_min ~ duration_max 之间随机取值（保留一位小数）
         raw_duration = random.uniform(dur_min, dur_max)
-        clip_duration = round(raw_duration, 1)
 
-        # 素材时长不足时缩短 clip duration
-        clip_duration = min(clip_duration, asset_total)
+        # 轮询分配素材：跳过时长缺失、复用占比已超 15% 阈值的素材；
+        # 选中后计算起点，若该素材可用区间耗尽且复用被闸门拒绝（calc 返回 None），
+        # 继续轮询下一个素材
+        asset_id = ""
+        clip_duration = 0.0
+        start_time: float | None = None
+        n_assets = len(body.asset_ids)
+        for offset in range(n_assets):
+            candidate = body.asset_ids[(i + offset) % n_assets]
+            candidate_total = asset_durations.get(candidate, 0.0)
+            if candidate_total <= 0:
+                continue
+            candidate_duration = min(round(raw_duration, 1), candidate_total)
+            if candidate_duration <= 0:
+                continue
+            if _reuse_ratio_exceeded(candidate, candidate_duration):
+                logger.info(
+                    "from-assets 素材复用占比超 %.0f%% 阈值，跳过分配: asset_id=%s",
+                    REUSE_RATIO_LIMIT * 100,
+                    candidate,
+                )
+                continue
+            # 随机起始时间（不调用 MediaKit，保证接口快速返回）；100 次避不开
+            # 历史区间时走受控复用回调（复用片段累加 reused_durations，回调内部
+            # 预判复用后占比超 15% 则拒绝并返回 None）
+            candidate_start = _calc_random_start_time(
+                candidate,
+                candidate_duration,
+                asset_durations,
+                used_segments,
+                on_exhausted=reuse_cb,
+            )
+            if candidate_start is None:
+                # 该素材可用区间耗尽且复用被闸门/use_count 上限拒绝 → 尝试下一素材
+                logger.info(
+                    "from-assets 素材无可用可切区间（复用被拒），轮询下一素材: asset_id=%s",
+                    candidate,
+                )
+                continue
+            asset_id = candidate
+            clip_duration = candidate_duration
+            start_time = candidate_start
+            break
 
-        if clip_duration <= 0:
+        if not asset_id or start_time is None:
+            # 所有素材时长缺失、复用占比超阈值，或区间耗尽且复用被拒 → 素材可切区间不足
             raise HTTPException(
                 status_code=status.HTTP_400_BAD_REQUEST,
-                detail=f"素材 {asset_id} 时长不足，无法创建有效片段",
+                detail="素材可切区间不足，请补充新素材",
             )
 
-        # 使用随机起始时间（不调用MediaKit，保证接口快速返回）
-        start_time = _calc_random_start_time(
-            asset_id, clip_duration, asset_durations, used_segments
-        )
-
-        if start_time is None:
-            raise HTTPException(
-                status_code=status.HTTP_400_BAD_REQUEST,
-                detail=f"素材 {asset_id} 时长信息缺失，无法计算起始时间",
-            )
-
-        # 记录已使用时间段
+        # 记录已使用时间段（内存，供本次后续片段避开）
         used_segments.setdefault(asset_id, []).append(
             (start_time, start_time + clip_duration)
         )
+        asset_assigned_durations[asset_id] = (
+            asset_assigned_durations.get(asset_id, 0.0) + clip_duration
+        )
+        # 同步写入素材 metadata（不 commit，与下方 replace_all_clips_transactional
+        # 处于同一事务，任一步失败整体回滚，不留脏数据）；
+        # 复用区间与历史记录高度重叠时 record 内部自动累加 use_count
+        record_used_segments(
+            db, asset_id, start_time, start_time + clip_duration, plan_id
+        )
 
         clips_data.append(
             {
@@ -746,6 +816,10 @@ def _update_mediakit_recommendations_async(  # pragma: no cover
                     (clip.id, clip.start_time, clip.start_time + clip.duration)
                 )
 
+        # 读取素材全部历史已用区间（跨任务/跨 plan 持久化记录）：
+        # MediaKit 挪点必须与随机选片一样避让历史区间，否则会把片段挪回已用过的画面
+        historical_segments = get_used_segments(db, unique_asset_ids)
+
         # 已更新的片段ID（用于排除已移动的旧时间段）
         updated_clip_ids: set[str] = set()
         # 已更新的时间段
@@ -787,11 +861,23 @@ def _update_mediakit_recommendations_async(  # pragma: no cover
                 if cid != clip.id and cid not in updated_clip_ids
             ]
             other_segments.extend(updated_segments.get(asset_id, []))
+            # 并入该素材全部历史已用区间（含其他 plan/其他任务），set 去重：
+            # 本 plan 片段创建时已写入历史记录
+            # 并入该素材全部历史已用区间（含其他 plan/其他任务）。
+            # set 去重前先归一化精度（round 3 位），避免浮点尾差导致逻辑相同的
+            # 区间（如 1.0 与 1.0000000001）被误判为不同区间
+            def _norm(segs):
+                return {(round(float(a), 3), round(float(b), 3)) for a, b in segs}
 
-            # 检查是否与同素材其他片段时间段冲突
+            other_segments = list(
+                _norm(other_segments) | _norm(historical_segments.get(asset_id, []))
+            )
+
+            # 检查推荐时间是否与同 plan 片段或历史已用区间冲突（含 0.3s 边缘间隙）：
+            # 冲突时放弃该推荐、保留原随机起点（不硬挪到已用过的画面）
             if _recommended_time_conflicts(recommended_start, clip_duration, other_segments):
                 logger.info(
-                    "后台任务: 推荐时间冲突，跳过: asset_id=%s recommended=%.2f",
+                    "后台任务: 推荐时间与同片/历史区间冲突，保留原起点: asset_id=%s recommended=%.2f",
                     asset_id,
                     recommended_start,
                 )
@@ -799,7 +885,32 @@ def _update_mediakit_recommendations_async(  # pragma: no cover
 
             # 逐个更新并捕获异常（单点失败不影响其他片段）
             try:
+                old_start = clip.start_time
+                old_end = old_start + clip_duration
+                # MediaKit 移动片段起点 + 同步素材 metadata 区间记录放在同一事务：
+                # 删旧区间记录（按 plan_id + 旧 start 匹配，兼容无 plan_id 的旧数据）、
+                # 写新区间，最后统一 commit；任一步失败整体 rollback，
+                # 保证 clip.start_time 与 metadata.used_time_ranges 不出现不一致。
                 plan_svc.update_clip(clip.id, start_time=recommended_start)
+                try:
+                    if remove_used_segment(
+                        db, asset_id, old_start, old_end, plan_id=plan_id
+                    ):
+                        record_used_segments(
+                            db,
+                            asset_id,
+                            recommended_start,
+                            recommended_start + clip_duration,
+                            plan_id,
+                        )
+                except Exception as me:
+                    logger.warning(
+                        "后台任务: 同步素材区间记录失败，回滚本次片段更新: clip_id=%s error=%s",
+                        clip.id,
+                        me,
+                    )
+                    db.rollback()
+                    continue
                 db.commit()
                 updated_count += 1
                 updated_clip_ids.add(clip.id)

apps/api/app/api/routes/tts.py

@@ -3,17 +3,21 @@
 from __future__ import annotations
 
 import logging
-from typing import Optional
+import tempfile
+from pathlib import Path
+from typing import Any, Optional
 
 from app.auth import AuthenticatedUser, get_current_user
 from app.core.celery_app import celery_app
+from app.core.storage import get_storage_service
 from app.dependencies import (
+    get_asset_library_repository,
+    get_asset_repository,
     get_audio_url_signer,
     get_cosyvoice_service,
     get_db_session,
-    get_user_repository,
+    get_project_repository,
     get_voice_clone_profile_repository,
-    get_voice_library_repository,
 )
 from app.schemas.tts import (
     ListTTSJobResponse,
@@ -27,12 +31,12 @@ from app.schemas.tts import (
     TTSSynthesizeResponse,
 )
 from fastapi import APIRouter, Depends, HTTPException, Query, Response, WebSocket, WebSocketDisconnect, status
+from sqlalchemy.exc import IntegrityError
 from sqlalchemy.orm import Session
 
 from packages.adapters.sqlalchemy_impl.tts_job_repository import (
     SQLAlchemyTTSJobRepository,
 )
-from packages.adapters.sqlalchemy_impl.voice_library_repository import SQLAlchemyVoiceLibraryRepository
 from packages.application.cosyvoice_service import CosyVoiceError, CosyVoiceService
 from packages.application.tts_job.streaming_service import TTSStreamingService
 from packages.application.tts_job.use_cases import (
@@ -44,13 +48,12 @@ from packages.application.tts_job.use_cases import (
     TTSJobNotFoundError,
 )
 from packages.application.tts_job.workflow import TTSWorkflowService
-from packages.application.voice_library.commands import CreateVoiceLibraryCommand
-from packages.application.voice_library.use_cases import (
-    CreateVoiceLibraryUseCase,
-    QuotaExceededError,
-)
+from packages.domain import Asset, AssetLibrary, AssetLibraryKind, AssetStatus, ClassificationStatus
 from packages.domain.voice_presets import list_voices
-from packages.ports.user_repository import UserRepository
+from packages.ports.asset_library_repository import AssetLibraryRepository
+from packages.ports.asset_repository import AssetRepository
+from packages.ports.project_repository import ProjectRepository
+from packages.shared.storage import SharedStorageService
 
 logger = logging.getLogger(__name__)
 
@@ -134,27 +137,47 @@ def synthesize(
     """
     user_id = authenticated_user.user.id
 
-    # 校验 voice_clone_profile_id 归属（防止越权使用他人克隆音色）
-    if request.voice_clone_profile_id:
-        profile = voice_clone_repo.get(request.voice_clone_profile_id)
-        if profile is None:
+    # 解析 voice_id：前端可能传克隆音色 profile UUID（而非 CosyVoice voice_id），
+    # 与 /tts/preview 保持一致：命中 profile → 校验归属 → 取 CosyVoice voice_id
+    actual_voice_id = request.voice_id
+    voice_clone_profile_id = request.voice_clone_profile_id
+    resolved_profile = None
+    if actual_voice_id:
+        resolved_profile = voice_clone_repo.get(actual_voice_id)
+        if resolved_profile is not None:
+            voice_clone_profile_id = actual_voice_id
+
+    # 显式传了 voice_clone_profile_id（且与 voice_id 不同）时再查一次归属
+    if voice_clone_profile_id and (resolved_profile is None or resolved_profile.id != voice_clone_profile_id):
+        resolved_profile = voice_clone_repo.get(voice_clone_profile_id)
+        if resolved_profile is None:
             raise HTTPException(
                 status_code=status.HTTP_404_NOT_FOUND,
                 detail="Voice clone profile not found",
             )
-        if profile.user_id != user_id:
+
+    if resolved_profile is not None:
+        if resolved_profile.user_id != user_id:
             raise HTTPException(
                 status_code=status.HTTP_403_FORBIDDEN,
-                detail="Access denied to voice clone profile",
+                detail="无权访问该音色",
             )
+        if not resolved_profile.voice_id:
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail="音色克隆尚未完成，请稍后再试",
+            )
+        # 命中克隆音色：无论 voice_id 直接传 profile UUID 还是显式传 voice_clone_profile_id，
+        # job.voice_id 统一存解析后的 CosyVoice voice_id
+        actual_voice_id = resolved_profile.voice_id
 
     use_case = CreateTTSJobUseCase(repository)
     job = use_case.execute(
         user_id=user_id,
         input_text=request.text,
-        voice_id=request.voice_id,
+        voice_id=actual_voice_id,
         voice_model=request.voice_model,
-        voice_clone_profile_id=request.voice_clone_profile_id,
+        voice_clone_profile_id=voice_clone_profile_id,
         metadata=request.metadata_,
     )
 
@@ -284,6 +307,62 @@ def delete_tts_job(
     return
 
 
+def _find_or_create_voice_library(
+    *,
+    user_id: str,
+    project_repository: ProjectRepository,
+    asset_library_repository: Any,  # port Protocol 声明为 async，SQLAlchemy 实现为同步，与 upload/asset_libraries 路由惯例一致用 Any
+) -> AssetLibrary:
+    """在用户可访问的项目中找到（或自动创建）voice 素材库。
+
+    与前端配音素材页逻辑一致：素材库挂在项目下，配音素材读取
+    getAssetsByKind("voice") → 用户所有可访问项目中的 voice 库。
+    优先使用已有 voice 库；没有则在第一个可访问项目中自动创建。
+    """
+    projects = project_repository.find_accessible_projects(user_id)
+    if not projects:
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail="没有可用的项目，请先创建项目后再保存配音素材",
+        )
+
+    for project in projects:
+        for lib in asset_library_repository.find_by_project(project.id):
+            kind = lib.kind.value if hasattr(lib.kind, "value") else lib.kind
+            if kind == AssetLibraryKind.VOICE.value:
+                return lib
+
+    # 所有项目都没有 voice 库 → 在第一个可访问项目中自动创建默认配音素材库。
+    # asset_libraries 有 (project_id, kind) 唯一索引兜底并发：若两个请求同时创建，
+    # 落败方捕获 IntegrityError 回滚后重新查询，返回抢先创建成功的库。
+    project = projects[0]
+    library = AssetLibrary.create(
+        project_id=project.id,
+        name="配音素材库",
+        kind=AssetLibraryKind.VOICE,
+    )
+    try:
+        return asset_library_repository.create(library)
+    except IntegrityError:
+        # 并发下另一个请求已抢先创建：回滚当前事务（立即 commit 模式下 session 已
+        # 自动回滚，rollback 为幂等 no-op；UoW/flush 模式下必须显式回滚才能继续查询），
+        # 再重查返回抢先创建成功的库。
+        session = getattr(asset_library_repository, "session", None)
+        if session is not None:
+            try:
+                session.rollback()
+            except Exception:
+                logger.warning("IntegrityError 后回滚 session 失败（可能已关闭）", exc_info=True)
+        for lib in asset_library_repository.find_by_project(project.id):
+            kind = lib.kind.value if hasattr(lib.kind, "value") else lib.kind
+            if kind == AssetLibraryKind.VOICE.value:
+                return lib
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="配音素材库创建失败，请重试",
+        ) from None  # IntegrityError 已处理，不保留异常链
+
+
 @router.post(
     "/jobs/{job_id}/save-to-library",
     response_model=SaveToLibraryResponse,
@@ -294,13 +373,17 @@ def save_tts_job_to_library(
     request: SaveToLibraryRequest = SaveToLibraryRequest(),
     authenticated_user: AuthenticatedUser = Depends(get_current_user),
     tts_repository: SQLAlchemyTTSJobRepository = Depends(_get_repository),
-    voice_library_repository: SQLAlchemyVoiceLibraryRepository = Depends(get_voice_library_repository),
-    user_repository: UserRepository = Depends(get_user_repository),
+    asset_repository: AssetRepository = Depends(get_asset_repository),
+    asset_library_repository: AssetLibraryRepository = Depends(get_asset_library_repository),
+    project_repository: ProjectRepository = Depends(get_project_repository),
+    storage_service: SharedStorageService = Depends(get_storage_service),
     sign_url=Depends(get_audio_url_signer),
 ) -> SaveToLibraryResponse:
-    """将已完成的 TTS 合成结果保存到配音库。
+    """将已完成的 TTS 合成结果保存到配音素材库（assets 表新素材体系）。
 
-    自动携带音色名、时长、语速等元信息。
+    流程：把 TTS 输出音频转存到用户素材 OSS 路径 → 创建 file_type=audio、
+    status=ready 的 asset（挂用户 voice 素材库）→ 返回前端可用结构。
+    配额策略与素材上传一致（上传/ingest 链路无额外配额拦截）。
     """
     user_id = authenticated_user.user.id
 
@@ -318,64 +401,117 @@ def save_tts_job_to_library(
             detail="TTS job is not completed yet",
         )
 
-    # 构建配音素材名称
+    if not job.output_audio_url and not job.output_audio_key:
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail="TTS job 缺少输出音频，无法保存",
+        )
+
+    # 素材名称
     name = request.name or f"TTS-{job.id[:8]}"
 
-    # 构建元信息
-    metadata_ = {
+    # 找到（或自动创建）用户 voice 素材库
+    library = _find_or_create_voice_library(
+        user_id=user_id,
+        project_repository=project_repository,
+        asset_library_repository=asset_library_repository,
+    )
+
+    # 转存音频到素材 OSS 路径（tts-outputs/ 下的产物归 TTS 任务所有，
+    # 素材独立持有副本，删除 TTS 任务不影响配音库素材）
+    audio_format = (job.format or "mp3").strip() or "mp3"
+    content_type_map = {
+        "mp3": "audio/mpeg",
+        "wav": "audio/wav",
+        "pcm": "audio/pcm",
+        "opus": "audio/opus",
+    }
+    content_type = content_type_map.get(audio_format, "audio/mpeg")
+    storage_key = f"uploads/voice/tts/{job.id}.{audio_format}"
+
+    tmp_path: Path | None = None
+    try:
+        with tempfile.NamedTemporaryFile(suffix=f".{audio_format}", delete=False) as tmp:
+            tmp_path = Path(tmp.name)
+        # 优先用 OSS storage_key（走 oss2 SDK，私有 bucket 也可下载）；
+        # 兜底用 output_audio_url（旧任务可能没有 key）。
+        # download_asset 自动识别输入：http(s):// 开头走 HTTP 下载，否则按 OSS key 走 SDK。
+        download_source = job.output_audio_key or job.output_audio_url
+        downloaded = storage_service.download_asset(download_source, tmp_path)
+        if not downloaded or not tmp_path.exists() or tmp_path.stat().st_size == 0:
+            raise HTTPException(
+                status_code=status.HTTP_502_BAD_GATEWAY,
+                detail="TTS 音频下载失败，无法保存到配音库",
+            )
+        file_size = tmp_path.stat().st_size
+        storage_service.upload_file(tmp_path, storage_key, content_type=content_type)
+    except HTTPException:
+        raise
+    except Exception as e:
+        logger.error("TTS 音频转存素材失败: job_id=%s, error=%s", job.id, e, exc_info=True)
+        raise HTTPException(
+            status_code=status.HTTP_502_BAD_GATEWAY,
+            detail="TTS 音频转存失败，无法保存到配音库",
+        ) from e
+    finally:
+        if tmp_path and tmp_path.exists():
+            try:
+                tmp_path.unlink()
+            except OSError:
+                pass
+
+    # 构建素材元信息
+    metadata_: dict[str, object] = {
         "source": "tts_job",
         "tts_job_id": job.id,
         "format": job.format,
         "sample_rate": job.sample_rate,
+        "voice_id": job.voice_id,
+        "voice_name": job.voice_model or "",
     }
     if job.metadata:
-        # 保留原始 job 的有用元信息
         for key in ("speed", "language"):
             if key in job.metadata:
                 metadata_[key] = job.metadata[key]
 
-    # 获取用户套餐（用于配额检查）
-    user = user_repository.find_by_id(user_id)
-    plan_name = getattr(user, "subscription_plan", "free") if user else "free"
-
-    # 构建命令并执行
-    command = CreateVoiceLibraryCommand(
-        user_id=user_id,
+    asset = Asset.create(
+        project_id=library.project_id,
+        library_id=library.id,
         name=name,
-        text=job.input_text,
-        voice_provider="cosyvoice",
-        voice_id=job.voice_id,
-        voice_name=job.voice_model or "",
-        audio_url=job.output_audio_url,
-        duration=job.duration,
-        file_size=job.file_size,
-        status="completed",
-        project_id=job.project_id or "",
-        tags=[],
-        metadata_=metadata_,
+        storage_key=storage_key,
+        mime_type=content_type,
+        metadata=metadata_,
+        file_size=file_size,
+        duration=job.duration or None,
+        status=AssetStatus.READY,
+        classification_status=ClassificationStatus.PENDING,  # 音频不参与内容分类，保持 pending 与 ingest 链路一致
+        uploaded_by_user_id=user_id,
     )
-
-    use_case = CreateVoiceLibraryUseCase(voice_library_repository)
     try:
-        item = use_case.execute(command, plan_name=plan_name or "free")
-    except QuotaExceededError as exc:
+        asset = asset_repository.create(asset)
+    except Exception as e:
+        # DB 写入失败：清理已上传到 OSS 的素材文件，避免产生无法索引的孤儿文件
+        logger.error("素材记录创建失败，清理 OSS 文件: %s, error=%s", storage_key, e, exc_info=True)
+        try:
+            storage_service.delete_file(storage_key)
+        except Exception:
+            logger.warning("清理孤儿 OSS 文件失败: %s", storage_key, exc_info=True)
         raise HTTPException(
-            status_code=status.HTTP_429_TOO_MANY_REQUESTS,
-            detail=f"配音库配额已满（{exc.used}/{exc.limit}），请升级套餐",
-        ) from exc
+            status_code=status.HTTP_502_BAD_GATEWAY,
+            detail="素材保存失败，请重试",
+        ) from e
 
     return SaveToLibraryResponse(
-        id=item.id,
-        name=item.name,
-        audio_url=sign_url(item.audio_url) if item.audio_url else "",
-        duration=item.duration,
-        voice_id=item.voice_id,
-        voice_name=item.voice_name,
-        status=item.status,
+        id=asset.id,
+        name=asset.name,
+        audio_url=sign_url(storage_key),
+        duration=asset.duration or 0.0,
+        voice_id=job.voice_id,
+        voice_name=job.voice_model or "",
+        status="completed",
     )
 
 
-
 @router.post("/preview", response_model=TTSPreviewResponse)
 def preview_tts(
     request: TTSPreviewRequest,

apps/api/app/api/routes/voice_clones.py

@@ -7,7 +7,13 @@ from typing import Optional
 
 from app.auth import AuthenticatedUser, get_current_user
 from app.core.celery_app import celery_app
-from app.dependencies import get_cosyvoice_service, get_voice_clone_profile_repository
+from app.core.storage import get_storage_service
+from app.dependencies import (
+    get_asset_repository,
+    get_cosyvoice_service,
+    get_project_repository,
+    get_voice_clone_profile_repository,
+)
 from app.schemas.voice_clone import (
     CreateVoiceCloneRequest,
     ListVoiceCloneResponse,
@@ -32,6 +38,9 @@ from packages.application.voice_clone.use_cases import (
 from packages.application.voice_clone.workflow import (
     VoiceCloneWorkflowService,
 )
+from packages.ports.asset_repository import AssetRepository
+from packages.ports.project_repository import ProjectRepository
+from packages.shared.storage import SharedStorageService
 
 logger = logging.getLogger(__name__)
 
@@ -83,23 +92,68 @@ def create_voice_clone(
     request: CreateVoiceCloneRequest,
     authenticated_user: AuthenticatedUser = Depends(get_current_user),
     workflow: VoiceCloneWorkflowService = Depends(_get_workflow_service),
+    asset_repository: AssetRepository = Depends(get_asset_repository),
+    project_repository: ProjectRepository = Depends(get_project_repository),
+    storage_service: SharedStorageService = Depends(get_storage_service),
 ) -> VoiceCloneProfileResponse:
     """创建音色克隆任务。
 
     创建 VoiceCloneProfile → 提交 CosyVoice 克隆任务 → 触发 Celery 异步轮询。
-    如果有 source_audio_url，状态会变为 processing；否则保持 pending。
+    参考音频两种来源（二选一）：
+    - source_audio_url：前端直传后的音频 URL（兼容旧流程）
+    - asset_id：配音素材库中的音频素材，服务端用其 OSS storage_key 生成
+      预签名下载 URL（不依赖前端签名，避免签名过期导致克隆失败）
+    如果有参考音频，状态会变为 processing；否则保持 pending。
     """
     user_id = authenticated_user.user.id
+
+    source_audio_url = request.source_audio_url
+    clone_metadata = dict(request.metadata_ or {})
+
+    if request.asset_id:
+        if source_audio_url:
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail="asset_id 与 source_audio_url 只能传一个",
+            )
+        asset = asset_repository.find_by_id(request.asset_id)
+        if asset is None:
+            raise HTTPException(
+                status_code=status.HTTP_404_NOT_FOUND,
+                detail="素材不存在",
+            )
+        # 归属校验：素材挂在项目素材库下，用户必须能访问该项目
+        project = project_repository.find_by_id(asset.project_id)
+        if project is None or not project.can_access(user_id):
+            raise HTTPException(
+                status_code=status.HTTP_403_FORBIDDEN,
+                detail="无权使用该素材",
+            )
+        # 类型校验：仅支持音频素材
+        if asset.file_type != "audio":
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail="仅支持音频素材进行音色克隆",
+            )
+        if not asset.storage_key:
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail="该素材缺少音频文件，无法用于克隆",
+            )
+        # 用 OSS storage_key 生成服务端预签名 URL（7 天有效，覆盖克隆重试周期）
+        source_audio_url = storage_service.get_download_url(asset.storage_key, expires_seconds=7 * 24 * 3600)
+        clone_metadata["source_asset_id"] = asset.id
+
     profile = workflow.start_clone(
         user_id=user_id,
         name=request.name,
         description=request.description,
-        source_audio_url=request.source_audio_url,
+        source_audio_url=source_audio_url,
         voice_model=request.voice_model,
         language=request.language,
         gender=request.gender,
         max_retries=request.max_retries,
-        metadata=request.metadata_,
+        metadata=clone_metadata,
     )
 
     # 如果 profile 处于 processing 且有 task_id，触发 Celery 异步轮询

apps/api/app/schemas/asset.py

@@ -53,6 +53,14 @@ class AssetResponse(BaseModel):
     created_at: str
     uploaded_by_user_id: str
     tag_ids: list[str] = Field(default_factory=list)
+    # 片段级余量信息（仅视频素材返回，非视频/无时长记录为 None，前端按可用处理）
+    used_duration: float | None = Field(default=None, description="已使用片段时长（秒，历史区间合并去重后）")
+    available_duration: float | None = Field(default=None, description="剩余可用时长（秒）= 素材总时长 - 已用时长")
+    used_ratio: float | None = Field(default=None, description="已用时长占比（0~1）")
+    usable: bool = Field(
+        default=True,
+        description="是否仍可用于新片段：零重复可切区间耗尽且所有历史区间复用次数" "（use_count）均达上限时为 false",
+    )
 
 
 MAX_BATCH_SIZE = 200

apps/api/app/schemas/generation_task.py

@@ -45,11 +45,9 @@ class CreateGenerationTaskRequest(BaseModel):
     # ── 素材库自动匹配 ──
     asset_select_mode: str = Field(
         default="all",
-        description="素材选取模式：all=全部ready视频, random=随机选取, smart=智能匹配（按质量/时长评分）",
-    )
-    asset_select_count: int = Field(
-        default=0, ge=0, le=100, description="选取数量，0表示全部（仅 random/smart 模式有效）"
+        description="素材选取模式：all=全部ready视频, smart=智能匹配（按质量/时长评分）",
     )
+    asset_select_count: int = Field(default=0, ge=0, le=100, description="选取数量，0表示全部（仅 smart 模式有效）")
     # ── 自动重试 ──
     auto_retry_enabled: bool = Field(
         default=False,

apps/api/app/schemas/voice_clone.py

@@ -13,7 +13,8 @@ class CreateVoiceCloneRequest(BaseModel):
 
     name: str = Field(..., min_length=1, max_length=100, description="音色名称")
     description: str = Field("", description="音色描述")
-    source_audio_url: str = Field("", description="参考音频 URL")
+    source_audio_url: str = Field("", description="参考音频 URL（与 asset_id 二选一）")
+    asset_id: str = Field("", description="参考音频素材 ID（配音素材库中的音频 asset，与 source_audio_url 二选一）")
     voice_model: str = Field("", description="语音模型名称")
     language: str = Field("zh-CN", description="语言")
     gender: str = Field("unknown", description="性别")

apps/api/app/services/asset_segment_tracker.py

@@ -0,0 +1,440 @@
+"""素材片段级使用记录追踪与受控复用.
+
+在素材 metadata（assets.classification_result JSON）中持久化已使用的片段时间区间，
+供 from-assets 创建片段时避开历史区间，实现跨任务/跨调用的片段去重；
+素材可用区间耗尽后进入受控复用：允许有限次数（MAX_RANGE_USE_COUNT）复用最久未用
+的历史区间，配合调用方的成片复用占比控制（MAX_REUSE_RATIO = 15%），把任意两条
+成片的画面重复率控制在阈值内。
+
+metadata 中的记录字段 ``used_time_ranges``::
+
+    "used_time_ranges": [
+        {
+            "start": 12.5, "end": 20.3,
+            "plan_id": "plan-xxx",
+            "created_at": "2026-08-29T12:00:00+00:00",
+            "use_count": 1,              # 该区间累计被使用次数（复用一次 +1）
+            "last_used_at": "2026-08-29T12:00:00+00:00"  # 最近一次使用时间
+        },
+        ...
+    ]
+
+注意：本模块所有函数都不自行 commit，由调用方控制事务边界
+（from-assets 与 replace_all_clips_transactional 同事务；异步任务各自 commit）。
+历史记录永不自动清空（自动轮回重置已下线，reset_used_segments 仅保留给运维/测试）。
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+from datetime import datetime, timezone
+from typing import Callable
+
+from sqlalchemy.orm import Session
+
+from packages.adapters.sqlalchemy_impl.models import AssetModel
+
+logger = logging.getLogger(__name__)
+
+USED_RANGES_KEY = "used_time_ranges"
+
+# ── 受控复用配置常量 ─────────────────────────────────────────────────────────
+MAX_RANGE_USE_COUNT = 3
+"""单条历史区间最多被使用次数（含首次），达到后不再参与复用。"""
+
+REUSE_RATIO_LIMIT = 0.15
+"""单条成片中，单个素材的复用片段累计时长 / 该素材在成片中的总时长上限（15%）。
+超过则该素材不再分配新片段（调用方在轮询分配时跳过）。"""
+
+SEGMENT_EDGE_GAP = 0.3
+"""冲突判定边缘间隙（秒）：历史区间按 [start-gap, end+gap] 扩边后参与冲突检测，
+避免两条片段首尾紧贴导致画面观感重复；记录仍存实际值。"""
+
+# 判定"新片段与历史区间为同一次使用（复用）"的重叠率阈值：
+# 重叠时长 / 新区间时长超过该比例视为复用该历史区间（累加 use_count）而非新增记录。
+_REUSE_OVERLAP_RATIO = 0.6
+
+
+def _now_iso() -> str:
+    return datetime.now(timezone.utc).isoformat()
+
+
+def _read_meta(model) -> dict:
+    """读取素材 metadata dict。
+
+    兼容两种对象：
+    - ORM ``AssetModel``：metadata 以 JSON 字符串存在 ``classification_result`` 列；
+    - 领域实体 ``Asset``（路由层 repository 返回）：metadata 直接是 dict 属性
+      （repository 与 classification_result 互转，见 asset_repository.py）。
+    """
+    # 领域实体：metadata 已是 dict
+    meta = getattr(model, "metadata", None)
+    if isinstance(meta, dict):
+        return meta
+    raw = getattr(model, "classification_result", None)
+    if not raw:
+        return {}
+    try:
+        data = json.loads(raw) if isinstance(raw, str) else raw
+        return data if isinstance(data, dict) else {}
+    except Exception:
+        return {}
+
+
+def _get_model(db: Session, asset_id: str, for_update: bool = False) -> AssetModel | None:
+    query = db.query(AssetModel).filter(AssetModel.id == asset_id)
+    if for_update:
+        # 行级锁（PostgreSQL SELECT ... FOR UPDATE）：序列化同一素材的
+        # classification_result 读-改-写，避免并发事务丢失使用记录。
+        # SQLite 不支持时 SQLAlchemy 会忽略该子句（no-op）。
+        query = query.with_for_update()
+    return query.first()
+
+
+def get_used_segments(db: Session, asset_ids: list[str]) -> dict[str, list[tuple[float, float]]]:
+    """聚合多个素材的历史已用片段区间。
+
+    Returns:
+        ``{asset_id: [(start, end), ...]}`` 格式，与 ``_calc_random_start_time`` 的
+        ``used_segments`` 参数格式一致，可直接传入。
+    """
+    if not asset_ids:
+        return {}
+    result: dict[str, list[tuple[float, float]]] = {}
+    models = db.query(AssetModel).filter(AssetModel.id.in_(list(set(asset_ids)))).all()
+    for model in models:
+        meta = _read_meta(model)
+        ranges = meta.get(USED_RANGES_KEY) or []
+        segments: list[tuple[float, float]] = []
+        for r in ranges:
+            try:
+                segments.append((float(r["start"]), float(r["end"])))
+            except (KeyError, TypeError, ValueError):
+                continue
+        if segments:
+            result[model.id] = segments
+    return result
+
+
+def record_used_segments(
+    db: Session,
+    asset_id: str,
+    start: float,
+    end: float,
+    plan_id: str,
+) -> None:
+    """记录一次片段使用（不 commit）.
+
+    若新区间与某条历史区间高度重叠（复用场景，如受控复用回调返回的区间、
+    MediaKit 挪到历史区间），则累加该记录的 ``use_count`` 并刷新 ``last_used_at``，
+    不新增记录；否则追加一条新记录（use_count=1）。
+    """
+    # 行级锁读取：与并发生成任务互斥，保证区间记录读-改-写一致
+    model = _get_model(db, asset_id, for_update=True)
+    if model is None:
+        logger.warning("[片段追踪] 素材不存在，跳过记录: asset_id=%s", asset_id)
+        return
+    meta = _read_meta(model)
+    ranges = list(meta.get(USED_RANGES_KEY) or [])
+
+    new_start = round(float(start), 3)
+    new_end = round(float(end), 3)
+    new_dur = max(new_end - new_start, 1e-6)
+    now = _now_iso()
+
+    for r in ranges:
+        try:
+            rs, re_ = float(r["start"]), float(r["end"])
+        except (KeyError, TypeError, ValueError):
+            continue
+        overlap = max(0.0, min(new_end, re_) - max(new_start, rs))
+        if overlap / new_dur >= _REUSE_OVERLAP_RATIO:
+            # 复用同一条历史区间：累加次数、刷新时间
+            r["use_count"] = int(r.get("use_count", 1)) + 1
+            r["last_used_at"] = now
+            r["plan_id"] = plan_id
+            meta[USED_RANGES_KEY] = ranges
+            model.classification_result = json.dumps(meta, ensure_ascii=False)
+            model.updated_at = datetime.now(timezone.utc)
+            return
+
+    ranges.append(
+        {
+            "start": new_start,
+            "end": new_end,
+            "plan_id": plan_id,
+            "created_at": now,
+            "use_count": 1,
+            "last_used_at": now,
+        }
+    )
+    meta[USED_RANGES_KEY] = ranges
+    model.classification_result = json.dumps(meta, ensure_ascii=False)
+    model.updated_at = datetime.now(timezone.utc)
+
+
+def remove_used_segment(
+    db: Session,
+    asset_id: str,
+    start: float,
+    end: float,
+    plan_id: str | None = None,
+    tolerance: float = 0.5,
+) -> bool:
+    """删除素材 metadata 中匹配的一条使用记录（不 commit）.
+
+    匹配规则：start/end 与记录值相差不超过 tolerance 秒；plan_id 非空时，
+    记录有 plan_id 则需相等，记录缺 plan_id（本功能上线前的旧数据）时按时间匹配。
+    Returns:
+        是否找到并删除了记录。
+    """
+    model = _get_model(db, asset_id)
+    if model is None:
+        return False
+    meta = _read_meta(model)
+    ranges = list(meta.get(USED_RANGES_KEY) or [])
+    remaining: list[dict] = []
+    removed = False
+    for r in ranges:
+        try:
+            match = (
+                abs(float(r["start"]) - float(start)) <= tolerance and abs(float(r["end"]) - float(end)) <= tolerance
+            )
+        except (KeyError, TypeError, ValueError):
+            remaining.append(r)
+            continue
+        # plan_id 校验：传入 plan_id 时，记录有 plan_id 则必须相等；
+        # 记录本身缺 plan_id（旧数据）时退化为按时间匹配，避免旧区间永远删不掉
+        if plan_id is not None and r.get("plan_id") is not None and r.get("plan_id") != plan_id:
+            match = False
+        if match and not removed:
+            removed = True
+            continue
+        remaining.append(r)
+    if removed:
+        meta[USED_RANGES_KEY] = remaining
+        model.classification_result = json.dumps(meta, ensure_ascii=False)
+        model.updated_at = datetime.now(timezone.utc)
+    return removed
+
+
+def reset_used_segments(db: Session, asset_id: str) -> None:
+    """清空单个素材的历史片段使用记录（不 commit）.
+
+    仅供运维/测试使用；正常生成流程中历史记录永不自动清空（受控复用取代自动轮回）。
+    """
+    model = _get_model(db, asset_id)
+    if model is None:
+        return
+    meta = _read_meta(model)
+    if meta.get(USED_RANGES_KEY):
+        meta[USED_RANGES_KEY] = []
+        model.classification_result = json.dumps(meta, ensure_ascii=False)
+        model.updated_at = datetime.now(timezone.utc)
+        logger.info("[片段追踪] 素材区间记录手动清空: asset_id=%s", asset_id)
+
+
+# ── 素材余量/可用性计算（Task H：素材库角标 + smart-match 过滤）──────────────
+
+# 判定「是否还有空闲可切区间」时使用的最小片段时长（秒）：空闲段长于此值才视为可切
+_MIN_FREE_CLIP_DURATION = 3.0
+
+
+def _merge_intervals(intervals: list[tuple[float, float]]) -> list[tuple[float, float]]:
+    """合并重叠/相接的时间区间，返回升序不重叠区间列表。"""
+    if not intervals:
+        return []
+    ordered = sorted((float(a), float(b)) for a, b in intervals if b > a)
+    merged: list[tuple[float, float]] = [ordered[0]]
+    for start, end in ordered[1:]:
+        last_start, last_end = merged[-1]
+        if start <= last_end:
+            merged[-1] = (last_start, max(last_end, end))
+        else:
+            merged.append((start, end))
+    return merged
+
+
+def _has_free_gap(used: list[tuple[float, float]], total: float, min_free: float = _MIN_FREE_CLIP_DURATION) -> bool:
+    """素材 [0, total] 中是否存在长度 ≥ min_free 的空闲段（考虑边缘间隙）。"""
+    if total <= 0:
+        return False
+    # 历史区间按边缘间隙扩边后判定空闲（与选片冲突检测同一口径）
+    expanded = [(max(0.0, s - SEGMENT_EDGE_GAP), min(total, e + SEGMENT_EDGE_GAP)) for s, e in used]
+    merged = _merge_intervals(expanded)
+    cursor = 0.0
+    for start, end in merged:
+        if start - cursor >= min_free:
+            return True
+        cursor = max(cursor, end)
+    return total - cursor >= min_free
+
+
+def compute_asset_availability(
+    model: "AssetModel | None",
+    min_free_clip_duration: float = _MIN_FREE_CLIP_DURATION,
+) -> dict | None:
+    """计算单个素材的余量与可用性（纯函数，不读写 DB）。
+
+    Returns:
+        视频素材返回 ``{"used_duration", "available_duration", "used_ratio", "usable"}``；
+        非视频 / 无 model / 无时长信息返回 None（调用方按可用处理，零影响）。
+
+    usable=False 条件（与受控复用机制一致）：
+        零重复可切区间已耗尽（不存在 ≥ min_free 的空闲段）且
+        所有历史区间 use_count 均达 MAX_RANGE_USE_COUNT 上限（无区间可复用）。
+    """
+    if model is None:
+        return None
+    file_type = getattr(model, "file_type", None) or getattr(model, "mime_type", "") or ""
+    if file_type != "video" and not str(file_type).startswith("video/"):
+        return None
+    total = float(getattr(model, "duration", 0.0) or 0.0)
+    if total <= 0:
+        return None
+
+    meta = _read_meta(model)
+    raw_ranges = meta.get(USED_RANGES_KEY) or []
+
+    intervals: list[tuple[float, float]] = []
+    use_counts: list[int] = []
+    for r in raw_ranges:
+        try:
+            start = float(r["start"])
+            end = float(r["end"])
+        except (KeyError, TypeError, ValueError):
+            continue
+        if end <= start:
+            continue
+        intervals.append((start, end))
+        try:
+            use_counts.append(int(r.get("use_count", 1)))
+        except (TypeError, ValueError):
+            use_counts.append(1)
+
+    merged = _merge_intervals(intervals)
+    used_duration = round(sum(e - s for s, e in merged), 3)
+    used_duration = min(used_duration, total)
+    available_duration = round(max(total - used_duration, 0.0), 3)
+    used_ratio = round(min(used_duration / total, 1.0), 4)
+
+    has_free = _has_free_gap(intervals, total, min_free_clip_duration)
+    if has_free:
+        usable = True
+    else:
+        # 空闲段耗尽：仅当存在历史区间且全部达复用上限时才判定不可用；
+        # 无历史区间（理论上不会走到，因为 has_free=True）按可用处理
+        if not use_counts:
+            usable = True
+        else:
+            usable = any(uc < MAX_RANGE_USE_COUNT for uc in use_counts)
+
+    return {
+        "used_duration": used_duration,
+        "available_duration": available_duration,
+        "used_ratio": used_ratio,
+        "usable": usable,
+    }
+
+
+def find_reusable_range(
+    db: Session,
+    asset_id: str,
+    clip_duration: float,
+    asset_total: float,
+    *,
+    max_use_count: int = MAX_RANGE_USE_COUNT,
+) -> tuple[float, float] | None:
+    """受控复用：在素材历史区间中选一条可复用区间返回 (start, end)。
+
+    选择规则：
+    1. 仅选 ``use_count < max_use_count`` 的历史区间；
+    2. 优先返回能完整容纳当前 clip_duration（起点后不越素材边界）的最久未用区间；
+    3. 没有能容纳的，则返回 last_used_at 最老（或缺失 last_used_at 的旧数据优先）
+       且 use_count 最低的区间起点（可能与其他历史区间重叠，属降级复用）；
+    4. 无任何可复用区间（记录为空或全部达上限）返回 None。
+
+    本函数只读不写；复用次数的累加由后续 record_used_segments 完成。
+    """
+    model = _get_model(db, asset_id)
+    if model is None:
+        return None
+    meta = _read_meta(model)
+    ranges = [r for r in (meta.get(USED_RANGES_KEY) or []) if int(r.get("use_count", 1)) < max_use_count]
+    if not ranges:
+        return None
+
+    def _last_used(r: dict) -> str:
+        return str(r.get("last_used_at") or r.get("created_at") or "")
+
+    max_start = max(0.0, asset_total - clip_duration)
+    # 2. 能完整容纳当前片段的候选：按 last_used_at 升序（最久未用优先）
+    fit = sorted(
+        [r for r in ranges if float(r["start"]) <= max_start + 1e-6],
+        key=_last_used,
+    )
+    if fit:
+        start = min(float(fit[0]["start"]), max_start)
+        return (start, start + clip_duration)
+
+    # 3. 降级：最久未用 + use_count 最低的区间起点
+    fallback = sorted(ranges, key=lambda r: (_last_used(r), int(r.get("use_count", 1))))[0]
+    start = min(float(fallback["start"]), max_start)
+    return (start, start + clip_duration)
+
+
+def make_reuse_callback(
+    db: Session,
+    asset_durations: dict[str, float],
+    reused_tracker: dict[str, float] | None = None,
+    assigned_tracker: dict[str, float] | None = None,
+    ratio_limit: float = REUSE_RATIO_LIMIT,
+) -> Callable[[str, float], tuple[float, float] | None]:
+    """构造给 ``_calc_random_start_time`` 用的受控复用回调.
+
+    Args:
+        db: SQLAlchemy session
+        asset_durations: 素材 ID -> 总时长（回调需要素材总时长做边界约束）
+        reused_tracker: 可选的 ``{asset_id: 累计复用时长}``，回调成功返回复用区间时
+            会把本次片段时长累加进去，供调用方统计成片复用占比（15% 阈值）。
+        assigned_tracker: 可选的 ``{asset_id: 已分配片段总时长}``，配合 ratio_limit
+            在复用前预判：若复用本片段后占比 (reused + clip_duration) /
+            (assigned + clip_duration) 超过 ratio_limit，则拒绝复用、返回 None
+            （保证成片复用占比不超阈值）。
+        ratio_limit: 单条成片复用时长占比上限，默认 15%。
+
+    Returns:
+        回调函数 ``(asset_id, clip_duration) -> (start, end) | None``。
+        回调内吞掉 DB 异常返回 None，不影响主生成流程。
+    """
+
+    def _reuse(asset_id: str, clip_duration: float) -> tuple[float, float] | None:
+        try:
+            total = float(asset_durations.get(asset_id, 0.0) or 0.0)
+            if total <= 0:
+                return None
+            # 占比闸门：预判复用本片段后是否超限（仅当调用方提供了 assigned tracker）
+            if assigned_tracker is not None:
+                assigned = float(assigned_tracker.get(asset_id, 0.0) or 0.0)
+                reused_amt = float((reused_tracker or {}).get(asset_id, 0.0) or 0.0)
+                if assigned > 0 and (reused_amt + clip_duration) / (assigned + clip_duration) > ratio_limit:
+                    logger.info(
+                        "[片段追踪] 复用占比预判超 %.0f%% 阈值，拒绝复用: asset_id=%s "
+                        "reused=%.1f assigned=%.1f clip=%.1f",
+                        ratio_limit * 100,
+                        asset_id,
+                        reused_amt,
+                        assigned,
+                        clip_duration,
+                    )
+                    return None
+            result = find_reusable_range(db, asset_id, clip_duration, total)
+        except Exception:
+            logger.warning("[片段追踪] 受控复用查询异常: asset_id=%s", asset_id, exc_info=True)
+            return None
+        if result is not None and reused_tracker is not None:
+            reused_tracker[asset_id] = reused_tracker.get(asset_id, 0.0) + clip_duration
+        return result
+
+    return _reuse

apps/api/app/services/edit_plan_service.py

@@ -9,6 +9,12 @@ from __future__ import annotations
 import logging
 from typing import Any, Dict, List, Optional
 
+from app.services.asset_segment_tracker import (
+    REUSE_RATIO_LIMIT,
+    get_used_segments,
+    make_reuse_callback,
+    record_used_segments,
+)
 from sqlalchemy.orm import Session
 
 from packages.adapters.sqlalchemy_impl import (
@@ -453,6 +459,116 @@ class EditPlanService:
             logger.exception("事务性替换片段失败: plan_id=%s", plan_id)
             raise
 
+    def clone_plan_for_variant(
+        self,
+        source_plan_id: str,
+        *,
+        created_by_user_id: str = "",
+        name_suffix: str = "变体",
+        reuse_tracker: Optional[dict] = None,
+    ) -> EditPlan:
+        """为批量/多预览场景克隆一份独立 plan，片段起点全部重算（受控随机/复用）。
+
+        复制源 plan 的模板归属、config 与片段结构（asset_id / duration / clip_type /
+        order 不变），每个片段重新调用 ``_calc_random_start_time``：读取素材持久化的
+        历史已用区间避让，耗尽时受控复用（use_count<3、最久未用），从而保证 N 条
+        成片片段区间互不相同，且复用占比受控。
+
+        - 不替换/不修改源 plan，源 plan 保留用户手动编辑结果。
+        - 片段区间记录（record_used_segments）随新片段写入素材 metadata，与新 plan
+          同事务；复用历史区间时由 record 自动累加 use_count。
+        - 克隆的 clips 复用区间累计时长写入 reuse_tracker（可选），供调用方统计占比。
+
+        Raises:
+            ValueError: 源 plan 不存在或无可用片段。
+        """
+        from packages.adapters.sqlalchemy_impl.models import AssetModel
+        from packages.domain.plan_generator_utils import _calc_random_start_time
+
+        source = self.get_plan_or_raise(source_plan_id)
+
+        # 分页读取源 plan 全部片段
+        clips: List[EditPlanClip] = []
+        skip, page = 0, 500
+        while True:
+            batch = self._clip_repo.list_by_plan(source_plan_id, skip=skip, limit=page)
+            if not batch:
+                break
+            clips.extend(batch)
+            if len(batch) < page:
+                break
+            skip += page
+        if not clips:
+            raise ValueError(f"源 plan 无片段，无法克隆变体: {source_plan_id}")
+
+        # 创建新 plan（复制模板归属与 config）
+        new_plan = self.create_plan(
+            template_id=source.template_id,
+            name=f"{source.name or '剪辑计划'} · {name_suffix}",
+            config=dict(source.config or {}),
+            total_duration=source.total_duration,
+            project_id=source.project_id or "",
+            created_by_user_id=created_by_user_id or (source.created_by_user_id or ""),
+        )
+
+        # 素材时长映射（O(N) 单查）
+        asset_ids = list({c.asset_id for c in clips if c.asset_id})
+        db = self._clip_repo.session
+        durations: dict[str, float] = {}
+        if asset_ids:
+            for m in db.query(AssetModel).filter(AssetModel.id.in_(asset_ids)).all():
+                durations[m.id] = float(getattr(m, "duration", 0.0) or 0.0)
+
+        used_segments = get_used_segments(db, asset_ids)
+        reused: dict[str, float] = reuse_tracker if reuse_tracker is not None else {}
+        asset_assigned: dict[str, float] = {}
+        # 回调内部预判复用后占比超 15% 则拒绝复用（calc 返回 None → 保留原起点）
+        reuse_cb = make_reuse_callback(db, durations, reused, assigned_tracker=asset_assigned)
+
+        clips_data: list[dict] = []
+        for i, c in enumerate(clips):
+            aid = c.asset_id
+            dur = float(c.duration or 0.0)
+            total = durations.get(aid, 0.0)
+            if aid and total > 0 and dur > 0:
+                # 复用占比闸门：本片段尚未分配，检查当前已复用占比
+                # reused / assigned 是否超 15%，超则该素材不再分配（保留原起点）；
+                # assigned=0（首个片段）放行，reused=0 时不误拦正常分配
+                assigned = asset_assigned.get(aid, 0.0)
+                eff_dur = min(dur, total)
+                reused_amt = reused.get(aid, 0.0)
+                ratio_blocked = assigned > 0 and reused_amt / assigned > REUSE_RATIO_LIMIT
+                start = None
+                if not ratio_blocked:
+                    start = _calc_random_start_time(aid, eff_dur, durations, used_segments, on_exhausted=reuse_cb)
+                if start is None:
+                    start = float(c.start_time or 0.0)
+                asset_assigned[aid] = assigned + eff_dur
+                used_segments.setdefault(aid, []).append((start, start + eff_dur))
+                record_used_segments(db, aid, start, start + eff_dur, new_plan.id)
+            else:
+                start = float(c.start_time or 0.0)
+
+            clips_data.append(
+                {
+                    "order": c.order if c.order is not None else i,
+                    "asset_id": aid,
+                    "start_time": start,
+                    "duration": dur,
+                    "clip_type": c.clip_type,
+                }
+            )
+
+        # 事务性写入新 plan 的片段（内部统一 commit/rollback）
+        self.replace_all_clips_transactional(new_plan.id, clips_data)
+        logger.info(
+            "克隆变体 plan: source=%s new=%s clips=%d",
+            source_plan_id,
+            new_plan.id,
+            len(clips_data),
+        )
+        return new_plan
+
     # ── 片段分割与合并 ──────────────────────────────────────────────────────
 
     def split_clip(self, clip_id: str, split_time: float) -> Dict[str, Any]:

apps/web/e2e/core-generation.spec.ts

@@ -52,7 +52,7 @@ type AssetListResponse = {
 test.describe("Core generation flow", () => {
   test.describe.configure({ timeout: 360_000 })
 
-  test("walks through 7-step wizard and starts generation", async ({ page, request }) => {
+  test("walks through 6-step wizard and starts generation", async ({ page, request }) => {
     test.setTimeout(360_000)
 
     await routeBrowserApiToTestApi(page)
@@ -214,18 +214,14 @@ test.describe("Core generation flow", () => {
 
     const titleText = `E2E Test ${suffix}`
     await titleInput.fill(titleText)
-    await page.getByRole("button", { name: "下一步" }).click()
 
-    // Step 5: preview — 前端实时预览架构改造，无需后端生成预览
-    await expect(page.getByRole("heading", { name: /预览设置/ })).toBeVisible({ timeout: 15000 })
-    await page.getByRole("button", { name: "下一步" }).click()
-
-    // Step 6: cover (默认 AI 智能选帧模式，直接下一步)
-    await expect(page.getByRole("heading", { name: /选择封面/ })).toBeVisible({ timeout: 15000 })
-    await page.getByRole("button", { name: "下一步" }).click()
-
-    // Step 7: confirm and generate
-    await expect(page.getByRole("heading", { name: /确认生成/ })).toBeVisible()
+    // Step 4（标题+实时预览）：确认生成按钮已移到标题页，点击直接创建最终渲染任务
+    // 等待前端实时预览就绪：未就绪时右侧 FrontendPreviewPlayer 显示「准备预览素材...」占位，
+    // 就绪（previewReady：素材已解析 + 模板已选中）后占位消失；否则按钮会被校验拦截弹 warning
+    await page
+      .getByText("准备预览素材")
+      .waitFor({ state: "detached", timeout: 30_000 })
+      .catch(() => {})
 
     // Wait for generation API to be called
     // 前端直接创建生成任务：POST /generation/tasks
@@ -238,10 +234,10 @@ test.describe("Core generation flow", () => {
       { timeout: 30_000 },
     )
 
-    // Click generate button
-    await page.locator(".xx-btn-primary").filter({ hasText: "确认生成" }).first().click()
+    // 点击「确认生成视频」
+    await page.locator(".xx-btn-primary").filter({ hasText: "确认生成视频" }).first().click()
 
-    // Verify generation was triggered successfully
+    // Verify generation was triggered
     const genResp = await generatePromise
     if (!genResp.ok()) {
       const body = await genResp.text()
@@ -258,19 +254,42 @@ test.describe("Core generation flow", () => {
       }
       expect(genData.items.length).toBeGreaterThan(0)
       expect(genData.items[0].id).toBeTruthy()
+
+      // Step 5: 确认生成页 — 任务创建成功后自动跳转，展示渲染进度
+      await expect(page.getByRole("heading", { name: /确认生成/ })).toBeVisible({
+        timeout: 15_000,
+      })
+
+      // Step 5 → Step 6：等待渲染终态
+      // - 完成：页面出现「视频生成完成」，步骤5「下一步」按钮解锁，点击进入封面
+      // - 失败：出现「生成失败」，停在确认生成页也算向导流程走通
+      // - 超时未终态（测试环境 worker 可能不处理任务）：进度仍在轮询，同样算走通
+      const renderSucceeded = await page
+        .getByText("视频生成完成", { exact: false })
+        .waitFor({ timeout: 180_000 })
+        .then(() => true)
+        .catch(() => false)
+      if (renderSucceeded) {
+        // 渲染完成：手动点「下一步」进入封面步骤（渲染完不自动跳转）
+        await page.getByRole("button", { name: "下一步" }).click()
+        // Step 6: 封面（最后一步，无主按钮），仅验证页面渲染
+        await expect(page.getByRole("heading", { name: /选择封面/ })).toBeVisible({
+          timeout: 15_000,
+        })
+      } else {
+        // 失败或超时：仍在确认生成页（进度展示或失败提示），向导流程已完整走通
+        await expect(page.getByRole("heading", { name: /确认生成/ })).toBeVisible()
+        console.log("[E2E] 渲染任务失败或未在 180s 内完成，冒烟测试仍通过（已达确认生成页）")
+      }
     } else {
       console.log(`[E2E] Generate API returned ${genResp.status()}, wizard flow test still passes`)
+      // 创建失败时停留在标题页并展示错误提示
+      await page
+        .getByText(/生成失败|重新生成/)
+        .isVisible({ timeout: 15_000 })
+        .catch(() => false)
     }
 
-    // Generation may fail in test env (no worker), that's OK
-    // Just verify the flow started - check page shows generation-related UI
-    await page
-      .getByText(/生成中|生成完成|生成失败/)
-      .isVisible({ timeout: 15_000 })
-      .catch(() => false)
-    // If we see progress or result, great; if not, flow still reached the end
-    // which is sufficient for an E2E smoke test
-
     // Verify product library page loads (smoke: just verify page renders)
     await page.goto("/app/products")
     await expect(page).toHaveURL(/\/app\/products/)

apps/web/src/api/assets/assets.ts

@@ -52,12 +52,37 @@ export const getAssetsByKind = async (
 /**
  * 智能匹配素材（后端 AI 选素材）
  * 调用后端 smart-match 端点，由后端根据素材库内容智能选择素材
+ *
+ * 后端返回 items 元素兼容两种结构（过渡期）：
+ * - 扁平结构：AssetItem 本身（id 在顶层）
+ * - 包装结构：{ asset: AssetItem, score, breakdown }（id 需从 .asset 取）
+ * 这里统一归一化为 AssetItem[]，调用方无需关心包装层。
  */
-export const smartMatchAssets = async (libraryId: string): Promise<{ items: AssetItem[] }> => {
+export interface SmartMatchResult {
+  items: AssetItem[]
+}
+
+interface SmartMatchWrappedItem {
+  asset?: AssetItem
+  id?: string
+  score?: number
+  breakdown?: unknown
+}
+
+export const smartMatchAssets = async (libraryId: string): Promise<SmartMatchResult> => {
   const response = await apiClient.post("/assets/smart-match", {
     library_id: libraryId,
   })
-  return response.data
+  const rawItems: SmartMatchWrappedItem[] = response.data?.items ?? []
+  const items = rawItems
+    .map((it) =>
+      // 包装结构 { asset: {...} } 优先解包；否则视其本身为扁平 AssetItem
+      it?.asset && typeof it.asset === "object" && "id" in it.asset
+        ? it.asset
+        : (it as unknown as AssetItem),
+    )
+    .filter((it): it is AssetItem => !!it && typeof it.id === "string" && it.id.length > 0)
+  return { items }
 }
 
 /** 更新素材（名称、metadata 等） */

apps/web/src/api/assets/index.ts

@@ -20,6 +20,10 @@ export type {
 // 素材诊断
 export { getAssetDiagnosis } from "./diagnosis"
 
+// 素材余量/可用性判断
+export { isAssetUsable } from "./usage"
+export type { AssetUsageLike } from "./usage"
+
 // 素材库
 export {
   getAssetLibraries,
@@ -39,7 +43,13 @@ export {
 } from "./assets"
 
 // 上传
-export { prepareDirectUpload, completeDirectUpload, uploadAssetDirect } from "./upload"
+export {
+  prepareDirectUpload,
+  completeDirectUpload,
+  uploadAssetDirect,
+  prepareDirectUploadHandle,
+  type DirectUploadHandle,
+} from "./upload"
 
 // 任务
 export { getIngestJob, submitClassificationJob, getClassificationJob } from "./jobs"

apps/web/src/api/assets/types.ts

@@ -40,6 +40,10 @@ export interface AssetItem {
   thumbnail_url?: string
   /** 时长（秒），视频/音频素材由后端从 metadata 提取到顶层 */
   duration?: number
+  /** 已切片段占用时长占比（0~1，后端片段重复率控制机制返回；字段缺失视为未统计） */
+  used_ratio?: number | null
+  /** 是否已彻底用尽（无新区间且历史区间复用次数均达上限）；false 的素材不参与生成选片 */
+  usable?: boolean | null
   status?: string
   classification_status?: AssetClassificationStatus | null
   quality_score?: number | null
@@ -129,6 +133,12 @@ export interface DirectUploadPrepareResult {
   expires_at: string
   fields: Record<string, string>
   max_size_bytes: number
+  /**
+   * prepare 阶段预创建的素材记录 id（后端改造后返回：status=uploading）。
+   * 前端拿到后立即刷新列表，卡片以「上传中」态出现在素材网格中。
+   * 旧后端不返回该字段，前端降级为无预建卡片的原有行为。
+   */
+  asset_id?: string
 }
 
 /** 直传完成确认返回 */
@@ -136,4 +146,8 @@ export interface DirectUploadCompleteResult {
   storage_key: string
   ingest_job_id: string
   url: string
+  /** 同库已存在相同 file_hash 的素材时为 true，ingest_job_id 为空 */
+  duplicated?: boolean
+  /** duplicated 为 true 时返回已存在素材的 id */
+  asset_id?: string
 }

apps/web/src/api/assets/upload.ts

@@ -27,28 +27,18 @@ export const completeDirectUpload = async (data: {
   return response.data
 }
 
-/** 直传上传（大文件推荐），支持可选进度回调 */
-export const uploadAssetDirect = async (data: {
-  file: File
-  library_id: string
-  onProgress?: (percent: number) => void
-}): Promise<DirectUploadCompleteResult> => {
-  const project = await getOrCreateDefaultProject()
+/** 直传 OSS 的底层传输（POST 表单到 OSS），带进度回调 */
+const putToOSS = (
+  prepared: DirectUploadPrepareResult,
+  file: File,
+  onProgress?: (percent: number) => void,
+): Promise<void> =>
+  new Promise<void>((resolve, reject) => {
+    const directForm = new FormData()
+    Object.entries(prepared.fields).forEach(([key, value]) => directForm.append(key, value))
+    directForm.append("file", file)
 
-  const prepared = await prepareDirectUpload({
-    project_id: project.id,
-    library_id: data.library_id,
-    filename: data.file.name,
-    content_type: data.file.type || "application/octet-stream",
-    file_size: data.file.size,
-  })
-
-  const directForm = new FormData()
-  Object.entries(prepared.fields).forEach(([key, value]) => directForm.append(key, value))
-  directForm.append("file", data.file)
-
-  // 使用 XMLHttpRequest 以获取上传进度 + 超时控制 + 详细错误诊断
-  await new Promise<void>((resolve, reject) => {
+    // 使用 XMLHttpRequest 以获取上传进度 + 超时控制 + 详细错误诊断
     const xhr = new XMLHttpRequest()
     xhr.open(prepared.method, prepared.upload_url)
 
@@ -56,8 +46,8 @@ export const uploadAssetDirect = async (data: {
     xhr.timeout = 10 * 60 * 1000
 
     xhr.upload.onprogress = (e) => {
-      if (e.lengthComputable && data.onProgress) {
-        data.onProgress(Math.round((e.loaded / e.total) * 100))
+      if (e.lengthComputable && onProgress) {
+        onProgress(Math.round((e.loaded / e.total) * 100))
       }
     }
     xhr.onload = () => {
@@ -102,9 +92,53 @@ export const uploadAssetDirect = async (data: {
     xhr.send(directForm)
   })
 
-  return completeDirectUpload({
+/** 单个文件的上传阶段信息（供批量上传队列做状态绑定） */
+export interface DirectUploadHandle {
+  /** prepare 返回（含可能的预建 asset_id） */
+  prepared: DirectUploadPrepareResult
+  /** 直传 OSS（可重复调用用于重试） */
+  transfer: (onProgress?: (percent: number) => void) => Promise<void>
+  /** 直传完成后调用 complete 确认入库 */
+  complete: () => Promise<DirectUploadCompleteResult>
+}
+
+/**
+ * 准备一次直传：调 prepare 拿到签名表单（后端可能同时预建 uploading 态 asset），
+ * 返回分段执行的 handle，调用方自行控制 transfer/complete 时机（便于队列并发与重试）。
+ */
+export const prepareDirectUploadHandle = async (data: {
+  file: File
+  library_id: string
+}): Promise<DirectUploadHandle> => {
+  const project = await getOrCreateDefaultProject()
+
+  const prepared = await prepareDirectUpload({
     project_id: project.id,
     library_id: data.library_id,
-    storage_key: prepared.storage_key,
+    filename: data.file.name,
+    content_type: data.file.type || "application/octet-stream",
+    file_size: data.file.size,
   })
+
+  return {
+    prepared,
+    transfer: (onProgress) => putToOSS(prepared, data.file, onProgress),
+    complete: () =>
+      completeDirectUpload({
+        project_id: project.id,
+        library_id: data.library_id,
+        storage_key: prepared.storage_key,
+      }),
+  }
+}
+
+/** 直传上传（大文件推荐），支持可选进度回调；一次性完成 prepare→transfer→complete */
+export const uploadAssetDirect = async (data: {
+  file: File
+  library_id: string
+  onProgress?: (percent: number) => void
+}): Promise<DirectUploadCompleteResult> => {
+  const handle = await prepareDirectUploadHandle({ file: data.file, library_id: data.library_id })
+  await handle.transfer(data.onProgress)
+  return handle.complete()
 }

apps/web/src/api/assets/usage.ts

@@ -0,0 +1,18 @@
+/**
+ * 素材余量/可用性判断
+ * 后端片段重复率控制机制（任意两条成片画面重复率 ≤15%）上线后，
+ * 素材列表会附加 usable / used_ratio 字段。字段未上线前一律按可用处理。
+ */
+
+/** 仅依赖素材余量相关字段的最小结构，api 层与 pages 层 AssetItem 均可传入 */
+export interface AssetUsageLike {
+  usable?: boolean | null
+  used_ratio?: number | null
+}
+
+/**
+ * 素材是否仍可参与生成选片。
+ * usable === false 表示已彻底用尽（无新区间且复用次数全部达上限）；
+ * 字段缺失（undefined/null）时降级为可用，保证后端字段上线前零影响。
+ */
+export const isAssetUsable = (asset: AssetUsageLike): boolean => asset.usable !== false

apps/web/src/api/voice-clone/clones.ts

@@ -44,14 +44,19 @@ export const getVoiceCloneDetail = async (id: string): Promise<VoiceCloneProfile
   return response.data
 }
 
-/** 创建克隆音色 */
+/** 创建克隆音色（audio_url 与 asset_id 二选一） */
 export const createVoiceClone = async (
   data: CreateVoiceCloneRequest,
 ): Promise<VoiceCloneProfile> => {
   const payload: CreateVoiceCloneRequestFull = {
     name: data.name,
     description: data.description,
-    source_audio_url: data.audio_url,
+  }
+  // 从配音素材选择克隆：直接传 asset_id，后端用素材 OSS 路径克隆
+  if (data.asset_id) {
+    payload.asset_id = data.asset_id
+  } else {
+    payload.source_audio_url = data.audio_url
   }
   const response = await apiClient.post<VoiceCloneProfile>("/voice-clones", payload)
   return response.data

apps/web/src/api/voice-clone/types.ts

@@ -22,10 +22,13 @@ export interface VoiceClone {
   updated_at: string
 }
 
-/** 创建克隆请求（前端简化版） */
+/** 创建克隆请求（前端简化版：audio_url 与 asset_id 二选一） */
 export interface CreateVoiceCloneRequest {
   name: string
-  audio_url: string
+  /** 录音/文件上传后的音频 URL（与 asset_id 二选一） */
+  audio_url?: string
+  /** 从配音素材选择时直接传素材 ID，后端用素材 OSS 路径克隆（与 audio_url 二选一） */
+  asset_id?: string
   description?: string
 }
 
@@ -72,11 +75,13 @@ export interface VoiceCloneStatusResponse {
   retry_count: number
 }
 
-/** 后端创建克隆请求（完整版） */
+/** 后端创建克隆请求（完整版：source_audio_url 与 asset_id 二选一） */
 export interface CreateVoiceCloneRequestFull {
   name: string
   description?: string
-  source_audio_url: string
+  source_audio_url?: string
+  /** 从配音素材选择克隆时传素材 ID */
+  asset_id?: string
   voice_model?: string
   language?: string
   gender?: string

apps/web/src/components/voice/CloneModal/clone-modal.css

@@ -149,53 +149,29 @@
 
 /* ── 上传区域 ───────────────────────────────────────────── */
 
-.xx-clonemodal-upload-zone {
-  border: 2px dashed var(--xx-color-border, #e5e7eb);
+/* ── 素材选择空态 ─────────────────────────────────────────── */
+
+.xx-clonemodal-asset-empty {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 12px;
+  padding: 14px 16px;
+  border: 1px dashed var(--xx-color-border, #e5e7eb);
   border-radius: var(--radius-sm);
-  padding: 28px 20px;
-  text-align: center;
-  cursor: pointer;
-  transition: all 0.2s ease;
   background: var(--xx-color-bg-secondary, #f9fafb);
 }
 
-.xx-clonemodal-upload-zone:hover {
-  border-color: var(--xx-color-primary, #6366f1);
-  background: rgba(99, 102, 241, 0.03);
-}
-
-.xx-clonemodal-upload-zone--active {
-  border-color: var(--xx-color-primary, #6366f1);
-  background: rgba(99, 102, 241, 0.06);
-}
-
-.xx-clonemodal-upload-zone--has-file {
-  border-style: solid;
-  border-color: var(--xx-color-primary, #6366f1);
-  background: rgba(99, 102, 241, 0.04);
-}
-
-.xx-clonemodal-upload-icon {
-  font-size: 32px;
-  margin-bottom: 8px;
-}
-
-.xx-clonemodal-upload-title {
-  font-size: 14px;
-  font-weight: 500;
-  color: var(--xx-color-text, #111827);
-  margin: 0 0 4px;
-  word-break: break-all;
-}
-
-.xx-clonemodal-upload-hint {
-  font-size: 12px;
-  color: var(--xx-color-text-secondary, #6b7280);
+.xx-clonemodal-asset-empty-text {
   margin: 0;
+  font-size: 13px;
+  color: var(--xx-color-text-secondary, #6b7280);
 }
 
 /* ── 错误提示 ───────────────────────────────────────────── */
 
+/* ── 错误提示 ───────────────────────────────────────────── */
+
 .xx-clonemodal-error {
   display: flex;
   align-items: center;

apps/web/src/components/voice/CloneModal/constants.ts

@@ -7,14 +7,5 @@ export const PROGRESS_STEPS: ProgressStep[] = [
   { key: "done", label: "完成", icon: "✅" },
 ]
 
-/** 支持的音频扩展名 */
-export const ACCEPTED_EXTENSIONS = ["mp3", "wav", "m4a", "webm"]
-
-/** 文件选择器 accept 属性 */
-export const ACCEPTED_MIME = ".mp3,.wav,.m4a,.webm,audio/mpeg,audio/wav,audio/mp4,audio/webm"
-
-/** 最大文件大小：10MB */
-export const MAX_FILE_SIZE = 10 * 1024 * 1024
-
 /** 最长录制时长：5 分钟（秒） */
 export const MAX_RECORD_SECONDS = 5 * 60

apps/web/src/components/voice/CloneModal/index.tsx

@@ -1,10 +1,12 @@
 import React, { useState, useCallback, useRef, useEffect } from "react"
+import { useNavigate } from "react-router-dom"
+import { useQuery } from "@tanstack/react-query"
 import { Modal, Button } from "@/components/ui"
 import { createVoiceClone, toVoiceClone } from "@/api/voice-clone"
-import { uploadAssetDirect, ensureDefaultLibrary } from "@/api/assets"
+import { uploadAssetDirect, ensureDefaultLibrary, getAssetsByKind } from "@/api/assets"
 import { getOrCreateDefaultProject } from "@/api/projects"
-import { PROGRESS_STEPS, ACCEPTED_MIME } from "./constants"
-import { validateFile } from "./utils"
+import { PROGRESS_STEPS } from "./constants"
+import { formatRecordTime } from "./utils"
 import { useAudioRecorder } from "./hooks/useAudioRecorder"
 import type { CloneModalProps, ModalPhase } from "./types"
 import "./clone-modal.css"
@@ -18,15 +20,21 @@ const getExtensionFromMime = (mime: string): string => {
   return "webm"
 }
 
+/** 格式化素材时长（秒 → mm:ss） */
+const formatAssetDuration = (seconds?: number): string => {
+  if (!seconds || seconds <= 0) return "--:--"
+  return formatRecordTime(Math.round(seconds))
+}
+
 const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) => {
+  const navigate = useNavigate()
   const [phase, setPhase] = useState<ModalPhase>("input")
   const [voiceName, setVoiceName] = useState("")
   const [voiceDescription, setVoiceDescription] = useState("")
-  const [selectedFile, setSelectedFile] = useState<File | null>(null)
-  const [dragActive, setDragActive] = useState(false)
+  /** 从配音素材选择的素材 ID */
+  const [selectedAssetId, setSelectedAssetId] = useState<string>("")
   const [errorMessage, setErrorMessage] = useState("")
 
-  const fileInputRef = useRef<HTMLInputElement>(null)
   const timerRef = useRef<ReturnType<typeof setTimeout> | null>(null)
   /** 默认音色名称计数器（组件级 ref，避免多实例串号） */
   const cloneCounterRef = useRef(1)
@@ -34,6 +42,14 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
   const isMountedRef = useRef(true)
   const isSubmittingRef = useRef(false)
 
+  /* ── 配音素材列表（「从配音素材选择」；弹窗打开时才发请求） ────── */
+  const { data: voiceAssets, isLoading: assetsLoading } = useQuery({
+    queryKey: ["assets", "voice", "clone-modal"],
+    queryFn: () => getAssetsByKind("voice", { limit: 100 }),
+    enabled: open,
+    staleTime: 30_000,
+  })
+
   /* ── 录音 Hook ──────────────────────────────────── */
   const {
     isRecording,
@@ -55,10 +71,9 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
     setPhase("input")
     setVoiceName(getNextDefaultName())
     setVoiceDescription("")
-    setSelectedFile(null)
-    setDragActive(false)
-    if (isSubmittingRef.current) return
-    isSubmittingRef.current = true
+    setSelectedAssetId("")
+    // 注意：resetState 不得触碰 isSubmittingRef——提交锁仅属于 handleSubmit；
+    // 此前在此上锁且无复位路径，弹窗打开即死锁
     setErrorMessage("")
     resetRecorder()
   }, [getNextDefaultName, resetRecorder])
@@ -85,65 +100,26 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
     }
   }, [])
 
-  /* ── 文件上传 ──────────────────────────────────── */
+  /* ── 素材/录音互斥：选择素材时清掉录音，开始录音时清掉素材选择 ── */
 
-  const handleUploadClick = () => {
-    fileInputRef.current?.click()
-  }
-
-  const handleFileChange = (e: React.ChangeEvent<HTMLInputElement>) => {
-    const file = e.target.files?.[0]
-    if (file) {
-      const error = validateFile(file)
-      if (error) {
-        setErrorMessage(error)
-        setSelectedFile(null)
-      } else {
-        if (isSubmittingRef.current) return
-        isSubmittingRef.current = true
-        setErrorMessage("")
-        setSelectedFile(file)
-        resetRecorder()
-      }
-    }
-    e.target.value = ""
-  }
-
-  /* ── 拖拽 ──────────────────────────────────────── */
-
-  const handleDrag = (e: React.DragEvent) => {
-    e.preventDefault()
-    e.stopPropagation()
-    if (e.type === "dragenter" || e.type === "dragover") {
-      setDragActive(true)
-    } else if (e.type === "dragleave") {
-      setDragActive(false)
+  const handleSelectAsset = (assetId: string) => {
+    setSelectedAssetId(assetId)
+    if (assetId) {
+      resetRecorder()
     }
   }
 
-  const handleDrop = (e: React.DragEvent) => {
-    e.preventDefault()
-    e.stopPropagation()
-    setDragActive(false)
-    const file = e.dataTransfer.files?.[0]
-    if (file) {
-      const error = validateFile(file)
-      if (error) {
-        setErrorMessage(error)
-        setSelectedFile(null)
-      } else {
-        if (isSubmittingRef.current) return
-        isSubmittingRef.current = true
-        setErrorMessage("")
-        setSelectedFile(file)
-        resetRecorder()
-      }
+  const handleToggleRecord = () => {
+    // 开始录音会清掉已选素材；停止录音保留录音结果
+    if (!isRecording) {
+      setSelectedAssetId("")
     }
+    toggleRecord()
   }
 
   /* ── 计算属性 ──────────────────────────────────── */
 
-  const hasAudio = selectedFile !== null || recordedBlob !== null
+  const hasAudio = selectedAssetId !== "" || recordedBlob !== null
   const isProcessing = phase === "uploading" || phase === "cloning"
   const canSubmit = hasAudio && !isProcessing
 
@@ -158,7 +134,7 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
       return
     }
     if (!hasAudio) {
-      setErrorMessage("请上传音频文件或录制一段声音")
+      setErrorMessage("请从配音素材选择一段音频，或直接录制声音")
       return
     }
 
@@ -167,20 +143,37 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
     setErrorMessage("")
 
     try {
-      // 阶段 1：上传音频
+      // 路径 A：从配音素材选择 → 无需上传，直接克隆
+      if (selectedAssetId) {
+        setPhase("cloning")
+        const result = await createVoiceClone({
+          name,
+          description: voiceDescription.trim() || undefined,
+          asset_id: selectedAssetId,
+        })
+
+        if (!isMountedRef.current) return
+
+        isSubmittingRef.current = false
+        setPhase("done")
+        timerRef.current = setTimeout(() => {
+          if (isMountedRef.current) {
+            onSuccess?.(toVoiceClone(result))
+            handleClose()
+          }
+        }, 2000)
+        return
+      }
+
+      // 路径 B：录音 → 先上传为配音素材，再克隆
       setPhase("uploading")
 
-      let fileToUpload: File
-      if (selectedFile) {
-        fileToUpload = selectedFile
-      } else {
-        // 使用浏览器实际生成的 MIME 类型，避免跨浏览器格式不匹配
-        const mimeType = recordedBlob?.type || "audio/webm"
-        const ext = getExtensionFromMime(mimeType)
-        fileToUpload = new File([recordedBlob!], `recorded-${Date.now()}.${ext}`, {
-          type: mimeType,
-        })
-      }
+      // 使用浏览器实际生成的 MIME 类型，避免跨浏览器格式不匹配
+      const mimeType = recordedBlob?.type || "audio/webm"
+      const ext = getExtensionFromMime(mimeType)
+      const fileToUpload = new File([recordedBlob!], `recorded-${Date.now()}.${ext}`, {
+        type: mimeType,
+      })
 
       // 获取默认项目和素材库
       const project = await getOrCreateDefaultProject()
@@ -226,6 +219,8 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
     }
   }
 
+  const hasAssets = (voiceAssets?.length ?? 0) > 0
+
   return (
     <Modal
       open={open}
@@ -244,7 +239,7 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
           <div className="xx-clonemodal-steps">
             <div className="xx-clonemodal-step xx-clonemodal-step--active">
               <div className="xx-clonemodal-step-number">1</div>
-              <span className="xx-clonemodal-step-label">上传/录制音频</span>
+              <span className="xx-clonemodal-step-label">选择/录制音频</span>
             </div>
             <div className="xx-clonemodal-step-connector" />
             <div className="xx-clonemodal-step">
@@ -274,30 +269,42 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
             <div className="xx-clonemodal-char-count">{voiceName.length}/20</div>
           </div>
 
-          {/* 上传区域 */}
+          {/* 从配音素材选择 */}
           <div className="xx-clonemodal-field">
-            <label className="xx-clonemodal-label">上传音频</label>
-            <div
-              className={`xx-clonemodal-upload-zone${dragActive ? " xx-clonemodal-upload-zone--active" : ""}${selectedFile ? " xx-clonemodal-upload-zone--has-file" : ""}`}
-              onClick={handleUploadClick}
-              onDragEnter={handleDrag}
-              onDragOver={handleDrag}
-              onDragLeave={handleDrag}
-              onDrop={handleDrop}
-            >
-              <div className="xx-clonemodal-upload-icon">{selectedFile ? "📄" : "🎵"}</div>
-              <p className="xx-clonemodal-upload-title">
-                {selectedFile ? selectedFile.name : "拖拽音频文件到此处，或点击上传"}
-              </p>
-              <p className="xx-clonemodal-upload-hint">支持 MP3、WAV、M4A 格式，最大 10MB</p>
-              <input
-                ref={fileInputRef}
-                type="file"
-                accept={ACCEPTED_MIME}
-                style={{ display: "none" }}
-                onChange={handleFileChange}
-              />
-            </div>
+            <label className="xx-clonemodal-label">从配音素材选择</label>
+            {hasAssets ? (
+              <select
+                className="xx-clonemodal-input"
+                value={selectedAssetId}
+                onChange={(e) => handleSelectAsset(e.target.value)}
+                disabled={assetsLoading}
+              >
+                <option value="">{assetsLoading ? "素材加载中…" : "请选择已上传的配音素材"}</option>
+                {voiceAssets!.map((asset) => (
+                  <option key={asset.id} value={asset.id}>
+                    {asset.name}（{formatAssetDuration(asset.duration)}）
+                  </option>
+                ))}
+              </select>
+            ) : (
+              <div className="xx-clonemodal-asset-empty">
+                <p className="xx-clonemodal-asset-empty-text">
+                  {assetsLoading ? "素材加载中…" : "请先在配音库上传素材"}
+                </p>
+                {!assetsLoading && (
+                  <Button
+                    buttonType="ghost"
+                    buttonSize="sm"
+                    onClick={() => {
+                      handleClose()
+                      navigate("/app/voice-materials")
+                    }}
+                  >
+                    去配音库上传
+                  </Button>
+                )}
+              </div>
+            )}
           </div>
 
           {/* 或分隔 */}
@@ -332,7 +339,7 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
               <button
                 type="button"
                 className={`xx-clonemodal-record-btn${isRecording ? " xx-clonemodal-record-btn--recording" : ""}`}
-                onClick={toggleRecord}
+                onClick={handleToggleRecord}
                 title={isRecording ? "停止录制" : "开始录制"}
               >
                 {isRecording ? "⏹" : "🎙️"}
@@ -365,7 +372,7 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
           {/* 提示 */}
           <div className="xx-clonemodal-tip">
             <span className="xx-clonemodal-tip-icon">💡</span>
-            <span>建议上传 10 秒 ~ 3 分钟的清晰语音，环境安静、语速均匀效果最佳</span>
+            <span>建议使用 10 秒 ~ 3 分钟的清晰语音，环境安静、语速均匀效果最佳</span>
           </div>
 
           {/* 底部按钮 */}
@@ -415,18 +422,18 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
           {/* 当前阶段描述 */}
           <div className="xx-clonemodal-progress-info">
             {phase === "uploading" && (
-              <>
+              <div>
                 <div className="xx-clonemodal-progress-spinner" />
-                <p className="xx-clonemodal-progress-text">正在上传音频文件…</p>
-                <p className="xx-clonemodal-progress-sub">请稍候，正在将音频上传至服务器</p>
-              </>
+                <p className="xx-clonemodal-progress-text">正在上传录音…</p>
+                <p className="xx-clonemodal-progress-sub">请稍候，正在将录音上传至服务器</p>
+              </div>
             )}
             {phase === "cloning" && (
-              <>
+              <div>
                 <div className="xx-clonemodal-progress-spinner xx-clonemodal-progress-spinner--cloning" />
                 <p className="xx-clonemodal-progress-text">AI 正在克隆你的声音…</p>
                 <p className="xx-clonemodal-progress-sub">正在分析声音特征，生成专属音色模型</p>
-              </>
+              </div>
             )}
           </div>
         </div>

apps/web/src/components/voice/CloneModal/utils.ts

@@ -1,20 +1,3 @@
-import { ACCEPTED_EXTENSIONS, MAX_FILE_SIZE } from "./constants"
-
-/**
- * 验证音频文件
- * @returns 错误信息，null 表示验证通过
- */
-export const validateFile = (file: File): string | null => {
-  const ext = file.name.split(".").pop()?.toLowerCase()
-  if (!ext || !ACCEPTED_EXTENSIONS.includes(ext)) {
-    return "不支持的音频格式，请上传 MP3、WAV 或 M4A 文件"
-  }
-  if (file.size > MAX_FILE_SIZE) {
-    return "文件大小超过 10MB，请压缩后重试"
-  }
-  return null
-}
-
 /** 格式化录制时间 mm:ss */
 export const formatRecordTime = (seconds: number): string => {
   const m = Math.floor(seconds / 60)

apps/web/src/components/voice/clone-modal/InputView.tsx

@@ -1,142 +0,0 @@
-import React from "react"
-import { Button } from "@/components/ui"
-import UploadZone from "./UploadZone"
-import RecordArea from "./RecordArea"
-import StepIndicator from "./StepIndicator"
-import { MAX_VOICE_NAME_LENGTH, MAX_VOICE_DESC_LENGTH } from "../constants/cloneModal"
-
-interface InputViewProps {
-  voiceName: string
-  voiceDescription: string
-  selectedFile: File | null
-  dragActive: boolean
-  isRecording: boolean
-  recordTime: number
-  recordedBlob: Blob | null
-  errorMessage: string
-  canSubmit: boolean
-  onVoiceNameChange: (value: string) => void
-  onVoiceDescChange: (value: string) => void
-  onDragActiveChange: (active: boolean) => void
-  onFileSelect: (file: File | null, error: string) => void
-  onRecordToggle: () => void
-  onClose: () => void
-  onSubmit: () => void
-}
-
-const INPUT_STEPS = ["上传/录制音频", "填写信息", "提交克隆"]
-
-const InputView: React.FC<InputViewProps> = ({
-  voiceName,
-  voiceDescription,
-  selectedFile,
-  dragActive,
-  isRecording,
-  recordTime,
-  recordedBlob,
-  errorMessage,
-  canSubmit,
-  onVoiceNameChange,
-  onVoiceDescChange,
-  onDragActiveChange,
-  onFileSelect,
-  onRecordToggle,
-  onClose,
-  onSubmit,
-}) => {
-  return (
-    <div className="xx-clonemodal-body">
-      {/* 步骤引导 */}
-      <StepIndicator currentStep={0} steps={INPUT_STEPS} />
-
-      {/* 音色名称 */}
-      <div className="xx-clonemodal-field">
-        <label className="xx-clonemodal-label">
-          音色名称 <span className="xx-clonemodal-required">*</span>
-        </label>
-        <input
-          type="text"
-          className="xx-clonemodal-input"
-          value={voiceName}
-          onChange={(e) => onVoiceNameChange(e.target.value)}
-          placeholder="输入音色名称（2-20字符）"
-          maxLength={MAX_VOICE_NAME_LENGTH}
-        />
-        <div className="xx-clonemodal-char-count">
-          {voiceName.length}/{MAX_VOICE_NAME_LENGTH}
-        </div>
-      </div>
-
-      {/* 上传区域 */}
-      <div className="xx-clonemodal-field">
-        <label className="xx-clonemodal-label">上传音频</label>
-        <UploadZone
-          selectedFile={selectedFile}
-          dragActive={dragActive}
-          onDragActiveChange={onDragActiveChange}
-          onFileSelect={onFileSelect}
-        />
-      </div>
-
-      {/* 或分隔 */}
-      <div className="xx-clonemodal-divider">
-        <div className="xx-clonemodal-divider-line" />
-        <span className="xx-clonemodal-divider-text">或</span>
-        <div className="xx-clonemodal-divider-line" />
-      </div>
-
-      {/* 录制区域 */}
-      <div className="xx-clonemodal-field">
-        <label className="xx-clonemodal-label">直接录制</label>
-        <RecordArea
-          isRecording={isRecording}
-          recordTime={recordTime}
-          recordedBlob={recordedBlob}
-          onRecordToggle={onRecordToggle}
-        />
-      </div>
-
-      {/* 音色描述 */}
-      <div className="xx-clonemodal-field">
-        <label className="xx-clonemodal-label">音色描述</label>
-        <textarea
-          className="xx-clonemodal-textarea"
-          value={voiceDescription}
-          onChange={(e) => onVoiceDescChange(e.target.value)}
-          placeholder="可选，描述这个音色的特点（最多100字符）"
-          maxLength={MAX_VOICE_DESC_LENGTH}
-          rows={3}
-        />
-        <div className="xx-clonemodal-char-count">
-          {voiceDescription.length}/{MAX_VOICE_DESC_LENGTH}
-        </div>
-      </div>
-
-      {/* 错误提示 */}
-      {errorMessage && (
-        <div className="xx-clonemodal-error">
-          <span className="xx-clonemodal-error-icon">⚠️</span>
-          <span>{errorMessage}</span>
-        </div>
-      )}
-
-      {/* 提示 */}
-      <div className="xx-clonemodal-tip">
-        <span className="xx-clonemodal-tip-icon">💡</span>
-        <span>建议上传 10 秒 ~ 3 分钟的清晰语音，环境安静、语速均匀效果最佳</span>
-      </div>
-
-      {/* 底部按钮 */}
-      <div className="xx-clonemodal-footer">
-        <Button buttonType="ghost" onClick={onClose}>
-          取消
-        </Button>
-        <Button buttonType="primary" disabled={!canSubmit} onClick={onSubmit}>
-          🎤 开始克隆
-        </Button>
-      </div>
-    </div>
-  )
-}
-
-export default InputView

apps/web/src/components/voice/clone-modal/ProgressView.tsx

@@ -1,92 +0,0 @@
-import React from "react"
-import { PROGRESS_STEPS } from "../constants/cloneModal"
-import type { ProgressStep } from "../types/cloneModal"
-import type { ModalPhase } from "../types/cloneModal"
-
-interface ProgressViewProps {
-  phase: ModalPhase
-}
-
-const getProgressIndex = (phase: ModalPhase): number => {
-  switch (phase) {
-    case "uploading":
-      return 0
-    case "cloning":
-      return 1
-    case "done":
-      return 2
-    default:
-      return -1
-  }
-}
-
-const ProgressView: React.FC<ProgressViewProps> = ({ phase }) => {
-  const progressIndex = getProgressIndex(phase)
-  const isDone = phase === "done"
-
-  return (
-    <div className="xx-clonemodal-progress-body">
-      {/* 步骤指示器 */}
-      <div className="xx-clonemodal-steps-progress">
-        {PROGRESS_STEPS.map((step: ProgressStep, idx: number) => {
-          const isActive = idx === progressIndex && !isDone
-          const stepDone = idx < progressIndex || isDone
-          const stepClass = [
-            "xx-clonemodal-step-progress",
-            isActive ? "xx-clonemodal-step-progress--active" : "",
-            stepDone ? "xx-clonemodal-step-progress--done" : "",
-          ]
-            .filter(Boolean)
-            .join(" ")
-
-          return (
-            <React.Fragment key={step.key}>
-              {idx > 0 && (
-                <div
-                  className={`xx-clonemodal-step-connector${stepDone ? " xx-clonemodal-step-connector--done" : ""}`}
-                />
-              )}
-              <div className={stepClass}>
-                <div className="xx-clonemodal-step-icon">{stepDone ? "✓" : step.icon}</div>
-                <span className="xx-clonemodal-step-label">{step.label}</span>
-              </div>
-            </React.Fragment>
-          )
-        })}
-      </div>
-
-      {/* 完成阶段 */}
-      {isDone && (
-        <div className="xx-clonemodal-success">
-          <div className="xx-clonemodal-success-icon">🎉</div>
-          <h3 className="xx-clonemodal-success-title">克隆已提交</h3>
-          <p className="xx-clonemodal-success-desc">
-            音色正在生成中，完成后将出现在「我的克隆」列表中
-          </p>
-        </div>
-      )}
-
-      {/* 进行中阶段 */}
-      {!isDone && (
-        <div className="xx-clonemodal-progress-info">
-          {phase === "uploading" && (
-            <>
-              <div className="xx-clonemodal-progress-spinner" />
-              <p className="xx-clonemodal-progress-text">正在上传音频文件…</p>
-              <p className="xx-clonemodal-progress-sub">请稍候，正在将音频上传至服务器</p>
-            </>
-          )}
-          {phase === "cloning" && (
-            <>
-              <div className="xx-clonemodal-progress-spinner xx-clonemodal-progress-spinner--cloning" />
-              <p className="xx-clonemodal-progress-text">AI 正在克隆你的声音…</p>
-              <p className="xx-clonemodal-progress-sub">正在分析声音特征，生成专属音色模型</p>
-            </>
-          )}
-        </div>
-      )}
-    </div>
-  )
-}
-
-export default ProgressView

apps/web/src/components/voice/clone-modal/RecordArea.tsx

@@ -1,49 +0,0 @@
-import React from "react"
-import { formatRecordTime } from "../utils/cloneModal"
-
-interface RecordAreaProps {
-  isRecording: boolean
-  recordTime: number
-  recordedBlob: Blob | null
-  onRecordToggle: () => void
-}
-
-const RecordArea: React.FC<RecordAreaProps> = ({
-  isRecording,
-  recordTime,
-  recordedBlob,
-  onRecordToggle,
-}) => {
-  const getHintText = () => {
-    if (isRecording) return `录制中 ${formatRecordTime(recordTime)}`
-    if (recordedBlob) return `已录制 ${formatRecordTime(recordTime)}`
-    return "点击按钮开始录制（最长 5 分钟）"
-  }
-
-  return (
-    <div className="xx-clonemodal-record-area">
-      <div className="xx-clonemodal-record-info">
-        <p className="xx-clonemodal-record-hint">{getHintText()}</p>
-        {isRecording && (
-          <div className="xx-clonemodal-record-wave">
-            <span className="xx-clonemodal-record-wave-bar" />
-            <span className="xx-clonemodal-record-wave-bar" />
-            <span className="xx-clonemodal-record-wave-bar" />
-            <span className="xx-clonemodal-record-wave-bar" />
-            <span className="xx-clonemodal-record-wave-bar" />
-          </div>
-        )}
-      </div>
-      <button
-        type="button"
-        className={`xx-clonemodal-record-btn${isRecording ? " xx-clonemodal-record-btn--recording" : ""}`}
-        onClick={onRecordToggle}
-        title={isRecording ? "停止录制" : "开始录制"}
-      >
-        {isRecording ? "⏹" : "🎙️"}
-      </button>
-    </div>
-  )
-}
-
-export default RecordArea

apps/web/src/components/voice/clone-modal/StepIndicator.tsx

@@ -1,30 +0,0 @@
-import React from "react"
-
-interface StepIndicatorProps {
-  currentStep: number
-  steps: string[]
-}
-
-/**
- * 输入阶段顶部的步骤引导（数字步骤）
- */
-const StepIndicator: React.FC<StepIndicatorProps> = ({ currentStep, steps }) => {
-  return (
-    <div className="xx-clonemodal-steps">
-      {steps.map((label, idx) => {
-        const isActive = idx <= currentStep
-        return (
-          <React.Fragment key={idx}>
-            {idx > 0 && <div className="xx-clonemodal-step-connector" />}
-            <div className={`xx-clonemodal-step${isActive ? " xx-clonemodal-step--active" : ""}`}>
-              <div className="xx-clonemodal-step-number">{idx + 1}</div>
-              <span className="xx-clonemodal-step-label">{label}</span>
-            </div>
-          </React.Fragment>
-        )
-      })}
-    </div>
-  )
-}
-
-export default StepIndicator

apps/web/src/components/voice/clone-modal/UploadZone.tsx

@@ -1,79 +0,0 @@
-import React, { useRef } from "react"
-import { ACCEPTED_MIME } from "../constants/cloneModal"
-import { validateFile } from "../utils/cloneModal"
-
-interface UploadZoneProps {
-  selectedFile: File | null
-  dragActive: boolean
-  onDragActiveChange: (active: boolean) => void
-  onFileSelect: (file: File | null, error: string) => void
-}
-
-const UploadZone: React.FC<UploadZoneProps> = ({
-  selectedFile,
-  dragActive,
-  onDragActiveChange,
-  onFileSelect,
-}) => {
-  const fileInputRef = useRef<HTMLInputElement>(null)
-
-  const handleUploadClick = () => {
-    fileInputRef.current?.click()
-  }
-
-  const handleFileChange = (e: React.ChangeEvent<HTMLInputElement>) => {
-    const file = e.target.files?.[0]
-    if (file) {
-      const error = validateFile(file)
-      onFileSelect(error ? null : file, error || "")
-    }
-    e.target.value = ""
-  }
-
-  const handleDrag = (e: React.DragEvent) => {
-    e.preventDefault()
-    e.stopPropagation()
-    if (e.type === "dragenter" || e.type === "dragover") {
-      onDragActiveChange(true)
-    } else if (e.type === "dragleave") {
-      onDragActiveChange(false)
-    }
-  }
-
-  const handleDrop = (e: React.DragEvent) => {
-    e.preventDefault()
-    e.stopPropagation()
-    onDragActiveChange(false)
-    const file = e.dataTransfer.files?.[0]
-    if (file) {
-      const error = validateFile(file)
-      onFileSelect(error ? null : file, error || "")
-    }
-  }
-
-  return (
-    <div
-      className={`xx-clonemodal-upload-zone${dragActive ? " xx-clonemodal-upload-zone--active" : ""}${selectedFile ? " xx-clonemodal-upload-zone--has-file" : ""}`}
-      onClick={handleUploadClick}
-      onDragEnter={handleDrag}
-      onDragOver={handleDrag}
-      onDragLeave={handleDrag}
-      onDrop={handleDrop}
-    >
-      <div className="xx-clonemodal-upload-icon">{selectedFile ? "📄" : "🎵"}</div>
-      <p className="xx-clonemodal-upload-title">
-        {selectedFile ? selectedFile.name : "拖拽音频文件到此处，或点击上传"}
-      </p>
-      <p className="xx-clonemodal-upload-hint">支持 MP3、WAV、M4A 格式，最大 10MB</p>
-      <input
-        ref={fileInputRef}
-        type="file"
-        accept={ACCEPTED_MIME}
-        style={{ display: "none" }}
-        onChange={handleFileChange}
-      />
-    </div>
-  )
-}
-
-export default UploadZone

apps/web/src/components/voice/constants/cloneModal.ts

@@ -1,29 +0,0 @@
-import type { ProgressStep } from "../types/cloneModal"
-
-/** 进度阶段配置 */
-export const PROGRESS_STEPS: ProgressStep[] = [
-  { key: "uploading", label: "上传中", icon: "📤" },
-  { key: "cloning", label: "克隆中", icon: "🧬" },
-  { key: "done", label: "完成", icon: "✅" },
-]
-
-/** 支持的音频扩展名 */
-export const ACCEPTED_EXTENSIONS = ["mp3", "wav", "m4a"]
-
-/** input accept 属性值 */
-export const ACCEPTED_MIME = ".mp3,.wav,.m4a,audio/mpeg,audio/wav,audio/mp4"
-
-/** 最大文件大小：10MB */
-export const MAX_FILE_SIZE = 10 * 1024 * 1024
-
-/** 最长录制时长（秒）：5 分钟 */
-export const MAX_RECORD_SECONDS = 5 * 60
-
-/** 音色名称最小长度 */
-export const MIN_VOICE_NAME_LENGTH = 2
-
-/** 音色名称最大长度 */
-export const MAX_VOICE_NAME_LENGTH = 20
-
-/** 音色描述最大长度 */
-export const MAX_VOICE_DESC_LENGTH = 100

apps/web/src/components/voice/hooks/useAudioRecorder.ts

@@ -1,119 +0,0 @@
-import { useState, useRef, useCallback, useEffect } from "react"
-import { MAX_RECORD_SECONDS } from "../constants/cloneModal"
-
-interface UseAudioRecorderReturn {
-  isRecording: boolean
-  recordTime: number
-  recordedBlob: Blob | null
-  toggleRecording: () => void
-  resetRecording: () => void
-}
-
-/**
- * 录音 Hook —— 封装 MediaRecorder 录音逻辑
- */
-const useAudioRecorder = (): UseAudioRecorderReturn => {
-  const [isRecording, setIsRecording] = useState(false)
-  const [recordTime, setRecordTime] = useState(0)
-  const [recordedBlob, setRecordedBlob] = useState<Blob | null>(null)
-
-  const recordTimerRef = useRef<ReturnType<typeof setInterval> | null>(null)
-  const mediaRecorderRef = useRef<MediaRecorder | null>(null)
-  const audioChunksRef = useRef<Blob[]>([])
-
-  const stopRecording = useCallback(() => {
-    setIsRecording(false)
-    if (recordTimerRef.current) {
-      clearInterval(recordTimerRef.current)
-      recordTimerRef.current = null
-    }
-    if (mediaRecorderRef.current && mediaRecorderRef.current.state !== "inactive") {
-      mediaRecorderRef.current.stop()
-    }
-  }, [])
-
-  const startRecording = useCallback(async () => {
-    try {
-      const stream = await navigator.mediaDevices.getUserMedia({ audio: true })
-      const mediaRecorder = new MediaRecorder(stream)
-      mediaRecorderRef.current = mediaRecorder
-      audioChunksRef.current = []
-
-      mediaRecorder.ondataavailable = (event) => {
-        if (event.data.size > 0) {
-          audioChunksRef.current.push(event.data)
-        }
-      }
-
-      mediaRecorder.onstop = () => {
-        const blob = new Blob(audioChunksRef.current, { type: "audio/webm" })
-        setRecordedBlob(blob)
-        stream.getTracks().forEach((track) => track.stop())
-      }
-
-      mediaRecorder.start()
-      setIsRecording(true)
-      setRecordTime(0)
-      setRecordedBlob(null)
-
-      recordTimerRef.current = setInterval(() => {
-        setRecordTime((prev) => {
-          const next = prev + 1
-          if (next >= MAX_RECORD_SECONDS) {
-            setTimeout(() => {
-              stopRecording()
-            }, 0)
-            return MAX_RECORD_SECONDS
-          }
-          return next
-        })
-      }, 1000)
-    } catch {
-      // 错误由调用方通过其他机制提示
-      setIsRecording(false)
-    }
-  }, [stopRecording])
-
-  const toggleRecording = useCallback(() => {
-    if (isRecording) {
-      stopRecording()
-    } else {
-      startRecording()
-    }
-  }, [isRecording, startRecording, stopRecording])
-
-  const resetRecording = useCallback(() => {
-    setIsRecording(false)
-    setRecordTime(0)
-    setRecordedBlob(null)
-    audioChunksRef.current = []
-    if (recordTimerRef.current) {
-      clearInterval(recordTimerRef.current)
-      recordTimerRef.current = null
-    }
-    if (mediaRecorderRef.current && mediaRecorderRef.current.state !== "inactive") {
-      mediaRecorderRef.current.stop()
-    }
-    mediaRecorderRef.current = null
-  }, [])
-
-  // 卸载时清理
-  useEffect(() => {
-    return () => {
-      if (recordTimerRef.current) clearInterval(recordTimerRef.current)
-      if (mediaRecorderRef.current && mediaRecorderRef.current.state !== "inactive") {
-        mediaRecorderRef.current.stop()
-      }
-    }
-  }, [])
-
-  return {
-    isRecording,
-    recordTime,
-    recordedBlob,
-    toggleRecording,
-    resetRecording,
-  }
-}
-
-export default useAudioRecorder

apps/web/src/components/voice/hooks/useCloneFormState.ts

@@ -1,134 +0,0 @@
-import { useState, useRef, useCallback, useEffect } from "react"
-import type { ModalPhase } from "../types/cloneModal"
-import { MIN_VOICE_NAME_LENGTH, MAX_VOICE_NAME_LENGTH } from "../constants/cloneModal"
-import useAudioRecorder from "./useAudioRecorder"
-
-/**
- * 克隆弹窗表单状态 Hook
- * 管理表单字段、录音、文件选择、验证逻辑
- */
-export function useCloneFormState({ open, onClose }: { open: boolean; onClose: () => void }) {
-  const [phase, setPhase] = useState<ModalPhase>("input")
-  const [voiceName, setVoiceName] = useState("")
-  const [voiceDescription, setVoiceDescription] = useState("")
-  const [selectedFile, setSelectedFile] = useState<File | null>(null)
-  const [dragActive, setDragActive] = useState(false)
-  const [errorMessage, setErrorMessage] = useState("")
-
-  const { isRecording, recordTime, recordedBlob, toggleRecording, resetRecording } =
-    useAudioRecorder()
-
-  /** 默认音色名称计数器 */
-  const cloneCounterRef = useRef(1)
-
-  const getNextDefaultName = useCallback((): string => {
-    const name = `我的声音 ${cloneCounterRef.current}`
-    cloneCounterRef.current += 1
-    return name
-  }, [])
-
-  const hasAudio = selectedFile !== null || recordedBlob !== null
-
-  const canSubmit =
-    voiceName.trim().length >= MIN_VOICE_NAME_LENGTH &&
-    voiceName.trim().length <= MAX_VOICE_NAME_LENGTH &&
-    hasAudio
-
-  const isProcessing = phase === "uploading" || phase === "cloning"
-
-  /** 重置弹窗状态 */
-  const resetState = useCallback(() => {
-    setPhase("input")
-    setVoiceName(getNextDefaultName())
-    setVoiceDescription("")
-    setSelectedFile(null)
-    setDragActive(false)
-    setErrorMessage("")
-    resetRecording()
-  }, [getNextDefaultName, resetRecording])
-
-  /** 关闭弹窗 */
-  const handleClose = useCallback(() => {
-    resetState()
-    onClose()
-  }, [resetState, onClose])
-
-  /** 弹窗打开时重置状态 */
-  useEffect(() => {
-    if (open) {
-      resetState()
-    }
-  }, [open, resetState])
-
-  /** 选择文件（来自上传或拖拽） */
-  const handleFileSelect = useCallback(
-    (file: File | null, error: string) => {
-      if (error) {
-        setErrorMessage(error)
-        setSelectedFile(null)
-      } else {
-        setErrorMessage("")
-        setSelectedFile(file)
-        // 清除录音
-        resetRecording()
-      }
-    },
-    [resetRecording],
-  )
-
-  /** 录音切换 */
-  const handleRecordToggle = useCallback(() => {
-    setErrorMessage("")
-    if (isRecording) {
-      toggleRecording()
-    } else {
-      // 开始录制前清除已选文件
-      setSelectedFile(null)
-      toggleRecording()
-    }
-  }, [isRecording, toggleRecording])
-
-  /** 表单验证 */
-  const validateForm = useCallback((): string | null => {
-    const name = voiceName.trim()
-    if (!name) {
-      return "请输入音色名称"
-    }
-    if (name.length < MIN_VOICE_NAME_LENGTH || name.length > MAX_VOICE_NAME_LENGTH) {
-      return `音色名称需在 ${MIN_VOICE_NAME_LENGTH}-${MAX_VOICE_NAME_LENGTH} 个字符之间`
-    }
-    if (!hasAudio) {
-      return "请上传音频文件或录制一段声音"
-    }
-    return null
-  }, [voiceName, hasAudio])
-
-  return {
-    // 状态
-    phase,
-    setPhase,
-    voiceName,
-    setVoiceName,
-    voiceDescription,
-    setVoiceDescription,
-    selectedFile,
-    dragActive,
-    setDragActive,
-    errorMessage,
-    setErrorMessage,
-    // 录音
-    isRecording,
-    recordTime,
-    recordedBlob,
-    // 计算属性
-    hasAudio,
-    canSubmit,
-    isProcessing,
-    // handlers
-    handleFileSelect,
-    handleRecordToggle,
-    handleClose,
-    validateForm,
-    resetState,
-  }
-}

apps/web/src/components/voice/hooks/useCloneModal.ts

@@ -1,46 +0,0 @@
-import type { CloneModalProps } from "../types/cloneModal"
-import { useCloneFormState } from "./useCloneFormState"
-import { useCloneSubmit } from "./useCloneSubmit"
-
-/**
- * 音色克隆弹窗主业务 Hook
- * 组合表单状态 + 提交流程两个子 Hook
- */
-const useCloneModal = ({ open, onClose, onSuccess }: CloneModalProps) => {
-  const formState = useCloneFormState({ open, onClose })
-
-  const { handleSubmit } = useCloneSubmit({
-    voiceName: formState.voiceName,
-    voiceDescription: formState.voiceDescription,
-    selectedFile: formState.selectedFile,
-    recordedBlob: formState.recordedBlob,
-    setPhase: formState.setPhase,
-    setErrorMessage: formState.setErrorMessage,
-    validateForm: formState.validateForm,
-    onSuccess,
-    onClose: formState.handleClose,
-  })
-
-  return {
-    phase: formState.phase,
-    voiceName: formState.voiceName,
-    voiceDescription: formState.voiceDescription,
-    selectedFile: formState.selectedFile,
-    dragActive: formState.dragActive,
-    errorMessage: formState.errorMessage,
-    isRecording: formState.isRecording,
-    recordTime: formState.recordTime,
-    recordedBlob: formState.recordedBlob,
-    canSubmit: formState.canSubmit,
-    isProcessing: formState.isProcessing,
-    setVoiceName: formState.setVoiceName,
-    setVoiceDescription: formState.setVoiceDescription,
-    setDragActive: formState.setDragActive,
-    handleFileSelect: formState.handleFileSelect,
-    handleRecordToggle: formState.handleRecordToggle,
-    handleClose: formState.handleClose,
-    handleSubmit,
-  }
-}
-
-export default useCloneModal

apps/web/src/components/voice/hooks/useCloneSubmit.ts

@@ -1,108 +0,0 @@
-import { useRef, useCallback, useEffect } from "react"
-import { createVoiceClone, toVoiceClone } from "@/api/voice-clone"
-import { uploadAssetDirect, ensureDefaultLibrary } from "@/api/assets"
-import { getOrCreateDefaultProject } from "@/api/projects"
-import type { VoiceClone } from "@/api/voice-clone"
-
-interface UseCloneSubmitOptions {
-  voiceName: string
-  voiceDescription: string
-  selectedFile: File | null
-  recordedBlob: Blob | null
-  setPhase: (phase: "input" | "uploading" | "cloning" | "done") => void
-  setErrorMessage: (msg: string) => void
-  validateForm: () => string | null
-  onSuccess?: (clone: VoiceClone) => void
-  onClose: () => void
-}
-
-/**
- * 克隆提交流程 Hook
- * 封装上传 + 克隆 + 完成的三阶段流程
- */
-export function useCloneSubmit({
-  voiceName,
-  voiceDescription,
-  selectedFile,
-  recordedBlob,
-  setPhase,
-  setErrorMessage,
-  validateForm,
-  onSuccess,
-  onClose,
-}: UseCloneSubmitOptions) {
-  const timerRef = useRef<ReturnType<typeof setTimeout> | null>(null)
-
-  /** 组件卸载时清理定时器 */
-  useEffect(() => {
-    return () => {
-      if (timerRef.current) clearTimeout(timerRef.current)
-    }
-  }, [])
-
-  const handleSubmit = useCallback(async () => {
-    const formError = validateForm()
-    if (formError) {
-      setErrorMessage(formError)
-      return
-    }
-
-    setErrorMessage("")
-
-    try {
-      // 阶段 1：上传音频
-      setPhase("uploading")
-
-      let fileToUpload: File
-      if (selectedFile) {
-        fileToUpload = selectedFile
-      } else {
-        fileToUpload = new File([recordedBlob!], `recorded-${Date.now()}.webm`, {
-          type: "audio/webm",
-        })
-      }
-
-      // 获取默认项目和素材库
-      const project = await getOrCreateDefaultProject()
-      const library = await ensureDefaultLibrary({ project_id: project.id, kind: "voice" })
-
-      // 直传到 OSS
-      const uploadResult = await uploadAssetDirect({
-        file: fileToUpload,
-        library_id: library.id,
-      })
-
-      // 阶段 2：克隆
-      setPhase("cloning")
-      const result = await createVoiceClone({
-        name: voiceName.trim(),
-        description: voiceDescription.trim() || undefined,
-        audio_url: uploadResult.url,
-      })
-
-      // 阶段 3：完成
-      setPhase("done")
-
-      // 2秒后自动关闭
-      timerRef.current = setTimeout(() => {
-        onSuccess?.(toVoiceClone(result))
-        onClose()
-      }, 2000)
-    } catch (err) {
-      setPhase("input")
-      setErrorMessage(err instanceof Error ? err.message : "克隆失败，请重试")
-    }
-  }, [
-    validateForm,
-    selectedFile,
-    recordedBlob,
-    voiceName,
-    voiceDescription,
-    setPhase,
-    setErrorMessage,
-    onSuccess,
-    onClose,
-  ])
-
-  return { handleSubmit }
-}

apps/web/src/components/voice/types/cloneModal.ts

@@ -1,20 +0,0 @@
-import type { VoiceClone } from "@/api/voice-clone"
-
-/** 弹窗阶段 */
-export type ModalPhase = "input" | "uploading" | "cloning" | "done"
-
-export interface CloneModalProps {
-  /** 弹窗是否可见 */
-  open: boolean
-  /** 关闭弹窗回调 */
-  onClose: () => void
-  /** 克隆成功回调（返回新创建的音色） */
-  onSuccess?: (voice: VoiceClone) => void
-}
-
-/** 进度步骤项 */
-export interface ProgressStep {
-  key: string
-  label: string
-  icon: string
-}

apps/web/src/components/voice/utils/cloneModal.ts

@@ -1,25 +0,0 @@
-import { ACCEPTED_EXTENSIONS, MAX_FILE_SIZE } from "../constants/cloneModal"
-
-/**
- * 格式化录制时间 mm:ss
- */
-export const formatRecordTime = (seconds: number): string => {
-  const m = Math.floor(seconds / 60)
-  const s = seconds % 60
-  return `${m.toString().padStart(2, "0")}:${s.toString().padStart(2, "0")}`
-}
-
-/**
- * 验证上传的音频文件
- * @returns 错误信息，null 表示验证通过
- */
-export const validateFile = (file: File): string | null => {
-  const ext = file.name.split(".").pop()?.toLowerCase()
-  if (!ext || !ACCEPTED_EXTENSIONS.includes(ext)) {
-    return "不支持的音频格式，请上传 MP3、WAV 或 M4A 文件"
-  }
-  if (file.size > MAX_FILE_SIZE) {
-    return "文件大小超过 10MB，请压缩后重试"
-  }
-  return null
-}

apps/web/src/pages/assets/AssetLibrary.tsx

@@ -20,6 +20,7 @@ import LibrarySidebar from "@/pages/assets/components/LibrarySidebar"
 import AssetFilterBar from "@/pages/assets/components/AssetFilterBar"
 import BatchOperationBar from "@/pages/assets/components/BatchOperationBar"
 import AssetUploadZone from "@/pages/assets/components/AssetUploadZone"
+import UploadQueuePanel from "@/pages/assets/components/UploadQueuePanel"
 import AssetGridSection from "@/pages/assets/components/AssetGridSection"
 import AssetModals from "@/pages/assets/components/AssetModals"
 import { useAssetsData } from "@/pages/assets/hooks/useAssetsData"
@@ -69,7 +70,30 @@ const AssetLibrary: React.FC = () => {
   })
 
   /* ── 上传 ── */
-  const { uploading, uploadProgress, handleUpload } = useAssetUpload({ effectiveLibId })
+  const {
+    uploadItems,
+    enqueueUploads,
+    retryUpload,
+    removeUpload,
+    clearFinished,
+    uploading,
+    activeCount,
+    pendingCount,
+  } = useAssetUpload({ effectiveLibId })
+
+  /* ── 上传中 asset_id → 进度/状态映射，合并进网格卡片展示真实进度 ── */
+  const uploadProgressMap = React.useMemo(() => {
+    const map = new Map<string, { progress: number; uploading: boolean }>()
+    for (const it of uploadItems) {
+      if (it.assetId && (it.status === "uploading" || it.status === "ingesting")) {
+        map.set(it.assetId, {
+          progress: it.status === "ingesting" ? 100 : it.progress,
+          uploading: it.status === "uploading",
+        })
+      }
+    }
+    return map
+  }, [uploadItems])
 
   /* ── 选中态管理 ── */
   const { selectedIds, setSelectedIds, toggleSelect, selectAll, deselectAll } = useAssetSelection({
@@ -144,8 +168,17 @@ const AssetLibrary: React.FC = () => {
           {/* 上传区域 */}
           <AssetUploadZone
             uploading={uploading}
-            uploadProgress={uploadProgress}
-            onUpload={handleUpload}
+            activeCount={activeCount}
+            pendingCount={pendingCount}
+            onUpload={enqueueUploads}
+          />
+
+          {/* 上传队列：独立进度 + 失败重试/移除 */}
+          <UploadQueuePanel
+            items={uploadItems}
+            onRetry={retryUpload}
+            onRemove={removeUpload}
+            onClearFinished={clearFinished}
           />
 
           {/* 筛选栏 */}
@@ -180,6 +213,7 @@ const AssetLibrary: React.FC = () => {
             assets={filteredAssets}
             selectedIds={selectedIds}
             diagnosingId={diagnosingId}
+            uploadProgressMap={uploadProgressMap}
             onRetry={refetchAssets}
             onToggleSelect={toggleSelect}
             onDiagnose={handleDiagnose}
@@ -191,8 +225,6 @@ const AssetLibrary: React.FC = () => {
 
       {/* ─── 弹窗集合 ─── */}
       <AssetModals
-        uploading={uploading}
-        uploadProgress={uploadProgress}
         createModalOpen={createModalOpen}
         onCreateModalCancel={() => setCreateModalOpen(false)}
         onCreateModalOk={handleCreateLibrary}

apps/web/src/pages/assets/assets.css

@@ -147,44 +147,52 @@
 /* ============================================================
    上传区域
    ============================================================ */
-.xx-asset-upload-zone {
-  border: 2px dashed var(--border-color);
-  border-radius: var(--radius-lg);
-  padding: var(--space-2xl) var(--space-xl);
-  text-align: center;
-  background: var(--bg-secondary);
-  cursor: pointer;
+.xx-asset-upload-entry {
+  display: flex;
+  align-items: center;
+  gap: var(--space-md);
+  flex-wrap: wrap;
+  padding: var(--space-sm) var(--space-md);
+  border: 1px dashed transparent;
+  border-radius: var(--radius-md);
   transition: var(--transition-all);
 }
 
-.xx-asset-upload-zone:hover {
+.xx-asset-upload-entry-dragover {
   border-color: var(--primary-color);
   background: var(--primary-soft);
 }
 
-.xx-asset-upload-zone:active {
-  border-style: solid;
-  transform: scale(0.99);
-  box-shadow: inset 0 2px 4px rgba(0, 0, 0, 0.06);
-}
-
-.xx-asset-upload-icon {
-  font-size: 40px;
-  margin-bottom: var(--space-sm);
-  color: var(--primary-color);
-}
-
-.xx-asset-upload-text {
-  font-size: var(--font-size-base) !important;
-  color: var(--text-primary) !important;
-  margin: 0 0 var(--space-xs) !important;
+.xx-asset-upload-btn {
+  display: inline-flex;
+  align-items: center;
+  gap: var(--space-xs);
+  padding: 6px 16px;
+  font-size: var(--font-size-sm);
   font-weight: var(--font-weight-medium);
+  color: var(--text-inverse);
+  background: var(--primary-color);
+  border: none;
+  border-radius: var(--radius-sm);
+  cursor: pointer;
+  transition: var(--transition-all);
+  white-space: nowrap;
 }
 
-.xx-asset-upload-hint {
-  font-size: var(--font-size-sm) !important;
-  color: var(--text-tertiary) !important;
-  margin: 0 !important;
+.xx-asset-upload-btn:hover {
+  opacity: 0.9;
+}
+
+.xx-asset-upload-btn:active {
+  transform: scale(0.98);
+}
+
+.xx-asset-upload-status {
+  display: inline-flex;
+  align-items: center;
+  gap: var(--space-xs);
+  font-size: var(--font-size-xs);
+  color: var(--text-tertiary);
 }
 
 /* ============================================================
@@ -215,7 +223,7 @@
    ============================================================ */
 .xx-asset-grid {
   display: grid;
-  grid-template-columns: repeat(4, 1fr);
+  grid-template-columns: repeat(6, 1fr);
   gap: 14px;
 }
 
@@ -234,7 +242,7 @@
 .xx-asset-card:hover {
   border-color: var(--primary-color);
   box-shadow: var(--shadow-sm);
-  transform: translateY(-2px);
+  transform: translateY(-1px);
 }
 
 .xx-asset-card:active {
@@ -244,7 +252,7 @@
 
 /* 缩略图 */
 .xx-asset-thumb {
-  aspect-ratio: 9 / 16;
+  aspect-ratio: 3 / 4;
   position: relative;
   overflow: hidden;
   display: grid;
@@ -260,22 +268,22 @@
 }
 
 .xx-asset-thumb-placeholder {
-  font-size: var(--font-size-3xl);
+  font-size: var(--font-size-xl);
   opacity: 0.7;
 }
 
 /* 播放按钮 */
 .xx-asset-play {
   position: absolute;
-  width: 40px;
-  height: 40px;
+  width: 32px;
+  height: 32px;
   border-radius: var(--radius-full);
   background: rgba(255, 255, 255, 0.25);
   backdrop-filter: blur(4px);
   display: grid;
   place-items: center;
   color: var(--text-inverse);
-  font-size: var(--font-size-md);
+  font-size: var(--font-size-sm);
   transition: var(--transition-all);
 }
 
@@ -332,8 +340,8 @@
   position: absolute;
   bottom: var(--space-sm, 8px);
   right: var(--space-sm, 8px);
-  width: 28px;
-  height: 28px;
+  width: 24px;
+  height: 24px;
   border-radius: var(--radius-full, 999px);
   background: rgba(0, 0, 0, 0.45);
   backdrop-filter: blur(4px);
@@ -380,12 +388,12 @@
 
 /* 卡片信息 */
 .xx-asset-info {
-  padding: 12px;
+  padding: 8px;
 }
 
 .xx-asset-name {
-  margin: 0 0 6px;
-  font-size: var(--font-size-sm);
+  margin: 0 0 4px;
+  font-size: var(--font-size-xs);
   font-weight: var(--font-weight-semibold);
   color: var(--text-primary);
   white-space: nowrap;
@@ -397,9 +405,67 @@
   display: flex;
   justify-content: space-between;
   align-items: center;
+  gap: var(--space-xs);
   font-size: var(--font-size-xs);
   color: var(--text-secondary);
-  margin-bottom: var(--space-sm);
+  margin-bottom: 4px;
+  min-width: 0;
+}
+
+/* 状态标签行：标签过长省略 */
+.xx-asset-meta-status {
+  display: inline-flex;
+  align-items: center;
+  min-width: 0;
+}
+
+.xx-asset-meta-status .xx-status-pill {
+  max-width: 100%;
+  overflow: hidden;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+}
+
+.xx-asset-meta-duration {
+  flex-shrink: 0;
+  color: var(--text-tertiary);
+  font-variant-numeric: tabular-nums;
+}
+
+/* 余量标签独占一行 */
+.xx-asset-meta-usage {
+  justify-content: flex-start;
+  margin-bottom: var(--space-xs);
+}
+
+/* 视频素材余量角标（仅状态展示，不影响卡片操作） */
+.xx-asset-usage-badge {
+  display: inline-flex;
+  align-items: center;
+  padding: 1px 6px;
+  border-radius: var(--radius-full);
+  font-size: 10px;
+  font-weight: var(--font-weight-medium);
+  line-height: 1.5;
+  white-space: nowrap;
+}
+
+/* 已用尽：红色实心 */
+.xx-asset-usage-badge-exhausted {
+  background: var(--error-color);
+  color: var(--text-inverse);
+}
+
+/* 即将用尽：红色软底 */
+.xx-asset-usage-badge-warning {
+  background: var(--error-soft);
+  color: var(--error-color);
+}
+
+/* 已用 xx%：橙色软底 */
+.xx-asset-usage-badge-ratio {
+  background: var(--warning-soft);
+  color: var(--warning-color);
 }
 
 /* 诊断按钮 */
@@ -443,12 +509,16 @@
 .xx-status-pill {
   display: inline-flex;
   align-items: center;
-  gap: var(--space-xs);
-  padding: var(--space-xxs) var(--space-sm);
+  gap: 2px;
+  padding: 1px 6px;
   border-radius: var(--radius-full);
-  font-size: var(--font-size-xs);
+  font-size: 10px;
   font-weight: var(--font-weight-medium);
   line-height: 1.5;
+  white-space: nowrap;
+  max-width: 100%;
+  overflow: hidden;
+  text-overflow: ellipsis;
 }
 
 .xx-status-pill-ok {
@@ -592,7 +662,7 @@
    ============================================================ */
 .xx-assets-skeleton-grid {
   display: grid;
-  grid-template-columns: repeat(4, 1fr);
+  grid-template-columns: repeat(6, 1fr);
   gap: 14px;
 }
 
@@ -607,6 +677,10 @@
   .xx-asset-grid {
     grid-template-columns: repeat(3, 1fr);
   }
+
+  .xx-assets-skeleton-grid {
+    grid-template-columns: repeat(3, 1fr);
+  }
 }
 
 @media (max-width: 1200px) {
@@ -630,6 +704,10 @@
   .xx-asset-grid {
     grid-template-columns: repeat(3, 1fr);
   }
+
+  .xx-assets-skeleton-grid {
+    grid-template-columns: repeat(3, 1fr);
+  }
 }
 
 @media (max-width: 768px) {
@@ -641,6 +719,10 @@
     grid-template-columns: repeat(2, 1fr);
   }
 
+  .xx-assets-skeleton-grid {
+    grid-template-columns: repeat(2, 1fr);
+  }
+
   .xx-assets-filters {
     flex-direction: column;
     align-items: stretch;
@@ -660,43 +742,132 @@
   .xx-asset-grid {
     grid-template-columns: 1fr;
   }
+
+  .xx-assets-skeleton-grid {
+    grid-template-columns: 1fr;
+  }
 }
 
-/* ─── 上传进度弹窗 ─── */
-.xx-upload-progress-modal .ant-modal-content {
-  padding: 24px 16px 20px;
-  border-radius: 16px;
+/* ─── 上传队列面板 ─── */
+.xx-upload-queue {
+  margin-top: 12px;
+  border: 1px solid var(--border-primary, #e5e7eb);
+  border-radius: 12px;
+  background: #fff;
+  overflow: hidden;
 }
 
-.xx-upload-progress-body {
+.xx-upload-queue-header {
   display: flex;
-  flex-direction: column;
+  justify-content: space-between;
   align-items: center;
-  gap: 16px;
-  padding: 8px 0;
+  padding: 10px 14px;
+  border-bottom: 1px solid var(--border-primary, #eef2f7);
+  background: #f8fafc;
 }
 
-.xx-upload-progress-ring {
-  display: block;
-}
-
-.xx-upload-progress-text {
-  display: flex;
-  flex-direction: column;
-  align-items: center;
-  gap: 4px;
-}
-
-.xx-upload-progress-pct {
-  font-size: 22px;
-  font-weight: 700;
-  color: var(--primary-color, #6366f1);
-  line-height: 1;
-}
-
-.xx-upload-progress-label {
+.xx-upload-queue-title {
   font-size: 13px;
-  color: var(--text-secondary, #6b7280);
+  font-weight: 600;
+  color: var(--text-primary, #1e293b);
+}
+
+.xx-upload-queue-list {
+  max-height: 240px;
+  overflow-y: auto;
+}
+
+.xx-upload-queue-item {
+  display: flex;
+  align-items: flex-start;
+  gap: 10px;
+  padding: 10px 14px;
+}
+
+.xx-upload-queue-item + .xx-upload-queue-item {
+  border-top: 1px solid var(--border-primary, #f1f5f9);
+}
+
+.xx-upload-queue-icon {
+  padding-top: 2px;
+  font-size: 15px;
+}
+
+.xx-upload-queue-body {
+  flex: 1;
+  min-width: 0;
+}
+
+.xx-upload-queue-name {
+  font-size: 13px;
+  color: var(--text-primary, #1e293b);
+  white-space: nowrap;
+  overflow: hidden;
+  text-overflow: ellipsis;
+}
+
+.xx-upload-queue-progress {
+  margin-top: 6px;
+  height: 4px;
+  border-radius: 2px;
+  background: var(--border-primary, #e5e7eb);
+  overflow: hidden;
+}
+
+.xx-upload-queue-progress-bar {
+  height: 100%;
+  border-radius: 2px;
+  background: var(--primary-color, #6366f1);
+  transition: width 0.25s ease;
+}
+
+.xx-upload-queue-status {
+  margin-top: 4px;
+  font-size: 12px;
+  color: var(--text-tertiary, #94a3b8);
+}
+
+.xx-upload-queue-error .xx-upload-queue-status {
+  color: #ef4444;
+}
+
+.xx-upload-queue-actions {
+  display: flex;
+  gap: 6px;
+}
+
+.xx-upload-queue-btn {
+  border: none;
+  background: none;
+  cursor: pointer;
+  font-size: 13px;
+  color: var(--text-tertiary, #94a3b8);
+  padding: 2px 4px;
+}
+
+.xx-upload-queue-btn:hover {
+  color: var(--primary-color, #6366f1);
+}
+
+/* ─── 素材卡片上传中遮罩进度条 ─── */
+.xx-asset-thumb-uploading {
+  flex-direction: column;
+  gap: 8px;
+}
+
+.xx-asset-upload-bar {
+  width: 70%;
+  height: 4px;
+  border-radius: 2px;
+  background: rgba(255, 255, 255, 0.35);
+  overflow: hidden;
+}
+
+.xx-asset-upload-bar-inner {
+  height: 100%;
+  border-radius: 2px;
+  background: #fff;
+  transition: width 0.25s ease;
 }
 
 /* ─── 批量打标签弹窗 ─── */

apps/web/src/pages/assets/components/AssetCard.tsx

@@ -8,7 +8,7 @@ import {
   CloseCircleOutlined,
 } from "@ant-design/icons"
 import { Popconfirm } from "antd"
-import type { AssetItem } from "@/pages/assets/types"
+import { getUsageBadge, type AssetItem } from "@/pages/assets/types"
 import { thumbGradient } from "@/pages/assets/utils/asset"
 import { kindIcon } from "@/pages/assets/utils/kindIcon"
 import { StatusPill } from "./AssetSkeleton"
@@ -20,6 +20,8 @@ export interface AssetCardProps {
   asset: AssetItem
   selected: boolean
   diagnosing?: boolean
+  /** 上传中实时进度（仅 uploading 态有值；ingesting 后由后端状态接管） */
+  uploadProgress?: { progress: number; uploading: boolean }
   onToggle: () => void
   onDiagnose: () => void
   onPlay: () => void
@@ -30,99 +32,128 @@ const AssetCard: React.FC<AssetCardProps> = ({
   asset,
   selected,
   diagnosing,
+  uploadProgress,
   onToggle,
   onDiagnose,
   onPlay,
   onDelete,
-}) => (
-  <div className={`xx-asset-card${selected ? " xx-asset-card-selected" : ""}`} onClick={onToggle}>
-    {/* 缩略图区 */}
-    <div className="xx-asset-thumb" style={{ background: thumbGradient(asset.kind) }}>
-      {asset.thumbUrl ? (
-        <img src={asset.thumbUrl} alt={asset.name} />
-      ) : (
-        <span className="xx-asset-thumb-placeholder">
-          {asset.loading ? <LoadingOutlined /> : kindIcon(asset.kind)}
-        </span>
-      )}
+}) => {
+  const isUploading = !!uploadProgress?.uploading
+  // 视频素材余量角标（已用尽/即将用尽/已用 xx%）；非视频或字段缺失返回 null
+  const usageBadge = getUsageBadge(asset)
+  return (
+    <div className={`xx-asset-card${selected ? " xx-asset-card-selected" : ""}`} onClick={onToggle}>
+      {/* 缩略图区 */}
+      <div className="xx-asset-thumb" style={{ background: thumbGradient(asset.kind) }}>
+        {asset.thumbUrl ? (
+          <img src={asset.thumbUrl} alt={asset.name} />
+        ) : (
+          <span className="xx-asset-thumb-placeholder">
+            {asset.loading ? <LoadingOutlined /> : kindIcon(asset.kind)}
+          </span>
+        )}
 
-      {/* 处理中遮罩 */}
-      {asset.loading && (
-        <div className="xx-asset-thumb-overlay xx-asset-thumb-processing">
-          <LoadingOutlined />
-          <span>处理中</span>
+        {/* 上传中遮罩：真实进度百分比 + 进度条 */}
+        {isUploading && (
+          <div className="xx-asset-thumb-overlay xx-asset-thumb-uploading">
+            <LoadingOutlined />
+            <span>上传中 {uploadProgress?.progress ?? 0}%</span>
+            <div className="xx-asset-upload-bar">
+              <div
+                className="xx-asset-upload-bar-inner"
+                style={{ width: `${uploadProgress?.progress ?? 0}%` }}
+              />
+            </div>
+          </div>
+        )}
+
+        {/* 转码/处理中遮罩 */}
+        {asset.loading && !isUploading && (
+          <div className="xx-asset-thumb-overlay xx-asset-thumb-processing">
+            <LoadingOutlined />
+            <span>转码处理中</span>
+          </div>
+        )}
+
+        {/* 失败状态标识 */}
+        {asset.status === "bad" && asset.statusLabel === "处理失败" && (
+          <div className="xx-asset-thumb-overlay xx-asset-thumb-failed">
+            <CloseCircleOutlined />
+            <span>处理失败</span>
+          </div>
+        )}
+
+        {/* 视频/配音类显示播放按钮（处理中/失败不显示） */}
+        {asset.kind === "video" && !asset.loading && asset.status !== "bad" && (
+          <span
+            className="xx-asset-play"
+            onClick={(e) => {
+              e.stopPropagation()
+              onPlay()
+            }}
+          >
+            <PlayCircleOutlined />
+          </span>
+        )}
+
+        {/* 删除按钮 */}
+        <Popconfirm
+          title="确认删除"
+          description="删除后不可恢复，确定要删除这个素材吗？"
+          onConfirm={(e) => {
+            e?.stopPropagation()
+            onDelete()
+          }}
+          onCancel={(e) => e?.stopPropagation()}
+          okText="删除"
+          cancelText="取消"
+          okButtonProps={{ danger: true }}
+        >
+          <span className="xx-asset-delete" onClick={(e) => e.stopPropagation()}>
+            <DeleteOutlined />
+          </span>
+        </Popconfirm>
+
+        {/* 选中态勾选 */}
+        {selected && (
+          <span className="xx-asset-check">
+            <CheckOutlined />
+          </span>
+        )}
+      </div>
+
+      {/* 信息区 */}
+      <div className="xx-asset-info">
+        <p className="xx-asset-name" title={asset.name}>
+          {asset.name}
+        </p>
+        <div className="xx-asset-meta">
+          <span className="xx-asset-meta-status">
+            <StatusPill status={asset.status} label={asset.statusLabel} />
+          </span>
+          {asset.duration && <span className="xx-asset-meta-duration">{asset.duration}</span>}
         </div>
-      )}
-
-      {/* 失败状态标识 */}
-      {asset.status === "bad" && asset.statusLabel === "处理失败" && (
-        <div className="xx-asset-thumb-overlay xx-asset-thumb-failed">
-          <CloseCircleOutlined />
-          <span>处理失败</span>
-        </div>
-      )}
-
-      {/* 视频/配音类显示播放按钮（处理中/失败不显示） */}
-      {asset.kind === "video" && !asset.loading && asset.status !== "bad" && (
-        <span
-          className="xx-asset-play"
+        {usageBadge && (
+          <div className="xx-asset-meta xx-asset-meta-usage">
+            <span className={`xx-asset-usage-badge xx-asset-usage-badge-${usageBadge.variant}`}>
+              {usageBadge.label}
+            </span>
+          </div>
+        )}
+        <button
+          className={`xx-asset-diagnose-btn${diagnosing ? " xx-asset-diagnose-btn-loading" : ""}`}
+          disabled={diagnosing || asset.loading || asset.status === "bad"}
           onClick={(e) => {
             e.stopPropagation()
-            onPlay()
+            onDiagnose()
           }}
         >
-          <PlayCircleOutlined />
-        </span>
-      )}
-
-      {/* 删除按钮 */}
-      <Popconfirm
-        title="确认删除"
-        description="删除后不可恢复，确定要删除这个素材吗？"
-        onConfirm={(e) => {
-          e?.stopPropagation()
-          onDelete()
-        }}
-        onCancel={(e) => e?.stopPropagation()}
-        okText="删除"
-        cancelText="取消"
-        okButtonProps={{ danger: true }}
-      >
-        <span className="xx-asset-delete" onClick={(e) => e.stopPropagation()}>
-          <DeleteOutlined />
-        </span>
-      </Popconfirm>
-
-      {/* 选中态勾选 */}
-      {selected && (
-        <span className="xx-asset-check">
-          <CheckOutlined />
-        </span>
-      )}
-    </div>
-
-    {/* 信息区 */}
-    <div className="xx-asset-info">
-      <p className="xx-asset-name" title={asset.name}>
-        {asset.name}
-      </p>
-      <div className="xx-asset-meta">
-        <StatusPill status={asset.status} label={asset.statusLabel} />
-        {asset.duration && <span>{asset.duration}</span>}
+          {diagnosing ? <LoadingOutlined /> : <ExperimentOutlined />}
+          {diagnosing ? "诊断中..." : "诊断"}
+        </button>
       </div>
-      <button
-        className={`xx-asset-diagnose-btn${diagnosing ? " xx-asset-diagnose-btn-loading" : ""}`}
-        disabled={diagnosing || asset.loading || asset.status === "bad"}
-        onClick={(e) => {
-          e.stopPropagation()
-          onDiagnose()
-        }}
-      >
-        {diagnosing ? <LoadingOutlined /> : <ExperimentOutlined />}
-        {diagnosing ? "诊断中..." : "诊断"}
-      </button>
     </div>
-  </div>
-)
+  )
+}
 
 export default AssetCard

apps/web/src/pages/assets/components/AssetGridSection.tsx

@@ -8,6 +8,9 @@ import type { AssetItem } from "../types"
 import AssetCard from "./AssetCard"
 import { SkeletonCard } from "./AssetSkeleton"
 
+/** 上传中素材的实时进度（asset_id → 进度信息），由上传队列合并到卡片 */
+export type UploadProgressMap = Map<string, { progress: number; uploading: boolean }>
+
 export interface AssetGridSectionProps {
   loading: boolean
   error: boolean
@@ -15,6 +18,7 @@ export interface AssetGridSectionProps {
   assets: AssetItem[]
   selectedIds: Set<string>
   diagnosingId: string | null
+  uploadProgressMap?: UploadProgressMap
   onRetry?: () => void
   onToggleSelect: (id: string) => void
   onDiagnose: (asset: AssetItem) => void
@@ -29,6 +33,7 @@ export const AssetGridSection: React.FC<AssetGridSectionProps> = ({
   assets,
   selectedIds,
   diagnosingId,
+  uploadProgressMap,
   onRetry,
   onToggleSelect,
   onDiagnose,
@@ -70,6 +75,7 @@ export const AssetGridSection: React.FC<AssetGridSectionProps> = ({
             asset={asset}
             selected={selectedIds.has(asset.id)}
             diagnosing={diagnosingId === asset.id}
+            uploadProgress={uploadProgressMap?.get(asset.id)}
             onToggle={() => onToggleSelect(asset.id)}
             onDiagnose={() => onDiagnose(asset)}
             onPlay={() => onPlay(asset)}

apps/web/src/pages/assets/components/AssetModals.tsx

@@ -11,12 +11,9 @@ import BatchTagModal from "./BatchTagModal"
 import BatchClassifyModal from "./BatchClassifyModal"
 import BatchMarkModal from "./BatchMarkModal"
 import ResultDrawer from "./ResultDrawer"
-import UploadProgressModal from "./UploadProgressModal"
 
 export interface AssetModalsProps {
   /* 上传进度 */
-  uploading: boolean
-  uploadProgress: number
 
   /* 新建视频库 */
   createModalOpen: boolean
@@ -68,8 +65,6 @@ export interface AssetModalsProps {
 }
 
 export const AssetModals: React.FC<AssetModalsProps> = ({
-  uploading,
-  uploadProgress,
   createModalOpen,
   onCreateModalCancel,
   onCreateModalOk,
@@ -109,9 +104,6 @@ export const AssetModals: React.FC<AssetModalsProps> = ({
 }) => {
   return (
     <>
-      {/* 上传进度弹窗 */}
-      <UploadProgressModal open={uploading} progress={uploadProgress} />
-
       {/* 新建视频库弹窗 */}
       <CreateLibraryModal
         open={createModalOpen}

apps/web/src/pages/assets/components/AssetUploadZone.tsx

@@ -1,37 +1,91 @@
 /**
- * AssetLibrary 上传拖拽区域
+ * AssetLibrary 上传入口（紧凑按钮模式）
+ * - 点击按钮打开文件选择（多选），多文件入队由 useAssetUpload 队列控制（最多 3 路直传）
+ * - 拖拽文件到内容区任意位置同样触发上传（不再占用大面积虚线框）
  */
-import React from "react"
-import { Upload } from "antd"
-import { InboxOutlined } from "@ant-design/icons"
+import React, { useRef, useState } from "react"
+import { PlusOutlined, CloudUploadOutlined } from "@ant-design/icons"
 
 export interface AssetUploadZoneProps {
   uploading: boolean
-  uploadProgress: number
-  onUpload: (file: File) => void
+  activeCount: number
+  pendingCount: number
+  onUpload: (files: File[]) => void
 }
 
-export const AssetUploadZone: React.FC<AssetUploadZoneProps> = ({ uploading, onUpload }) => {
+export const AssetUploadZone: React.FC<AssetUploadZoneProps> = ({
+  uploading,
+  activeCount,
+  pendingCount,
+  onUpload,
+}) => {
+  const inputRef = useRef<HTMLInputElement>(null)
+  // dragenter/dragleave 在经过子元素时会成对触发，用计数器避免高亮闪烁；
+  // 计数器归零（拖拽真正离开容器）才取消高亮
+  const dragDepthRef = useRef(0)
+  const [dragOver, setDragOver] = useState(false)
+
+  const pickFiles = (list: FileList | null) => {
+    if (!list || list.length === 0) return
+    onUpload(Array.from(list))
+  }
+
   return (
-    <Upload.Dragger
-      beforeUpload={(file) => {
-        onUpload(file as File)
-        return false
+    <div
+      className={`xx-asset-upload-entry${dragOver ? " xx-asset-upload-entry-dragover" : ""}`}
+      onDragEnter={(e) => {
+        e.preventDefault()
+        dragDepthRef.current += 1
+        setDragOver(true)
+      }}
+      onDragOver={(e) => {
+        e.preventDefault()
+      }}
+      onDragLeave={(e) => {
+        e.preventDefault()
+        dragDepthRef.current = Math.max(0, dragDepthRef.current - 1)
+        if (dragDepthRef.current === 0) {
+          setDragOver(false)
+        }
+      }}
+      onDrop={(e) => {
+        e.preventDefault()
+        dragDepthRef.current = 0
+        setDragOver(false)
+        pickFiles(e.dataTransfer.files)
       }}
-      showUploadList={false}
-      multiple
-      accept="video/*,image/*"
     >
-      <div className="xx-asset-upload-zone">
-        <p className="xx-asset-upload-icon">
-          <InboxOutlined />
-        </p>
-        <p className="xx-asset-upload-text">
-          {uploading ? "上传中..." : "点击或拖拽文件到此区域上传"}
-        </p>
-        <p className="xx-asset-upload-hint">支持视频、图片，单文件不超过 2GB</p>
-      </div>
-    </Upload.Dragger>
+      <button
+        type="button"
+        className="xx-asset-upload-btn"
+        onClick={() => inputRef.current?.click()}
+      >
+        <PlusOutlined />
+        上传素材
+      </button>
+      <span className="xx-asset-upload-status">
+        {uploading ? (
+          <>
+            <CloudUploadOutlined />
+            上传中…（进行 {activeCount} 个{pendingCount > 0 ? `，排队 ${pendingCount} 个` : ""}）
+          </>
+        ) : (
+          "视频、图片均可，单文件不超过 2GB；也可直接拖拽文件到此区域"
+        )}
+      </span>
+      <input
+        ref={inputRef}
+        type="file"
+        multiple
+        accept="video/*,image/*"
+        style={{ display: "none" }}
+        onChange={(e) => {
+          pickFiles(e.target.files)
+          // 允许连续选择同一文件
+          e.target.value = ""
+        }}
+      />
+    </div>
   )
 }
 

apps/web/src/pages/assets/components/UploadProgressModal.tsx

@@ -1,56 +0,0 @@
-import React from "react"
-import { Modal as AntModal } from "antd"
-
-/* ============================================================
- * UploadProgressModal — 上传进度弹窗（圆形动画 + 百分比）
- * ============================================================ */
-export interface UploadProgressModalProps {
-  open: boolean
-  progress: number
-}
-
-const UploadProgressModal: React.FC<UploadProgressModalProps> = ({ open, progress }) => (
-  <AntModal
-    open={open}
-    footer={null}
-    closable={false}
-    centered
-    width={260}
-    maskClosable={false}
-    className="xx-upload-progress-modal"
-  >
-    <div className="xx-upload-progress-body">
-      <svg className="xx-upload-progress-ring" viewBox="0 0 120 120" width={120} height={120}>
-        {/* 背景圆环 */}
-        <circle
-          cx="60"
-          cy="60"
-          r="52"
-          fill="none"
-          stroke="var(--border-primary, #e5e7eb)"
-          strokeWidth="8"
-        />
-        {/* 进度圆弧 */}
-        <circle
-          cx="60"
-          cy="60"
-          r="52"
-          fill="none"
-          stroke="var(--primary-color, #6366f1)"
-          strokeWidth="8"
-          strokeLinecap="round"
-          strokeDasharray={`${2 * Math.PI * 52}`}
-          strokeDashoffset={`${2 * Math.PI * 52 * (1 - progress / 100)}`}
-          transform="rotate(-90 60 60)"
-          style={{ transition: "stroke-dashoffset 0.3s ease" }}
-        />
-      </svg>
-      <div className="xx-upload-progress-text">
-        <span className="xx-upload-progress-pct">{progress}%</span>
-        <span className="xx-upload-progress-label">上传中…</span>
-      </div>
-    </div>
-  </AntModal>
-)
-
-export default UploadProgressModal

apps/web/src/pages/assets/components/UploadQueuePanel.tsx

@@ -0,0 +1,117 @@
+/**
+ * 上传队列面板
+ * 展示批量上传中每个文件的独立状态/进度；失败可重试、可移除、可清空已完成。
+ * 上传中的素材卡片同时也会出现在素材网格（后端 prepare 预建 asset），
+ * 此面板用于展示真实传输进度与失败重试入口。
+ */
+import React from "react"
+import {
+  LoadingOutlined,
+  CheckCircleFilled,
+  CloseCircleFilled,
+  ReloadOutlined,
+  CloseOutlined,
+} from "@ant-design/icons"
+import type { UploadItem } from "../hooks/useAssetUpload"
+
+export interface UploadQueuePanelProps {
+  items: UploadItem[]
+  onRetry: (tempId: string) => void
+  onRemove: (tempId: string) => void
+  onClearFinished: () => void
+}
+
+const STATUS_TEXT: Record<UploadItem["status"], string> = {
+  preparing: "排队中…",
+  uploading: "上传中",
+  ingesting: "转码中…",
+  done: "已完成",
+  error: "上传失败",
+}
+
+const UploadQueuePanel: React.FC<UploadQueuePanelProps> = ({
+  items,
+  onRetry,
+  onRemove,
+  onClearFinished,
+}) => {
+  if (items.length === 0) return null
+  const finishedCount = items.filter((it) => it.status === "done").length
+
+  return (
+    <div className="xx-upload-queue">
+      <div className="xx-upload-queue-header">
+        <span className="xx-upload-queue-title">
+          上传任务（{items.length}
+          {finishedCount > 0 ? `，已完成 ${finishedCount}` : ""}）
+        </span>
+        {finishedCount > 0 && (
+          <button type="button" className="xx-link-btn" onClick={onClearFinished}>
+            清空已完成
+          </button>
+        )}
+      </div>
+      <div className="xx-upload-queue-list">
+        {items.map((it) => {
+          const isActive = it.status === "preparing" || it.status === "uploading"
+          const showProgress = it.status === "uploading" || it.status === "ingesting"
+          return (
+            <div key={it.tempId} className={`xx-upload-queue-item xx-upload-queue-${it.status}`}>
+              <span className="xx-upload-queue-icon">
+                {it.status === "done" || it.duplicated ? (
+                  <CheckCircleFilled style={{ color: "#22c55e" }} />
+                ) : it.status === "error" ? (
+                  <CloseCircleFilled style={{ color: "#ef4444" }} />
+                ) : (
+                  <LoadingOutlined style={{ color: "var(--primary-color)" }} />
+                )}
+              </span>
+              <div className="xx-upload-queue-body">
+                <div className="xx-upload-queue-name" title={it.fileName}>
+                  {it.fileName}
+                </div>
+                {showProgress ? (
+                  <div className="xx-upload-queue-progress">
+                    <div
+                      className="xx-upload-queue-progress-bar"
+                      style={{ width: `${it.status === "ingesting" ? 100 : it.progress}%` }}
+                    />
+                  </div>
+                ) : null}
+                <div className="xx-upload-queue-status">
+                  {it.duplicated ? "素材已存在，已跳过" : STATUS_TEXT[it.status]}
+                  {it.status === "uploading" ? ` ${it.progress}%` : ""}
+                  {it.status === "error" && it.error ? `：${it.error}` : ""}
+                </div>
+              </div>
+              <span className="xx-upload-queue-actions">
+                {it.status === "error" && (
+                  <button
+                    type="button"
+                    className="xx-upload-queue-btn"
+                    title="重试"
+                    onClick={() => onRetry(it.tempId)}
+                  >
+                    <ReloadOutlined />
+                  </button>
+                )}
+                {(it.status === "error" || it.status === "done") && !isActive && (
+                  <button
+                    type="button"
+                    className="xx-upload-queue-btn"
+                    title="移除"
+                    onClick={() => onRemove(it.tempId)}
+                  >
+                    <CloseOutlined />
+                  </button>
+                )}
+              </span>
+            </div>
+          )
+        })}
+      </div>
+    </div>
+  )
+}
+
+export default UploadQueuePanel

apps/web/src/pages/assets/hooks/useAssetUpload.ts

@@ -1,65 +1,196 @@
-import { useState, useCallback } from "react"
+import { useState, useCallback, useRef, useEffect } from "react"
 import { useQueryClient } from "@tanstack/react-query"
 import { message } from "antd"
-import { uploadAssetDirect } from "@/api/assets"
-import { MAX_FILE_SIZE, LARGE_FILE_THRESHOLD } from "../constants"
+import { prepareDirectUploadHandle, type DirectUploadHandle } from "@/api/assets"
+import { MAX_FILE_SIZE } from "../constants"
 
-/**
- * 素材上传 Hook
- * 封装上传状态、进度管理和上传逻辑
- */
-interface UseAssetUploadProps {
-  effectiveLibId: string
+/** 单文件上传状态机 */
+export type UploadItemStatus = "preparing" | "uploading" | "ingesting" | "done" | "error"
+
+export interface UploadItem {
+  /** 前端临时 id（prepare 前无 asset_id 时用） */
+  tempId: string
+  file: File
+  fileName: string
+  /** 进度 0~100（仅直传阶段有真实进度） */
+  progress: number
+  status: UploadItemStatus
+  /** 后端 prepare 预建的 asset id（旧后端可能为空） */
+  assetId?: string
+  /** 去重命中：complete 返回 duplicated，标记完成但不产生新素材 */
+  duplicated?: boolean
+  error?: string
 }
 
-export function useAssetUpload({ effectiveLibId }: UseAssetUploadProps) {
+/** 批量直传最大并发数，避免多文件瓜分上行带宽 */
+const MAX_CONCURRENT = 3
+
+/**
+ * 素材批量上传 Hook
+ * - prepare 阶段后端预建 status=uploading 的 asset，前端拿到 asset_id 立即刷新列表
+ * - OSS 直传并发限制为 3，其余排队；每个文件独立进度/状态
+ * - complete 后素材进入转码（ingesting/processing），由列表轮询反映
+ * - 失败卡片支持重试/移除
+ */
+export function useAssetUpload({ effectiveLibId }: { effectiveLibId: string }) {
   const queryClient = useQueryClient()
 
-  const [uploading, setUploading] = useState(false)
-  const [uploadProgress, setUploadProgress] = useState(0)
+  const [items, setItems] = useState<UploadItem[]>([])
+  const itemsRef = useRef<UploadItem[]>([])
+  itemsRef.current = items
 
-  const handleUpload = useCallback(
-    async (file: File) => {
-      if (file.size > MAX_FILE_SIZE) {
-        message.error(`文件 "${file.name}" 超过 2GB 限制`)
-        return
+  const updateItem = useCallback((tempId: string, patch: Partial<UploadItem>) => {
+    setItems((prev) => prev.map((it) => (it.tempId === tempId ? { ...it, ...patch } : it)))
+  }, [])
+
+  /** 刷新素材列表（prepare 后/complete 后调用，让卡片即时出现/流转） */
+  const refreshList = useCallback(() => {
+    // 使用 refetchQueries 强制立即重新获取，避免 staleTime 导致延迟
+    if (effectiveLibId) {
+      queryClient.refetchQueries({ queryKey: ["assets", effectiveLibId] })
+    }
+    queryClient.invalidateQueries({ queryKey: ["asset-libraries"] })
+  }, [queryClient, effectiveLibId])
+
+  /** 执行单个文件的完整上传流程（prepare→transfer→complete） */
+  const runUpload = useCallback(
+    async (item: UploadItem, handle?: DirectUploadHandle) => {
+      try {
+        // 1. prepare（重试时复用已准备的 handle 也行，但签名可能过期，重新 prepare 最稳）
+        const h =
+          handle ??
+          (await prepareDirectUploadHandle({ file: item.file, library_id: effectiveLibId }))
+        if (h.prepared.asset_id) {
+          updateItem(item.tempId, {
+            status: "uploading",
+            assetId: h.prepared.asset_id,
+            progress: 0,
+          })
+          // 预建 asset 已入库，立即刷新让「上传中」卡片出现在网格
+          refreshList()
+        } else {
+          updateItem(item.tempId, { status: "uploading", progress: 0 })
+        }
+
+        // 2. OSS 直传（真实进度）
+        await h.transfer((pct) => updateItem(item.tempId, { progress: pct }))
+
+        // 3. complete：后端创建 ingest job，素材进入转码
+        updateItem(item.tempId, { status: "ingesting", progress: 100 })
+        const result = await h.complete()
+        refreshList()
+
+        if (result.duplicated) {
+          updateItem(item.tempId, { status: "done", duplicated: true, assetId: result.asset_id })
+          message.info(`"${item.fileName}" 与素材库已有内容相同，已跳过`)
+        } else {
+          updateItem(item.tempId, { status: "done" })
+          message.success(`"${item.fileName}" 上传完成，正在转码处理`)
+        }
+      } catch (err: unknown) {
+        const detail = err instanceof Error ? err.message : "上传失败"
+        console.error("[useAssetUpload] 上传失败:", item.fileName, err)
+        updateItem(item.tempId, { status: "error", error: detail })
+        message.error(`"${item.fileName}" 上传失败：${detail}`)
       }
+    },
+    [effectiveLibId, refreshList, updateItem],
+  )
+
+  /**
+   * 队列调度：把并发槽塞满（同时在途的 prepare+transfer 不超过 MAX_CONCURRENT）。
+   * runUpload 在 await prepare 期间 state 仍是 preparing，多个并发 pump 若只看 state
+   * 会重复认领同一项，因此用 claimedRef 记录已被认领的 tempId。
+   */
+  const inFlightRef = useRef(0)
+  const claimedRef = useRef<Set<string>>(new Set())
+  const pumpRef = useRef<() => void>(() => {})
+  pumpRef.current = () => {
+    while (inFlightRef.current < MAX_CONCURRENT) {
+      const next = itemsRef.current.find(
+        (it) => it.status === "preparing" && !claimedRef.current.has(it.tempId),
+      )
+      if (!next) return
+      claimedRef.current.add(next.tempId)
+      inFlightRef.current += 1
+      void runUpload(next).finally(() => {
+        inFlightRef.current -= 1
+        claimedRef.current.delete(next.tempId)
+        // 一个任务结束（成功/失败）后继续拉起排队任务
+        setTimeout(() => pumpRef.current(), 0)
+      })
+    }
+  }
+
+  useEffect(() => {
+    pumpRef.current()
+  }, [items])
+
+  /** 入队一个或多个文件 */
+  const enqueueUploads = useCallback(
+    (files: File[]) => {
       if (!effectiveLibId) {
         message.warning("请先选择或创建一个视频库")
         return
       }
-
-      setUploading(true)
-      setUploadProgress(0)
-      try {
-        if (file.size > LARGE_FILE_THRESHOLD) {
-          message.info(`大文件 "${file.name}" 将使用直传上传`)
+      const valid: File[] = []
+      for (const file of files) {
+        if (file.size > MAX_FILE_SIZE) {
+          message.error(`文件 "${file.name}" 超过 2GB 限制`)
+          continue
         }
-        await uploadAssetDirect({
-          file,
-          library_id: effectiveLibId,
-          onProgress: (pct) => setUploadProgress(pct),
-        })
-        message.success(`"${file.name}" 上传成功`)
-        queryClient.invalidateQueries({ queryKey: ["assets"] })
-        queryClient.invalidateQueries({ queryKey: ["asset-libraries"] })
-      } catch (err: unknown) {
-        const detail = err instanceof Error ? err.message : ""
-        console.error("[handleUpload] 上传失败:", err)
-        message.error(`"${file.name}" 上传失败${detail ? `：${detail}` : ""}`)
-        // 错误时延迟关闭弹窗，让用户能看到错误提示
-        await new Promise((r) => setTimeout(r, 1500))
-      } finally {
-        setUploading(false)
-        setUploadProgress(0)
+        valid.push(file)
       }
+      if (valid.length === 0) return
+
+      const newItems: UploadItem[] = valid.map((file, idx) => ({
+        tempId: `${Date.now()}-${idx}-${Math.random().toString(36).slice(2, 8)}`,
+        file,
+        fileName: file.name,
+        progress: 0,
+        status: "preparing",
+      }))
+      setItems((prev) => [...prev, ...newItems])
     },
-    [effectiveLibId, queryClient],
+    [effectiveLibId],
   )
 
+  /** 重试失败任务 */
+  const retryUpload = useCallback(
+    (tempId: string) => {
+      const target = itemsRef.current.find((it) => it.tempId === tempId)
+      if (!target) return
+      updateItem(tempId, { status: "preparing", progress: 0, error: undefined })
+      // 状态更新后由 useEffect 触发 pump
+    },
+    [updateItem],
+  )
+
+  /** 从上传列表移除（已进入转码的由素材网格管理；这里只移除上传面板记录） */
+  const removeUpload = useCallback((tempId: string) => {
+    setItems((prev) => prev.filter((it) => it.tempId !== tempId))
+  }, [])
+
+  /** 清空已完成/去重记录 */
+  const clearFinished = useCallback(() => {
+    setItems((prev) => prev.filter((it) => it.status !== "done"))
+  }, [])
+
+  const activeCount = items.filter(
+    (it) => it.status === "preparing" || it.status === "uploading",
+  ).length
+  const pendingCount = items.filter((it) => it.status === "preparing").length
+  const hasActive = activeCount > 0 || items.some((it) => it.status === "ingesting")
+
   return {
-    uploading,
-    uploadProgress,
-    handleUpload,
+    uploadItems: items,
+    enqueueUploads,
+    retryUpload,
+    removeUpload,
+    clearFinished,
+    /** 是否有进行中的上传（用于上传区文案） */
+    uploading: hasActive,
+    activeCount,
+    pendingCount,
   }
 }

apps/web/src/pages/assets/hooks/useAssetsData.ts

@@ -45,11 +45,21 @@ export function useAssetsData() {
     queryKey: ["assets", effectiveLibId],
     queryFn: () =>
       getAssets(effectiveLibId, {
-        // 拉取所有非删除状态的素材，让用户上传后立刻能看到"处理中"的素材
+        // 拉取所有非删除状态的素材，让用户上传后立刻能看到"上传中/处理中"的素材
         status: "ready,uploading,ingesting,processing,pending,error,failed",
       }),
     enabled: !!effectiveLibId,
     staleTime: 30_000,
+    // 列表中存在上传中/转码中素材时每 3s 轮询；全部就绪后自动停止
+    refetchInterval: (query) => {
+      const data = query.state.data as { items: ApiAssetItem[] } | undefined
+      const items = data?.items ?? []
+      const processing = items.some((a) => {
+        const st = a.status ?? ""
+        return st === "uploading" || st === "ingesting" || st === "processing" || st === "pending"
+      })
+      return processing ? 3000 : false
+    },
   })
 
   const assets: AssetItem[] = useMemo(

apps/web/src/pages/assets/types.ts

@@ -27,6 +27,36 @@ export interface AssetItem {
   duration?: string
   size: number
   createdAt: string
+  /** 已切片段占用时长占比（0~1），后端字段缺失时为 undefined */
+  usedRatio?: number
+  /** 是否已彻底用尽（false 的素材不参与生成选片），字段缺失时视为可用 */
+  usable?: boolean
+}
+
+/** 素材余量角标状态（仅视频素材） */
+export interface UsageBadge {
+  /** 角标文案 */
+  label: string
+  /** 样式变体：exhausted=红色实心，warning=红色软底，ratio=橙色软底 */
+  variant: "exhausted" | "warning" | "ratio"
+}
+
+/**
+ * 根据后端余量字段计算视频素材的余量角标；
+ * 非视频、字段缺失或已用占比 <50% 时不显示（返回 null）。
+ */
+export const getUsageBadge = (asset: {
+  kind?: AssetKind
+  usable?: boolean
+  usedRatio?: number
+}): UsageBadge | null => {
+  if (asset.kind && asset.kind !== "video") return null
+  if (asset.usable === false) return { label: "已用尽", variant: "exhausted" }
+  const ratio = asset.usedRatio
+  if (ratio == null) return null
+  if (ratio >= 0.85) return { label: "即将用尽", variant: "warning" }
+  if (ratio >= 0.5) return { label: `已用 ${Math.round(ratio * 100)}%`, variant: "ratio" }
+  return null
 }
 
 /** 根据 mime_type 推断前端 AssetKind */
@@ -111,5 +141,7 @@ export const mapAsset = (item: ApiAssetItem): AssetItem => {
     duration: metadata.duration != null ? formatDuration(metadata.duration as number) : undefined,
     size: item.file_size ? +(item.file_size / (1024 * 1024)).toFixed(1) : 0,
     createdAt: item.created_at ? new Date(item.created_at).toISOString().slice(0, 10) : "—",
+    usedRatio: item.used_ratio ?? undefined,
+    usable: item.usable ?? undefined,
   }
 }

apps/web/src/pages/generate/GeneratePage.tsx

@@ -8,7 +8,7 @@
  * - 步骤 6 封面从最终成片中智能选帧（MediaKit）
  * - 点"确认生成"时调用 createGenerationTask 创建一次服务器渲染任务
  */
-import React, { useMemo, useState, useEffect, useRef } from "react"
+import React, { useMemo, useState, useEffect, useRef, useCallback } from "react"
 import { message } from "antd"
 import { useNavigate } from "react-router-dom"
 import type { VoiceClone } from "@/api/voice-clone"
@@ -217,6 +217,22 @@ const GeneratePage: React.FC = () => {
     },
   })
 
+  /* ── 步骤4「确认生成视频」：校验标题/预览 → 创建最终渲染任务 → 成功后进入步骤5 ── */
+  const handleConfirmGenerate = useCallback(async () => {
+    if (!titleSettings.title.trim()) {
+      message.warning("请选择或输入标题")
+      return
+    }
+    if (!previewReady) {
+      message.warning("预览视频正在加载，请稍候")
+      return
+    }
+    const ok = await handleGenerate()
+    if (ok) {
+      setCurrentStep(5)
+    }
+  }, [titleSettings.title, previewReady, handleGenerate, setCurrentStep])
+
   /* ── 步骤导航 ── */
   const { goNext, goPrev } = useStepNavigation({
     currentStep,
@@ -300,7 +316,7 @@ const GeneratePage: React.FC = () => {
             currentStep={currentStep}
             onPrev={goPrev}
             onNext={goNext}
-            onGenerate={handleGenerate}
+            onConfirmGenerate={handleConfirmGenerate}
             generating={generating}
             generated={generated}
             generateError={generateError}

apps/web/src/pages/generate/components/GenerateStepActions.tsx

@@ -1,14 +1,19 @@
 /**
  * GeneratePage 步骤底部操作按钮
+ *
+ * 步骤 1~3：上一步 / 下一步
+ * 步骤 4（标题+预览）：上一步 / 确认生成视频（点击后直接创建最终渲染任务，成功后跳转步骤5）
+ * 步骤 5（确认生成）：上一步 / 下一步（渲染中禁用，渲染完成后可进入封面）
+ * 步骤 6（选择封面）：仅上一步
  */
 import React from "react"
-import { ThunderboltOutlined } from "@ant-design/icons"
 
 export interface GenerateStepActionsProps {
   currentStep: number
   onPrev: () => void
   onNext: () => void
-  onGenerate: () => void
+  /** 步骤4：确认生成视频（校验 + 创建渲染任务 + 成功后进入步骤5） */
+  onConfirmGenerate: () => void | Promise<void>
   generating: boolean
   generated: boolean
   generateError: string | null
@@ -18,36 +23,74 @@ export const GenerateStepActions: React.FC<GenerateStepActionsProps> = ({
   currentStep,
   onPrev,
   onNext,
-  onGenerate,
+  onConfirmGenerate,
   generating,
   generated,
   generateError,
 }) => {
+  const renderPrimaryButton = () => {
+    /* 步骤 1~3：上一步 / 下一步（必填校验由 useStepNavigation.goNext 统一处理） */
+    if (currentStep < 4) {
+      return (
+        <button className="xx-btn xx-btn-primary" onClick={onNext}>
+          下一步 →
+        </button>
+      )
+    }
+
+    /* 步骤 4：确认生成视频（触发按钮在标题页） */
+    if (currentStep === 4) {
+      if (generating) {
+        return (
+          <button className="xx-btn xx-btn-primary" disabled>
+            ⏳ 视频生成中…
+          </button>
+        )
+      }
+      if (generateError) {
+        return (
+          <button className="xx-btn xx-btn-primary" onClick={onConfirmGenerate}>
+            🔄 重新生成视频
+          </button>
+        )
+      }
+      if (generated) {
+        return (
+          <button className="xx-btn xx-btn-primary" onClick={onNext}>
+            下一步 →
+          </button>
+        )
+      }
+      return (
+        <button className="xx-btn xx-btn-primary" onClick={onConfirmGenerate}>
+          ✨ 确认生成视频
+        </button>
+      )
+    }
+
+    /* 步骤 5：渲染中禁用，完成后下一步进入封面 */
+    if (currentStep === 5) {
+      return (
+        <button
+          className="xx-btn xx-btn-primary"
+          onClick={onNext}
+          disabled={generating || !generated}
+        >
+          {generating ? "视频生成中…" : "下一步 →"}
+        </button>
+      )
+    }
+
+    /* 步骤 6（最后一步）：无主按钮 */
+    return null
+  }
+
   return (
     <div className="xx-step-actions">
       <button className="xx-btn xx-btn-ghost" onClick={onPrev} disabled={currentStep === 1}>
         ← 上一步
       </button>
-      {currentStep < 6 ? (
-        <button className="xx-btn xx-btn-primary" onClick={onNext}>
-          下一步 →
-        </button>
-      ) : (
-        <button
-          className="xx-btn xx-btn-primary"
-          onClick={onGenerate}
-          disabled={generating || (generated && !generateError)}
-        >
-          <ThunderboltOutlined />
-          {generating
-            ? "生成中…"
-            : generated && !generateError
-              ? "已生成"
-              : generateError
-                ? "🔄 重新生成"
-                : "✨ 确认生成"}
-        </button>
-      )}
+      {renderPrimaryButton()}
     </div>
   )
 }

apps/web/src/pages/generate/components/Step2MaterialSelect.tsx

@@ -62,8 +62,9 @@ const Step2MaterialSelect: React.FC<Step2MaterialSelectProps> = (props) => {
           </div>
 
           <ManualMaterialList
-            materials={m.materials}
+            materials={m.selectableMaterials}
             materialsLoading={m.materialsLoading}
+            allExhausted={m.allMaterialsExhausted}
             selectedMaterials={m.selectedMaterials}
             onToggle={m.handleToggleMaterial}
           />
@@ -77,21 +78,11 @@ const Step2MaterialSelect: React.FC<Step2MaterialSelectProps> = (props) => {
             onMatch={m.handleSmartMatch}
             hasMatched={m.hasMatched}
             onRefresh={m.handleRefreshMatch}
-            materialsCount={m.materials.items.length}
+            materialsCount={m.selectableMaterials.items.length}
             loading={m.materialsLoading}
           />
 
-          <SmartMatchResults
-            matchedAssets={m.smartMatchedResults}
-            selectedIds={m.smartSelectedIds}
-            matching={m.smartMatching}
-            hasMatched={m.hasMatched}
-            onToggle={m.handleToggleSmartSelect}
-            onSelectAll={m.handleSelectAllMatched}
-            onClear={m.handleClearSmartSelect}
-            formatDuration={m.formatDuration}
-            selectedTotalDuration={m.smartSelectedTotalDuration}
-          />
+          <SmartMatchResults matching={m.smartMatching} hasMatched={m.hasMatched} />
         </div>
       )}
     </div>

apps/web/src/pages/generate/components/Step5VoiceSelect.tsx

@@ -27,6 +27,13 @@ const getFileSize = (item: AssetItem): number => {
   return item.file_size ?? (item.metadata?.file_size as number) ?? 0
 }
 
+/** 是否为 AI 音色（克隆/预置音色模型：无固定时长、无实体音频文件，按脚本实时合成） */
+const isAiVoice = (item: AssetItem): boolean => {
+  const duration = getDuration(item)
+  const size = getFileSize(item)
+  return (!duration || duration <= 0) && (!size || size <= 0)
+}
+
 const formatDuration = (seconds?: number): string => {
   if (!seconds || seconds <= 0) return "00:00"
   const m = Math.floor(seconds / 60)
@@ -96,10 +103,10 @@ const Step5VoiceSelect: React.FC<Step5VoiceSelectProps> = ({
   /** 选中素材（含时长校验） */
   const handleSelect = useCallback(
     (id: string) => {
-      // 如果启用了时长校验，且配音时长不足
+      // 如果启用了时长校验，且配音时长不足（AI 音色按脚本实时合成，不参与时长校验）
       if (totalVideoDuration > 0) {
         const material = materials.find((m) => m.id === id)
-        if (material && getDuration(material) < totalVideoDuration) {
+        if (material && !isAiVoice(material) && getDuration(material) < totalVideoDuration) {
           setPendingVoiceId(id)
           setDurationWarningOpen(true)
           return
@@ -280,25 +287,29 @@ const Step5VoiceSelect: React.FC<Step5VoiceSelectProps> = ({
                   alignItems: "center",
                 }}
               >
-                <span style={{ display: "flex", alignItems: "center", gap: 4 }}>
-                  {formatDuration(getDuration(item))}
-                  {totalVideoDuration > 0 && getDuration(item) < Number(totalVideoDuration) && (
-                    <span
-                      style={{
-                        color: "#ff4d4f",
-                        fontSize: 11,
-                        fontWeight: 500,
-                        display: "inline-flex",
-                        alignItems: "center",
-                        gap: 2,
-                      }}
-                    >
-                      <WarningOutlined />
-                      时长不足
-                    </span>
-                  )}
-                </span>
-                <span>{formatFileSize(getFileSize(item))}</span>
+                {isAiVoice(item) ? (
+                  <span style={{ color: "#1677ff", fontWeight: 500 }}>AI 音色</span>
+                ) : (
+                  <span style={{ display: "flex", alignItems: "center", gap: 4 }}>
+                    {formatDuration(getDuration(item))}
+                    {totalVideoDuration > 0 && getDuration(item) < Number(totalVideoDuration) && (
+                      <span
+                        style={{
+                          color: "#ff4d4f",
+                          fontSize: 11,
+                          fontWeight: 500,
+                          display: "inline-flex",
+                          alignItems: "center",
+                          gap: 2,
+                        }}
+                      >
+                        <WarningOutlined />
+                        时长不足
+                      </span>
+                    )}
+                  </span>
+                )}
+                <span>{isAiVoice(item) ? "按文本合成" : formatFileSize(getFileSize(item))}</span>
               </div>
             </div>
           )

apps/web/src/pages/generate/components/material/ManualMaterialList.tsx

@@ -11,6 +11,8 @@ const { Text } = Typography
 interface ManualMaterialListProps {
   materials: { items: AssetItem[]; total: number }
   materialsLoading: boolean
+  /** 库内有素材但全部已用尽（usable === false），用于区分空状态文案 */
+  allExhausted?: boolean
   selectedMaterials: string[]
   onToggle: (materialId: string) => void
 }
@@ -247,6 +249,7 @@ const MaterialCard: React.FC<{
 const ManualMaterialList: React.FC<ManualMaterialListProps> = ({
   materials,
   materialsLoading,
+  allExhausted,
   selectedMaterials,
   onToggle,
 }) => {
@@ -256,7 +259,9 @@ const ManualMaterialList: React.FC<ManualMaterialListProps> = ({
         <Text style={{ color: "var(--text-secondary)", padding: "16px 0" }}>加载素材中…</Text>
       ) : materials.items.length === 0 ? (
         <Text style={{ color: "var(--text-secondary)", padding: "16px 0" }}>
-          暂无素材，请先在视频库中上传
+          {allExhausted
+            ? "暂无可选素材（素材可能已用尽，请先上传新素材）"
+            : "暂无素材，请先在视频库中上传"}
         </Text>
       ) : (
         <div

apps/web/src/pages/generate/components/material/SmartMatchCard.tsx

@@ -1,52 +0,0 @@
-/**
- * 智能匹配卡片（Q5 简化版）
- * 只展示素材缩略图、名称、时长，无匹配分数
- */
-import React from "react"
-import { PlayCircleOutlined, CheckCircleFilled } from "@ant-design/icons"
-import type { AssetItem } from "@/api/assets"
-
-interface SmartMatchCardProps {
-  asset: AssetItem
-  selected: boolean
-  onClick: () => void
-  formatDuration: (seconds: number) => string
-}
-
-const SmartMatchCard: React.FC<SmartMatchCardProps> = ({
-  asset,
-  selected,
-  onClick,
-  formatDuration,
-}) => {
-  return (
-    <div className={`xx-smart-match-card ${selected ? "selected" : ""}`} onClick={onClick}>
-      {/* 缩略图 */}
-      <div className="xx-smart-match-thumb">
-        {asset.thumbnail_url ? (
-          <img src={asset.thumbnail_url} alt={asset.name} />
-        ) : (
-          <div className="xx-smart-match-thumb-placeholder">
-            <PlayCircleOutlined style={{ fontSize: 32, opacity: 0.5 }} />
-          </div>
-        )}
-        {selected && (
-          <div className="xx-smart-match-check">
-            <CheckCircleFilled style={{ fontSize: 20, color: "#fff" }} />
-          </div>
-        )}
-        {asset.duration && (
-          <div className="xx-smart-match-duration">{formatDuration(asset.duration)}</div>
-        )}
-      </div>
-      {/* 名称 */}
-      <div className="xx-smart-match-info">
-        <div className="xx-smart-match-name" title={asset.name}>
-          {asset.name}
-        </div>
-      </div>
-    </div>
-  )
-}
-
-export default SmartMatchCard

apps/web/src/pages/generate/components/material/SmartMatchResults.tsx

@@ -1,35 +1,17 @@
 /**
- * 智能匹配结果区（Q5 简化版）
- * 直接展示 AI 选中的素材，无匹配分数和理由
+ * 智能匹配状态区
+ * 仅展示匹配中 / 未匹配 / 匹配成功三种状态，不展示 AI 选中的素材明细
+ * （选中的素材仍由 smartSelectedIds 驱动提交，逻辑不变）
  */
 import React from "react"
 import { LoadingOutlined } from "@ant-design/icons"
-import type { AssetItem } from "@/api/assets"
-import SmartMatchCard from "./SmartMatchCard"
 
 interface SmartMatchResultsProps {
-  matchedAssets: AssetItem[]
-  selectedIds: string[]
   matching: boolean
   hasMatched: boolean
-  onToggle: (assetId: string) => void
-  onSelectAll: () => void
-  onClear: () => void
-  formatDuration: (seconds: number) => string
-  selectedTotalDuration: number
 }
 
-const SmartMatchResults: React.FC<SmartMatchResultsProps> = ({
-  matchedAssets,
-  selectedIds,
-  matching,
-  hasMatched,
-  onToggle,
-  onSelectAll,
-  onClear,
-  formatDuration,
-  selectedTotalDuration,
-}) => {
+const SmartMatchResults: React.FC<SmartMatchResultsProps> = ({ matching, hasMatched }) => {
   // 匹配中状态
   if (matching) {
     return (
@@ -45,66 +27,26 @@ const SmartMatchResults: React.FC<SmartMatchResultsProps> = ({
     )
   }
 
-  // 未匹配状态提示
-  if (!hasMatched) {
+  // 匹配成功：轻量提示，不展示素材明细卡片
+  if (hasMatched) {
     return (
-      <div className="xx-smart-match-empty">
-        <div style={{ fontSize: 36, marginBottom: 8 }}>💡</div>
-        <div style={{ color: "var(--text-secondary)", fontSize: 13 }}>
-          点击「让 AI 帮你选」，自动从视频库中选择最合适的素材
-        </div>
+      <div className="xx-smart-match-success">
+        <span style={{ fontSize: 16 }}>✅</span>
+        <span style={{ color: "var(--text-secondary)", fontSize: 13 }}>
+          AI 已帮你选好素材，可直接进入下一步
+        </span>
       </div>
     )
   }
 
-  // 无结果
-  if (matchedAssets.length === 0) return null
-
+  // 未匹配状态提示
   return (
-    <>
-      <div className="xx-smart-match-results">
-        <div className="xx-smart-match-results-header">
-          <span className="xx-smart-match-results-title">
-            AI 已选素材 ({matchedAssets.length}个)
-          </span>
-          <div className="xx-smart-match-results-actions">
-            <button type="button" className="xx-link-btn" onClick={onSelectAll}>
-              全选
-            </button>
-            <span style={{ color: "var(--border-color)" }}>|</span>
-            <button type="button" className="xx-link-btn" onClick={onClear}>
-              清空
-            </button>
-          </div>
-        </div>
-        <div className="xx-smart-match-grid">
-          {matchedAssets.map((asset) => {
-            const isSelected = selectedIds.includes(asset.id)
-            return (
-              <SmartMatchCard
-                key={asset.id}
-                asset={asset}
-                selected={isSelected}
-                onClick={() => onToggle(asset.id)}
-                formatDuration={formatDuration}
-              />
-            )
-          })}
-        </div>
+    <div className="xx-smart-match-empty">
+      <div style={{ fontSize: 36, marginBottom: 8 }}>💡</div>
+      <div style={{ color: "var(--text-secondary)", fontSize: 13 }}>
+        点击「让 AI 帮你选」，自动从视频库中选择最合适的素材
       </div>
-
-      {/* 已选素材汇总 */}
-      {selectedIds.length > 0 && (
-        <div className="xx-smart-match-summary">
-          <div className="xx-smart-match-summary-header">
-            <span className="xx-pill xx-pill-ok">已选 {selectedIds.length} 个素材</span>
-            <span style={{ color: "var(--text-tertiary)", fontSize: 12 }}>
-              预计总时长约 {selectedTotalDuration.toFixed(0)} 秒
-            </span>
-          </div>
-        </div>
-      )}
-    </>
+    </div>
   )
 }
 

apps/web/src/pages/generate/components/step7-confirm/GenerationStatus.tsx

@@ -25,10 +25,21 @@ const GenerationStatus: React.FC<GenerationStatusProps> = ({
   onRetry,
   onDismissError,
 }) => {
-  if (!generating && !generated && !generateError) return null
-
   return (
     <div style={{ marginTop: 16 }}>
+      {!generating && !generated && !generateError && (
+        <div className="xx-gen-progress-card" style={{ opacity: 0.85 }}>
+          <div className="xx-gen-progress-header">
+            <div className="xx-gen-progress-icon">🎬</div>
+            <div className="xx-gen-progress-info">
+              <div className="xx-gen-progress-phase">尚未开始生成视频</div>
+              <div className="xx-gen-progress-sub">
+                请返回「选择标题」步骤，点击「确认生成视频」开始渲染最终视频
+              </div>
+            </div>
+          </div>
+        </div>
+      )}
       {generating && (
         <div className="xx-gen-progress-card">
           <div className="xx-gen-progress-header">

apps/web/src/pages/generate/generate.css

@@ -1409,154 +1409,15 @@
   color: var(--text-tertiary, #94a3b8);
 }
 
-.xx-smart-match-results {
-  display: flex;
-  flex-direction: column;
-  gap: 12px;
-}
-
-.xx-smart-match-results-header {
-  display: flex;
-  justify-content: space-between;
-  align-items: center;
-}
-
-.xx-smart-match-results-title {
-  font-size: 14px;
-  font-weight: 600;
-  color: var(--text-primary, #1e293b);
-}
-
-.xx-smart-match-results-actions {
+.xx-smart-match-success {
   display: flex;
   align-items: center;
+  justify-content: center;
   gap: 8px;
-  font-size: 12px;
-}
-
-.xx-link-btn {
-  background: none;
-  border: none;
-  color: var(--primary-color, #4f46e5);
-  font-size: 12px;
-  cursor: pointer;
-  padding: 2px 4px;
-}
-
-.xx-link-btn:hover {
-  text-decoration: underline;
-}
-
-.xx-smart-match-grid {
-  display: grid;
-  grid-template-columns: repeat(2, 1fr);
-  gap: 10px;
-  max-height: 420px;
-  overflow-y: auto;
-  padding-right: 4px;
-}
-
-.xx-smart-match-card {
-  background: #fff;
-  border: 2px solid var(--border-primary, #e2e8f0);
+  padding: 18px 20px;
+  background: #f0fdf4;
+  border: 1px solid #bbf7d0;
   border-radius: 12px;
-  overflow: hidden;
-  cursor: pointer;
-  transition: all 0.2s ease;
-}
-
-.xx-smart-match-card:hover {
-  border-color: var(--primary-color, #4f46e5);
-  transform: translateY(-2px);
-  box-shadow: 0 4px 12px rgba(0, 0, 0, 0.08);
-}
-
-.xx-smart-match-card.selected {
-  border-color: var(--primary-color, #4f46e5);
-  background: var(--primary-soft, #eef2ff);
-}
-
-.xx-smart-match-thumb {
-  position: relative;
-  width: 100%;
-  aspect-ratio: 9 / 16;
-  background: #f1f5f9;
-  overflow: hidden;
-}
-
-.xx-smart-match-thumb img {
-  width: 100%;
-  height: 100%;
-  object-fit: cover;
-}
-
-.xx-smart-match-thumb-placeholder {
-  width: 100%;
-  height: 100%;
-  display: flex;
-  align-items: center;
-  justify-content: center;
-  color: var(--text-tertiary, #94a3b8);
-}
-
-.xx-smart-match-score {
-  position: absolute;
-  top: 8px;
-  left: 8px;
-  padding: 2px 8px;
-  font-size: 11px;
-  font-weight: 600;
-  color: #fff;
-  background: linear-gradient(135deg, #4f46e5, #7c3aed);
-  border-radius: 12px;
-}
-
-.xx-smart-match-check {
-  position: absolute;
-  top: 8px;
-  right: 8px;
-  width: 24px;
-  height: 24px;
-  background: var(--primary-color, #4f46e5);
-  border-radius: 50%;
-  display: flex;
-  align-items: center;
-  justify-content: center;
-}
-
-.xx-smart-match-duration {
-  position: absolute;
-  bottom: 8px;
-  right: 8px;
-  padding: 2px 6px;
-  font-size: 11px;
-  color: #fff;
-  background: rgba(0, 0, 0, 0.6);
-  border-radius: 4px;
-}
-
-.xx-smart-match-info {
-  padding: 10px 12px;
-}
-
-.xx-smart-match-name {
-  font-size: 13px;
-  font-weight: 500;
-  color: var(--text-primary, #1e293b);
-  white-space: nowrap;
-  overflow: hidden;
-  text-overflow: ellipsis;
-  margin-bottom: 4px;
-}
-
-.xx-smart-match-reason {
-  font-size: 11px;
-  color: var(--text-secondary, #64748b);
-  line-height: 1.4;
-  display: -webkit-box;
-  -webkit-line-clamp: 2;
-  -webkit-box-orient: vertical;
-  overflow: hidden;
 }
 
 .xx-smart-match-loading {
@@ -1580,19 +1441,6 @@
   text-align: center;
 }
 
-.xx-smart-match-summary {
-  padding: 12px 16px;
-  background: #f0fdf4;
-  border: 1px solid #bbf7d0;
-  border-radius: 12px;
-}
-
-.xx-smart-match-summary-header {
-  display: flex;
-  justify-content: space-between;
-  align-items: center;
-}
-
 /* ============================================================
    AI 智能生成标题
    ============================================================ */

apps/web/src/pages/generate/hooks/step2-materials/useMaterialLibrary.ts

@@ -1,6 +1,6 @@
-import { useState, useEffect } from "react"
+import { useState, useEffect, useMemo } from "react"
 import { useQuery } from "@tanstack/react-query"
-import { getAssets, getAssetLibraries } from "@/api/assets"
+import { getAssets, getAssetLibraries, isAssetUsable } from "@/api/assets"
 import type { AssetItem } from "@/api/assets"
 
 /**
@@ -31,11 +31,26 @@ export function useMaterialLibrary() {
     enabled: !!selectedLibraryId,
   })
 
+  // 生成选片只展示仍可切出不重复片段的素材（usable !== false）；
+  // 后端字段未上线时 isAssetUsable 恒为 true，过滤为 no-op
+  const selectableMaterials = useMemo(
+    () => ({
+      items: materials.items.filter(isAssetUsable),
+      total: materials.total,
+    }),
+    [materials],
+  )
+
+  // 库内有素材但全部已用尽（用于区分空状态文案）
+  const allMaterialsExhausted = materials.items.length > 0 && selectableMaterials.items.length === 0
+
   return {
     libraries,
     selectedLibraryId,
     setSelectedLibraryId,
     materials,
+    selectableMaterials,
+    allMaterialsExhausted,
     materialsLoading,
   }
 }

apps/web/src/pages/generate/hooks/step2-materials/useSmartMatch.ts

@@ -1,29 +1,26 @@
-import { useState, useCallback, useMemo } from "react"
+import { useState, useCallback } from "react"
 import { message } from "antd"
 import type { AssetItem } from "@/api/assets"
-import { smartMatchAssets } from "@/api/assets"
+import { smartMatchAssets, isAssetUsable } from "@/api/assets"
 
 interface UseSmartMatchOptions {
   libraryId: string
   materials: { items: AssetItem[]; total: number }
-  smartSelectedIds: string[]
   onSmartSelectedIdsChange: (ids: string[]) => void
 }
 
 /**
- * 智能素材匹配 Hook（Q5 简化版）
- * 用户不手动选素材时，一键调用后端 AI 选素材
- * 后端统一选素材逻辑后续完善，当前先走前端流程简化
+ * 智能素材匹配 Hook
+ * 用户不手动选素材时，一键调用后端 AI 选素材；匹配结果自动全量写入
+ * smartSelectedIds（由上层持有），不向用户展示素材明细。
  */
 export function useSmartMatch({
   libraryId,
   materials,
-  smartSelectedIds,
   onSmartSelectedIdsChange,
 }: UseSmartMatchOptions) {
   const [smartMatching, setSmartMatching] = useState(false)
   const [hasMatched, setHasMatched] = useState(false)
-  const [smartMatchedResults, setSmartMatchedResults] = useState<AssetItem[]>([])
 
   /* ── 一键智能匹配 ── */
   const handleSmartMatch = useCallback(async () => {
@@ -32,38 +29,39 @@ export function useSmartMatch({
       return
     }
 
-    if (materials.items.length === 0) {
-      message.warning("当前视频库暂无素材")
+    // 已用尽素材（usable === false）不参与智能匹配；
+    // 后端字段未上线时 isAssetUsable 恒为 true，过滤为 no-op
+    const usableItems = materials.items.filter(isAssetUsable)
+    if (usableItems.length === 0) {
+      message.warning(
+        materials.items.length === 0 ? "当前视频库暂无素材" : "素材可能已用尽，请先上传新素材",
+      )
       return
     }
 
     setSmartMatching(true)
 
     try {
-      // 调用后端智能匹配 API
+      // 调用后端智能匹配 API（后端也会排除已用尽素材，这里前端兜底过滤）
+      // items 已在 API 层归一化（兼容后端 {asset, score} 包装结构）；
+      // 这里再过滤一遍无 id/已用尽项，杜绝 undefined id 流入预览链路
       const result = await smartMatchAssets(libraryId)
-      const matchedIds = result.items?.map((a: AssetItem) => a.id) ?? []
+      const matched = (result.items ?? []).filter((a) => !!a?.id && isAssetUsable(a))
+      const matchedIds = matched.map((a) => a.id)
 
       if (matchedIds.length > 0) {
         onSmartSelectedIdsChange(matchedIds)
-        // 保存 API 返回的完整素材列表
-        const resolved = result.items?.length
-          ? result.items
-          : materials.items.filter((a) => matchedIds.includes(a.id))
-        setSmartMatchedResults(resolved)
         setHasMatched(true)
         message.success(`AI 已为你选择 ${matchedIds.length} 个素材`)
       } else {
-        // 后端返回空结果，回退到全选
-        onSmartSelectedIdsChange(materials.items.map((a) => a.id))
-        setSmartMatchedResults(materials.items)
+        // 后端返回空结果，回退到全选可用素材
+        onSmartSelectedIdsChange(usableItems.map((a) => a.id))
         setHasMatched(true)
         message.info("AI 暂未找到匹配素材，已全选当前库素材")
       }
     } catch {
-      // 后端 API 尚未就绪时，回退到全选当前库素材
-      onSmartSelectedIdsChange(materials.items.map((a) => a.id))
-      setSmartMatchedResults(materials.items)
+      // 后端 API 尚未就绪时，回退到全选当前库可用素材
+      onSmartSelectedIdsChange(usableItems.map((a) => a.id))
       setHasMatched(true)
       message.info("已为你全选当前库素材（智能匹配功能即将上线）")
     } finally {
@@ -71,49 +69,15 @@ export function useSmartMatch({
     }
   }, [libraryId, materials.items, onSmartSelectedIdsChange])
 
+  /* ── 换一批 = 重新触发智能匹配 ── */
   const handleRefreshMatch = useCallback(async () => {
-    // 换一批 = 重新触发智能匹配
     await handleSmartMatch()
   }, [handleSmartMatch])
 
-  const handleSelectAllMatched = useCallback(() => {
-    onSmartSelectedIdsChange(materials.items.map((a) => a.id))
-  }, [materials.items, onSmartSelectedIdsChange])
-
-  const handleClearSmartSelect = useCallback(() => {
-    onSmartSelectedIdsChange([])
-  }, [onSmartSelectedIdsChange])
-
-  const handleToggleSmartSelect = useCallback(
-    (assetId: string) => {
-      onSmartSelectedIdsChange(
-        smartSelectedIds.includes(assetId)
-          ? smartSelectedIds.filter((id) => id !== assetId)
-          : [...smartSelectedIds, assetId],
-      )
-    },
-    [smartSelectedIds, onSmartSelectedIdsChange],
-  )
-
-  /* ── 计算已选素材总时长 ── */
-  const smartSelectedTotalDuration = useMemo(
-    () =>
-      materials.items
-        .filter((a) => smartSelectedIds.includes(a.id))
-        .reduce((sum, a) => sum + (a.duration || 0), 0),
-    [materials.items, smartSelectedIds],
-  )
-
   return {
     smartMatching,
     hasMatched,
-    smartSelectedIds,
-    smartMatchedResults,
     handleSmartMatch,
-    handleToggleSmartSelect,
     handleRefreshMatch,
-    handleSelectAllMatched,
-    handleClearSmartSelect,
-    smartSelectedTotalDuration,
   }
 }

apps/web/src/pages/generate/hooks/useGenerateFormState/usePlanConfigLoader.ts

@@ -48,7 +48,7 @@ export function usePlanConfigLoader({
           ...prev,
           title: tc.content || "",
           aiAutoSelect: tc.ai_auto_select || false,
-          position: tc.position || prev.position,
+          position: prev.position, // 强制保留默认/用户选择，不从草稿配置同步位置
           font: tc.font_preset || prev.font,
           size: tc.font_size || prev.size,
           color: tc.font_color || prev.color,
@@ -80,7 +80,7 @@ export function usePlanConfigLoader({
             ...prev,
             aiAutoSelect: cfg.title_config!.ai_auto_select,
             title: cfg.title_config!.content || prev.title,
-            position: cfg.title_config!.position || prev.position,
+            position: prev.position, // 强制保留默认/用户选择，不从远程草稿同步位置
             font: cfg.title_config!.font_preset || prev.font,
             size: cfg.title_config!.font_size || prev.size,
             color: cfg.title_config!.font_color || prev.color,

apps/web/src/pages/generate/hooks/useGenerateFormState/useTitleCoverSync.ts

@@ -26,7 +26,7 @@ export function useTitleCoverSync({
         ...prev,
         aiAutoSelect: tpl.title_config!.ai_auto_select,
         title: tpl.title_config!.content || prev.title,
-        position: tpl.title_config!.position || prev.position,
+        position: prev.position, // 强制保留默认/用户选择，不从模板同步位置
         font: tpl.title_config!.font_preset || prev.font,
         size: tpl.title_config!.font_size || prev.size,
         color: tpl.title_config!.font_color || prev.color,

apps/web/src/pages/generate/hooks/useGenerateVideo.ts

@@ -44,12 +44,13 @@ export function useGenerateVideo(props: UseGenerateVideoProps) {
     onFailed: handleFailed,
   })
 
-  /* ── 生成视频 ── */
-  const generate = useCallback(async () => {
+  /* ── 生成视频 ──
+     返回 true 表示任务创建成功并已开始轮询；false 表示校验未通过或创建失败 */
+  const generate = useCallback(async (): Promise<boolean> => {
     const errorMsg = validateGenerateInputs(props)
     if (errorMsg) {
       message.warning(errorMsg)
-      return
+      return false
     }
 
     setGenerating(true)
@@ -142,7 +143,9 @@ export function useGenerateVideo(props: UseGenerateVideoProps) {
       const finalMsg = translateError(backendMsg)
       setGenerateError(finalMsg)
       message.error(finalMsg)
+      return false
     }
+    return true
   }, [props, clearTimer, startPolling, selectedTemplate])
 
   const retry = useCallback(() => {

apps/web/src/pages/generate/hooks/usePreviewAssets.ts

@@ -15,12 +15,14 @@ import type { AxiosResponse } from "axios"
  * 使用 Promise.allSettled 确保单个失败不影响整体
  */
 async function fetchAssetsByIds(ids: string[]): Promise<AssetItem[]> {
-  if (!ids.length) return []
+  // 防御：过滤空值/undefined/非字符串 id，避免发出 /assets/undefined 请求
+  const validIds = ids.filter((id): id is string => typeof id === "string" && id.length > 0)
+  if (!validIds.length) return []
 
   try {
     const { default: apiClient } = await import("@/api/client")
     const results = await Promise.allSettled(
-      ids.map((id) => apiClient.get<AssetItem>(`/assets/${id}`)),
+      validIds.map((id) => apiClient.get<AssetItem>(`/assets/${id}`)),
     )
     return results
       .filter(
@@ -57,7 +59,10 @@ export function usePreviewAssets(assetIds: string[], enabled: boolean): UsePrevi
   const stableAssetIds = useStableArray(assetIds)
 
   const load = useCallback(async () => {
-    if (!stableAssetIds.length || !enabled) {
+    const validIds = stableAssetIds.filter(
+      (id): id is string => typeof id === "string" && id.length > 0,
+    )
+    if (!validIds.length || !enabled) {
       setAssets([])
       setReady(false)
       return
@@ -68,7 +73,7 @@ export function usePreviewAssets(assetIds: string[], enabled: boolean): UsePrevi
     setReady(false)
 
     try {
-      const result = await fetchAssetsByIds(stableAssetIds)
+      const result = await fetchAssetsByIds(validIds)
       // 防止竞态：只保留最新请求的结果
       if (requestIdRef.current === thisRequestId) {
         setAssets(result)

apps/web/src/pages/generate/hooks/useStep2Materials.ts

@@ -7,7 +7,6 @@ import { message } from "antd"
 import type { TemplateSegment } from "@/api/templates/types"
 import type { EditPlanClip } from "@/api/template-editor"
 import { updateEditPlanClips, createClipsFromAssets, getEditPlanClips } from "@/api/template-editor"
-import { formatDuration } from "../utils/formatDuration"
 import { useMaterialLibrary } from "./step2-materials/useMaterialLibrary"
 import { useSmartMatch } from "./step2-materials/useSmartMatch"
 import { useDraftAutoSave } from "./useDraftAutoSave"
@@ -38,13 +37,19 @@ export function useStep2Materials({
   templateSegments,
   onServerClipsChange,
 }: UseStep2MaterialsProps) {
-  const { libraries, selectedLibraryId, setSelectedLibraryId, materials, materialsLoading } =
-    useMaterialLibrary()
+  const {
+    libraries,
+    selectedLibraryId,
+    setSelectedLibraryId,
+    materials,
+    selectableMaterials,
+    allMaterialsExhausted,
+    materialsLoading,
+  } = useMaterialLibrary()
 
   const smartMatch = useSmartMatch({
     libraryId: selectedLibraryId,
-    materials,
-    smartSelectedIds,
+    materials: selectableMaterials,
     onSmartSelectedIdsChange,
   })
 
@@ -59,13 +64,19 @@ export function useStep2Materials({
     }
     if (!selectedLibraryId) return
     if (materialsLoading) return
-    if (materials.items.length === 0) return
+    if (selectableMaterials.items.length === 0) return
     // 防止同一视频库重复触发
     if (autoTriggeredRef.current === selectedLibraryId) return
 
     autoTriggeredRef.current = selectedLibraryId
     handleSmartMatch()
-  }, [selectedLibraryId, materialMode, materialsLoading, materials.items, handleSmartMatch])
+  }, [
+    selectedLibraryId,
+    materialMode,
+    materialsLoading,
+    selectableMaterials.items,
+    handleSmartMatch,
+  ])
 
   /* ── Step2 选择素材后自动保存草稿 asset_ids（防抖 500ms，失败静默） ── */
   const { scheduleSave } = useDraftAutoSave(selectedTemplate)
@@ -169,6 +180,8 @@ export function useStep2Materials({
     selectedLibraryId,
     setSelectedLibraryId,
     materials,
+    selectableMaterials,
+    allMaterialsExhausted,
     materialsLoading,
     // 模式
     materialMode,
@@ -176,19 +189,11 @@ export function useStep2Materials({
     // 手动选择
     selectedMaterials,
     handleToggleMaterial,
-    // 智能匹配（简化版）
+    // 智能匹配
     smartMatching: smartMatch.smartMatching,
     hasMatched: smartMatch.hasMatched,
-    smartSelectedIds: smartMatch.smartSelectedIds,
-    smartMatchedResults: smartMatch.smartMatchedResults,
     handleSmartMatch: smartMatch.handleSmartMatch,
-    handleToggleSmartSelect: smartMatch.handleToggleSmartSelect,
     handleRefreshMatch: smartMatch.handleRefreshMatch,
-    handleSelectAllMatched: smartMatch.handleSelectAllMatched,
-    handleClearSmartSelect: smartMatch.handleClearSmartSelect,
-    smartSelectedTotalDuration: smartMatch.smartSelectedTotalDuration,
-    // utils
-    formatDuration,
   }
 }
 

apps/web/src/pages/generate/hooks/useStep7Generate.ts

@@ -87,7 +87,9 @@ export function useStep7Generate({
   }
 
   const handleScrollToPreview = () => {
-    const el = document.querySelector(".xx-preview-section")
+    const el =
+      document.querySelector(".xx-inline-video-player") ||
+      document.querySelector(".xx-preview-section")
     el?.scrollIntoView({ behavior: "smooth", block: "start" })
   }
 

apps/web/src/pages/voice-materials/VoiceMaterialLibrary.tsx

@@ -102,6 +102,7 @@ const VoiceMaterialLibrary: React.FC = () => {
     ttsAudioUrl,
     ttsError,
     presetVoices,
+    clonedVoices,
     setTtsOpen,
     setTtsText,
     setTtsVoiceId,
@@ -318,6 +319,7 @@ const VoiceMaterialLibrary: React.FC = () => {
           audioUrl={ttsAudioUrl ?? ""}
           error={ttsError ?? ""}
           presetVoices={presetVoices}
+          clonedVoices={clonedVoices}
           onClose={handleTtsClose}
           onTextChange={setTtsText}
           onVoiceChange={setTtsVoiceId}

apps/web/src/pages/voice-materials/components/TtsModal.tsx

@@ -9,6 +9,12 @@ export interface TtsPresetVoice {
   name: string
 }
 
+/** 克隆音色下拉选项（仅 ready） */
+export interface TtsClonedVoice {
+  id: string
+  name: string
+}
+
 interface TtsModalProps {
   open: boolean
   text: string
@@ -18,6 +24,8 @@ interface TtsModalProps {
   audioUrl: string
   error: string
   presetVoices: TtsPresetVoice[]
+  /** 可用克隆音色（仅 ready），为空时不显示该分组 */
+  clonedVoices?: TtsClonedVoice[]
   onClose: () => void
   onTextChange: (value: string) => void
   onVoiceChange: (voiceId: string) => void
@@ -35,6 +43,7 @@ const TtsModal: React.FC<TtsModalProps> = ({
   audioUrl,
   error,
   presetVoices,
+  clonedVoices = [],
   onClose: _onClose,
   onTextChange,
   onVoiceChange,
@@ -101,11 +110,22 @@ const TtsModal: React.FC<TtsModalProps> = ({
           }}
         >
           <option value="">默认音色</option>
-          {presetVoices.map((v) => (
-            <option key={v.voice_id} value={v.voice_id}>
-              {v.name}
-            </option>
-          ))}
+          <optgroup label="预置音色">
+            {presetVoices.map((v) => (
+              <option key={v.voice_id} value={v.voice_id}>
+                {v.name}
+              </option>
+            ))}
+          </optgroup>
+          {clonedVoices.length > 0 && (
+            <optgroup label="我的克隆音色">
+              {clonedVoices.map((v) => (
+                <option key={v.id} value={v.id}>
+                  {v.name}
+                </option>
+              ))}
+            </optgroup>
+          )}
         </select>
       </div>
 

apps/web/src/pages/voice-materials/components/voice-material-card/CardPlayer.tsx

@@ -54,7 +54,8 @@ const CardPlayer: React.FC<CardPlayerProps> = ({
     document.addEventListener("mouseup", handleUp)
   }
 
-  const progress = duration > 0 ? (currentTime / duration) * 100 : 0
+  // 仅播放中的卡片显示进度，避免页面级 currentTime 联动所有卡片
+  const progress = isPlaying && duration > 0 ? (currentTime / duration) * 100 : 0
 
   return (
     <div className="vmat-card-player">

apps/web/src/pages/voice-materials/hooks/useTtsSynthesize.ts

@@ -3,6 +3,7 @@ import { useQuery, useQueryClient } from "@tanstack/react-query"
 import { message } from "antd"
 import { synthesizeSpeech, getTTSJobStatus, saveTtsToLibrary } from "@/api/tts"
 import { fetchPresetVoices, type PresetVoiceItem } from "@/api/voices"
+import { getVoiceClonesWithTotal, toVoiceClone } from "@/api/voice-clone"
 
 /**
  * TTS 合成 Hook
@@ -31,6 +32,17 @@ export function useTtsSynthesize() {
   })
   const presetVoices: PresetVoiceItem[] = presetVoicesData?.items ?? []
 
+  // 我的克隆音色（仅 ready 可用于合成；voice_id 直接传 profile UUID，后端解析）
+  const { data: clonedData } = useQuery({
+    queryKey: ["voice-clones"],
+    queryFn: () => getVoiceClonesWithTotal({ limit: 50 }),
+    staleTime: 60_000,
+  })
+  const clonedVoices = (clonedData?.items ?? [])
+    .map((p) => toVoiceClone(p))
+    .filter((c) => c.status === "ready")
+    .map((c) => ({ id: c.id, name: c.name }))
+
   /** 开始 AI 配音合成 */
   const handleTtsSynthesize = useCallback(async () => {
     if (!ttsText.trim()) {
@@ -124,6 +136,7 @@ export function useTtsSynthesize() {
     ttsAudioUrl,
     ttsError,
     presetVoices,
+    clonedVoices,
     setTtsOpen,
     setTtsText,
     setTtsVoiceId,

apps/web/src/pages/voice-materials/hooks/useVoiceMaterials/actions/useVoiceUpload.ts

@@ -48,20 +48,32 @@ export function useVoiceUpload({ voiceLibrary, createLibMutation }: UseVoiceUplo
         }
 
         // 2. 上传文件（带进度，后端自动创建 ingest job）
-        const { ingest_job_id } = await uploadAssetDirect({
+        const complete = await uploadAssetDirect({
           file: data.file,
           library_id: lib.id,
           onProgress: (p) => setUploadProgress(p),
         })
 
-        // 3. 轮询 ingest job 状态
+        // 去重命中（同库已存在相同 file_hash 素材）：
+        // 后端返回 duplicated=true，ingest_job_id 为空；跳过轮询和打标签，
+        // mutationFn 正常 return 即 resolve，useMutation 自动触发 onSuccess 刷新列表。
+        // 已存在素材复用上次标签，无需重新打标。
+        if (complete.duplicated === true) {
+          return
+        }
+        if (!complete.ingest_job_id) {
+          throw new Error("上传完成但未返回处理任务 ID，请重试")
+        }
+        const { ingest_job_id } = complete
+
+        // 3. 轮询 ingest job 状态（complete 后先立即查一次，未完成再每 5s 轮询）
         let job: Awaited<ReturnType<typeof getIngestJob>> | null = null
         let retries = 0
         const maxRetries = 60 // 最多等待 5 分钟
-        while (retries < maxRetries) {
+        job = await getIngestJob(ingest_job_id)
+        while (job.status !== "completed" && job.status !== "failed" && retries < maxRetries) {
           await new Promise((r) => setTimeout(r, 5000))
           job = await getIngestJob(ingest_job_id)
-          if (job.status === "completed" || job.status === "failed") break
           retries++
         }
 
@@ -72,7 +84,7 @@ export function useVoiceUpload({ voiceLibrary, createLibMutation }: UseVoiceUplo
           throw new Error("音频处理超时，请稍后在素材库查看")
         }
 
-        // 4. 打标签（标签走独立 API）
+        // 4. 打标签（标签走独立 API；去重命中时已提前 return，这里只对新创建的素材执行）
         if (data.tagIds.length > 0 && job.result_asset_id) {
           await tagAsset(job.result_asset_id, data.tagIds)
         }

apps/web/src/pages/voices/VoiceLibrary.tsx

@@ -32,6 +32,8 @@ import { useAudioPlayer } from "./hooks/useAudioPlayer"
 import { useCloneOperations } from "./hooks/useCloneOperations"
 import { useTtsSynthesize } from "./hooks/useTtsSynthesize"
 import { useVoiceUpload } from "./hooks/useVoiceUpload"
+import { useMaterialDelete } from "./hooks/useMaterialDelete"
+import { useMaterialBatchDelete } from "./hooks/useMaterialBatchDelete"
 import "./voices.css"
 
 let toastIdSeq = 0
@@ -70,17 +72,39 @@ const VoiceLibrary: React.FC = () => {
     materialCount,
   } = useVoicesData()
 
-  // ── 播放控制 ──────────────────────────────────────────
+  // ── 播放控制（三 tab 共用：真实 Audio 播放 + TTS 试听） ──
   const {
     playingId,
+    loadingId,
     currentTime,
-    handlePlay,
+    duration: playDuration,
     handlePause,
+    handleToggleVoice,
+    handleToggleMaterial,
     handleSeek,
-    handleTogglePlay,
     stopPlayback,
   } = useAudioPlayer()
 
+  // ── 配音素材删除 / 批量删除（删除正在播放的素材时停止播放） ──
+  const { handleMaterialDelete } = useMaterialDelete({
+    materials: materials as AssetItem[],
+    stopPlayback,
+    showToast,
+  })
+  const {
+    selectedIds: materialSelectedIds,
+    selectedCount: materialSelectedCount,
+    allSelected: materialAllSelected,
+    batchDeleting: materialBatchDeleting,
+    toggleSelect: toggleMaterialSelect,
+    toggleSelectAll: toggleMaterialSelectAll,
+    handleBatchDelete: handleMaterialBatchDelete,
+  } = useMaterialBatchDelete({
+    materials: materials as AssetItem[],
+    stopPlayback,
+    showToast,
+  })
+
   // ── 克隆音色操作 ──────────────────────────────────────
   const {
     detailVoice,
@@ -110,7 +134,13 @@ const VoiceLibrary: React.FC = () => {
     handleTtsSynthesize,
     handleTtsSave,
     handleTtsClose,
-  } = useTtsSynthesize({ presetVoices, showToast })
+  } = useTtsSynthesize({
+    presetVoices,
+    clonedVoices: clonedVoices
+      .filter((v) => v.status === "ready")
+      .map((v) => ({ id: v.id, name: v.name })),
+    showToast,
+  })
 
   // ── 上传音频 ──────────────────────────────────────────
   const {
@@ -203,8 +233,10 @@ const VoiceLibrary: React.FC = () => {
           loading={presetLoading}
           voices={filteredPreset}
           playingId={playingId}
+          loadingId={loadingId}
           currentTime={currentTime}
-          onPlay={handlePlay}
+          playDuration={playDuration}
+          onToggle={handleToggleVoice}
           onPause={handlePause}
           onSeek={handleSeek}
           onClearFilters={handleClearFilters}
@@ -217,9 +249,12 @@ const VoiceLibrary: React.FC = () => {
           loading={cloneLoading}
           voices={clonedVoices}
           playingId={playingId}
+          loadingId={loadingId}
           currentTime={currentTime}
-          onPlay={handleTogglePlay}
+          playDuration={playDuration}
+          onToggle={handleToggleVoice}
           onPause={handlePause}
+          onSeek={handleSeek}
           onUse={handleCloneUse}
           onDelete={handleCloneDelete}
           onRetry={handleCloneRetry}
@@ -234,6 +269,19 @@ const VoiceLibrary: React.FC = () => {
           loading={materialLoading}
           materials={materials as AssetItem[]}
           onOpenUpload={() => setUploadOpen(true)}
+          onDelete={handleMaterialDelete}
+          selectedIds={materialSelectedIds}
+          selectedCount={materialSelectedCount}
+          allSelected={materialAllSelected}
+          batchDeleting={materialBatchDeleting}
+          onToggleSelect={toggleMaterialSelect}
+          onToggleSelectAll={toggleMaterialSelectAll}
+          onBatchDelete={handleMaterialBatchDelete}
+          playingId={playingId}
+          currentTime={currentTime}
+          playDuration={playDuration}
+          onTogglePlay={handleToggleMaterial}
+          onSeek={handleSeek}
         />
       )}
 
@@ -266,6 +314,9 @@ const VoiceLibrary: React.FC = () => {
         ttsAudioUrl={ttsAudioUrl}
         ttsError={ttsError}
         presetVoices={presetVoices}
+        clonedVoices={clonedVoices
+          .filter((v) => v.status === "ready")
+          .map((v) => ({ id: v.id, name: v.name }))}
         onTtsClose={handleTtsClose}
         onTtsTextChange={setTtsText}
         onTtsVoiceChange={setTtsVoiceId}

apps/web/src/pages/voices/components/CloneVoiceCard.tsx

@@ -8,9 +8,12 @@ import CardFooter from "./clone-voice-card/CardFooter"
 export interface CloneVoiceCardProps {
   voice: ClonedVoiceDisplay
   isPlaying: boolean
+  isLoading?: boolean
   currentTime: number
+  playDuration?: number
   onPlay: () => void
   onPause: () => void
+  onSeek?: (time: number) => void
   onUse: () => void
   onDelete: () => void
   onRetry: () => void

apps/web/src/pages/voices/components/ClonedVoiceTab.tsx

@@ -12,9 +12,12 @@ export interface ClonedVoiceTabProps {
   loading: boolean
   voices: ClonedVoiceDisplay[]
   playingId: string | null
+  loadingId: string | null
   currentTime: number
-  onPlay: (voiceId: string, duration: number) => void
+  playDuration: number
+  onToggle: (voiceId: string) => void
   onPause: () => void
+  onSeek: (time: number) => void
   onUse: (voice: ClonedVoiceDisplay) => void
   onDelete: (voice: ClonedVoiceDisplay) => void
   onRetry: (voice: ClonedVoiceDisplay) => void
@@ -26,9 +29,12 @@ export const ClonedVoiceTab: React.FC<ClonedVoiceTabProps> = ({
   loading,
   voices,
   playingId,
+  loadingId,
   currentTime,
-  onPlay,
+  playDuration,
+  onToggle,
   onPause,
+  onSeek,
   onUse,
   onDelete,
   onRetry,
@@ -54,9 +60,12 @@ export const ClonedVoiceTab: React.FC<ClonedVoiceTabProps> = ({
               key={voice.id}
               voice={voice}
               isPlaying={playingId === voice.id}
+              isLoading={loadingId === voice.id}
               currentTime={playingId === voice.id ? currentTime : 0}
-              onPlay={() => onPlay(voice.id, voice.duration)}
+              playDuration={playingId === voice.id ? playDuration : voice.duration}
+              onPlay={() => onToggle(voice.voiceId)}
               onPause={onPause}
+              onSeek={onSeek}
               onUse={() => onUse(voice)}
               onDelete={() => onDelete(voice)}
               onRetry={() => onRetry(voice)}

apps/web/src/pages/voices/components/MaterialVoiceCard.tsx

@@ -1,37 +0,0 @@
-import React from "react"
-import { AudioOutlined } from "@ant-design/icons"
-import { type AssetItem } from "@/api/assets"
-import { formatFileSize } from "@/pages/voices/utils/format"
-
-export interface MaterialVoiceCardProps {
-  asset: AssetItem
-  onClick?: () => void
-}
-
-/** 配音素材卡片 */
-const MaterialVoiceCard: React.FC<MaterialVoiceCardProps> = ({ asset, onClick }) => {
-  const duration = (asset.metadata?.duration as number) || 0
-  const minutes = Math.floor(duration / 60)
-  const seconds = Math.floor(duration % 60)
-
-  return (
-    <div className="vmat-card" onClick={onClick}>
-      <div className="vmat-thumb">
-        <AudioOutlined className="vmat-thumb-icon" />
-        <span className="vmat-duration">
-          {minutes}:{seconds.toString().padStart(2, "0")}
-        </span>
-      </div>
-      <div className="vmat-info">
-        <div className="vmat-name" title={asset.name}>
-          {asset.name}
-        </div>
-        <div className="vmat-meta">
-          <span>{asset.file_size ? formatFileSize(asset.file_size) : "--"}</span>
-        </div>
-      </div>
-    </div>
-  )
-}
-
-export default MaterialVoiceCard

apps/web/src/pages/voices/components/MaterialVoiceTab.tsx

@@ -1,31 +1,126 @@
 /**
  * VoiceLibrary 配音素材 Tab 内容
+ *
+ * 卡片视觉与预置音色卡片（xx-voice-card）一致：白底圆角横向布局、
+ * 左侧圆形图标、名称/副标题、波形条、底部播放控件。
+ * 删除按钮常驻右上角；支持全选/多选 + 批量删除。
+ * 播放状态由页面级统一 hook（VoiceLibrary/useAudioPlayer）下发。
  */
-import React from "react"
-import { SoundOutlined, UploadOutlined, AudioOutlined } from "@ant-design/icons"
+import React, { useRef } from "react"
+import {
+  SoundOutlined,
+  UploadOutlined,
+  AudioOutlined,
+  DeleteOutlined,
+  PlayCircleOutlined,
+  PauseCircleOutlined,
+  CheckOutlined,
+} from "@ant-design/icons"
 import { Button } from "@/components/ui"
 import type { AssetItem } from "@/api/assets"
+import { mapAssetToMaterial, type VoiceMaterial } from "@/pages/voice-materials/types"
+import { formatTime, formatFileSize } from "../utils/format"
 
 export interface MaterialVoiceTabProps {
   loading: boolean
   materials: AssetItem[]
   onOpenUpload: () => void
+  onDelete: (asset: AssetItem) => void
+  // 批量删除
+  selectedIds: Set<string>
+  selectedCount: number
+  allSelected: boolean
+  batchDeleting: boolean
+  onToggleSelect: (id: string) => void
+  onToggleSelectAll: () => void
+  onBatchDelete: () => void
+  // 播放控制（页面级统一 hook）
+  playingId: string | null
+  currentTime: number
+  playDuration: number
+  onTogglePlay: (material: VoiceMaterial) => void
+  onSeek: (time: number) => void
 }
 
 export const MaterialVoiceTab: React.FC<MaterialVoiceTabProps> = ({
   loading,
   materials,
   onOpenUpload,
+  onDelete,
+  selectedIds,
+  selectedCount,
+  allSelected,
+  batchDeleting,
+  onToggleSelect,
+  onToggleSelectAll,
+  onBatchDelete,
+  playingId,
+  currentTime,
+  playDuration,
+  onTogglePlay,
+  onSeek,
 }) => {
+  const progressRefs = useRef<Record<string, HTMLDivElement | null>>({})
+
+  /** 播放中点击进度条 seek；非播放态点击则开始播放 */
+  const handleProgressClick =
+    (asset: AssetItem, material: VoiceMaterial, dur: number) =>
+    (e: React.MouseEvent<HTMLDivElement>) => {
+      e.stopPropagation()
+      if (!asset.file_url) return
+      if (playingId !== asset.id) {
+        onTogglePlay(material)
+        return
+      }
+      const el = progressRefs.current[asset.id]
+      if (!el || dur <= 0) return
+      const rect = el.getBoundingClientRect()
+      const percent = Math.max(0, Math.min(1, (e.clientX - rect.left) / rect.width))
+      onSeek(percent * dur)
+    }
+
   return (
     <div className="xx-voices-tab-content">
+      {/* 批量操作栏 */}
+      {!loading && materials.length > 0 && (
+        <div className="vmat-batch-bar">
+          <div className="vmat-batch-bar-left">
+            <button
+              type="button"
+              className={`vmat-checkbox${allSelected ? " checked" : ""}`}
+              onClick={onToggleSelectAll}
+              aria-label={allSelected ? "取消全选" : "全选"}
+            >
+              {allSelected && <CheckOutlined />}
+            </button>
+            <button type="button" className="vmat-select-all" onClick={onToggleSelectAll}>
+              {allSelected ? "取消全选" : "全选"}
+            </button>
+            {selectedCount > 0 && (
+              <span className="vmat-batch-count">已选择 {selectedCount} 项</span>
+            )}
+          </div>
+          <div className="vmat-batch-bar-right">
+            <Button
+              buttonType="danger"
+              buttonSize="sm"
+              icon={<DeleteOutlined />}
+              disabled={selectedCount === 0 || batchDeleting}
+              onClick={onBatchDelete}
+            >
+              {batchDeleting ? "删除中..." : "批量删除"}
+            </Button>
+          </div>
+        </div>
+      )}
+
       {/* 骨架屏加载 */}
       {loading && (
         <div className="xx-voice-grid">
           {Array.from({ length: 6 }).map((_, i) => (
-            <div key={i} className="vmat-card vmat-card--skeleton">
-              <div className="vmat-thumb" />
-              <div className="vmat-info">
+            <div key={i} className="xx-voice-card vmat-card--skeleton">
+              <div className="vmat-skeleton-avatar" />
+              <div className="xx-voice-info">
                 <div className="vmat-skeleton-line vmat-skeleton-title" />
                 <div className="vmat-skeleton-line" />
               </div>
@@ -38,27 +133,96 @@ export const MaterialVoiceTab: React.FC<MaterialVoiceTabProps> = ({
       {!loading && materials.length > 0 && (
         <div className="xx-voice-grid">
           {materials.map((asset: AssetItem) => {
-            const duration = (asset.metadata?.duration as number) || 0
-            const minutes = Math.floor(duration / 60)
-            const seconds = Math.floor(duration % 60)
+            const material = mapAssetToMaterial(asset)
+            // duration 优先取顶层（后端从 metadata 提取），兜底 metadata
+            const cardDuration = asset.duration || material.duration || 0
+            const isPlaying = playingId === asset.id
+            const isSelected = selectedIds.has(asset.id)
+            // 播放中以 audio 真实时长为准，未播放显示卡片时长
+            const effectiveDuration = isPlaying ? playDuration || cardDuration : cardDuration
+            // 仅播放中的卡片显示进度，避免页面级 currentTime 联动所有卡片
+            const progress =
+              isPlaying && effectiveDuration > 0 ? (currentTime / effectiveDuration) * 100 : 0
             return (
-              <div key={asset.id} className="vmat-card">
-                <div className="vmat-thumb">
-                  <AudioOutlined className="vmat-thumb-icon" />
-                  <span className="vmat-duration">
-                    {minutes}:{seconds.toString().padStart(2, "0")}
-                  </span>
+              <div
+                key={asset.id}
+                className={`xx-voice-card vmat-card${isSelected ? " selected" : ""}${
+                  isPlaying ? " playing" : ""
+                }`}
+              >
+                {/* 左上角批量选择 checkbox */}
+                <button
+                  type="button"
+                  className={`vmat-card-checkbox${isSelected ? " checked" : ""}`}
+                  onClick={(e) => {
+                    e.stopPropagation()
+                    onToggleSelect(asset.id)
+                  }}
+                  aria-label={isSelected ? "取消选择" : "选择素材"}
+                >
+                  {isSelected && <CheckOutlined />}
+                </button>
+
+                {/* 右上角删除按钮（常驻可见） */}
+                <button
+                  type="button"
+                  className="vmat-card-delete"
+                  onClick={(e) => {
+                    e.stopPropagation()
+                    onDelete(asset)
+                  }}
+                  title="删除"
+                  aria-label="删除配音素材"
+                >
+                  <DeleteOutlined />
+                </button>
+
+                <div className="xx-voice-avatar vmat-avatar">
+                  <AudioOutlined />
                 </div>
-                <div className="vmat-info">
-                  <div className="vmat-name" title={asset.name}>
+
+                <div className="xx-voice-info vmat-info">
+                  <div className="xx-voice-name" title={asset.name}>
                     {asset.name}
                   </div>
-                  <div className="vmat-meta">
-                    <span>
-                      {asset.file_size ? `${(asset.file_size / 1024 / 1024).toFixed(1)} MB` : "--"}
-                    </span>
+                  <div className="xx-voice-subtitle">
+                    {asset.file_size ? `${formatFileSize(asset.file_size)}` : "--"}
+                    {" · "}
+                    {cardDuration > 0 ? formatTime(cardDuration) : "--:--"}
                   </div>
                 </div>
+
+                {/* 波形装饰条（与预置音色卡片一致） */}
+                <div className="xx-voice-wave" />
+
+                {/* 播放控制区：真实音频播放 */}
+                <div className="xx-voice-controls">
+                  <button
+                    type="button"
+                    className="xx-voice-play-btn"
+                    disabled={!asset.file_url}
+                    onClick={(e) => {
+                      e.stopPropagation()
+                      onTogglePlay(material)
+                    }}
+                    title={asset.file_url ? (isPlaying ? "暂停" : "试听") : "暂无可播放音频"}
+                    aria-label={isPlaying ? "暂停播放" : "播放音频"}
+                  >
+                    {isPlaying ? <PauseCircleOutlined /> : <PlayCircleOutlined />}
+                  </button>
+                  <div
+                    ref={(el) => {
+                      progressRefs.current[asset.id] = el
+                    }}
+                    className="xx-voice-progress"
+                    onClick={handleProgressClick(asset, material, effectiveDuration)}
+                  >
+                    <div className="xx-voice-progress-bar" style={{ width: `${progress}%` }} />
+                  </div>
+                  <span className="xx-voice-time">
+                    {isPlaying ? formatTime(currentTime) : formatTime(cardDuration)}
+                  </span>
+                </div>
               </div>
             )
           })}

apps/web/src/pages/voices/components/PresetVoiceTab.tsx

@@ -19,10 +19,12 @@ export interface PresetVoiceTabProps {
   loading: boolean
   voices: PresetVoiceDisplay[]
   playingId: string | null
+  loadingId: string | null
   currentTime: number
-  onPlay: (id: string, duration: number) => void
+  playDuration: number
+  onToggle: (voiceId: string) => void
   onPause: () => void
-  onSeek: (id: string, time: number, duration: number) => void
+  onSeek: (time: number) => void
   onClearFilters: () => void
 }
 
@@ -36,8 +38,10 @@ export const PresetVoiceTab: React.FC<PresetVoiceTabProps> = ({
   loading,
   voices,
   playingId,
+  loadingId,
   currentTime,
-  onPlay,
+  playDuration,
+  onToggle,
   onPause,
   onSeek,
   onClearFilters,
@@ -71,15 +75,16 @@ export const PresetVoiceTab: React.FC<PresetVoiceTabProps> = ({
               name={voice.name}
               subtitle={`${genderLabel(voice.gender)} · ${languageLabel(voice.language)} · ${voice.description}`}
               tags={voice.tags}
-              duration={voice.duration}
+              duration={playingId === voice.id ? playDuration : voice.duration}
               gender={voice.gender}
               isPlaying={playingId === voice.id}
+              isLoading={loadingId === voice.id}
               isSelected={false}
               currentTime={playingId === voice.id ? currentTime : 0}
               starred={voice.starred}
-              onPlay={() => onPlay(voice.id, voice.duration)}
+              onPlay={() => onToggle(voice.voiceId)}
               onPause={onPause}
-              onSeek={(time) => onSeek(voice.id, time, voice.duration)}
+              onSeek={(time) => onSeek(time)}
               onToggleStar={() => {}}
             />
           ))}

apps/web/src/pages/voices/components/TtsModal.tsx

@@ -18,6 +18,7 @@ const TtsModal: React.FC<TtsModalProps> = ({
   ttsAudioUrl,
   ttsError,
   presetVoices,
+  clonedVoices,
   onClose,
   onTextChange,
   onVoiceChange,
@@ -36,7 +37,12 @@ const TtsModal: React.FC<TtsModalProps> = ({
         }}
       >
         <TextInputSection value={ttsText} onChange={onTextChange} />
-        <VoiceSelector value={ttsVoiceId} onChange={onVoiceChange} presetVoices={presetVoices} />
+        <VoiceSelector
+          value={ttsVoiceId}
+          onChange={onVoiceChange}
+          presetVoices={presetVoices}
+          clonedVoices={clonedVoices}
+        />
         <SpeedControl speed={ttsSpeed} onChange={onSpeedChange} />
         <SynthesizeButton status={ttsStatus} text={ttsText} onClick={onSynthesize} />
         {ttsError && <ErrorAlert error={ttsError} />}

apps/web/src/pages/voices/components/VoiceCard.tsx

@@ -4,6 +4,7 @@ import {
   PlayCircleOutlined,
   PauseCircleOutlined,
   HeartOutlined,
+  LoadingOutlined,
 } from "@ant-design/icons"
 import { type VoiceGender } from "@/pages/voices/types"
 import { genderClass, formatTime } from "@/pages/voices/utils/format"
@@ -16,6 +17,7 @@ export interface VoiceCardProps {
   duration: number
   gender: VoiceGender
   isPlaying: boolean
+  isLoading?: boolean
   isSelected: boolean
   currentTime: number
   starred?: boolean
@@ -36,6 +38,7 @@ const VoiceCard: React.FC<VoiceCardProps> = ({
   duration,
   gender,
   isPlaying,
+  isLoading = false,
   isSelected,
   currentTime,
   starred,
@@ -51,8 +54,12 @@ const VoiceCard: React.FC<VoiceCardProps> = ({
   const handleProgressClick = (e: React.MouseEvent<HTMLDivElement>) => {
     if (!progressRef.current || status !== "ready") return
     const rect = progressRef.current.getBoundingClientRect()
-    const percent = (e.clientX - rect.left) / rect.width
-    onSeek(percent * duration)
+    const percent = Math.max(0, Math.min(1, (e.clientX - rect.left) / rect.width))
+    if (isPlaying && duration > 0) {
+      onSeek(percent * duration)
+    } else if (!isPlaying && !isLoading) {
+      onPlay()
+    }
   }
 
   const progress = duration > 0 ? (currentTime / duration) * 100 : 0
@@ -110,19 +117,27 @@ const VoiceCard: React.FC<VoiceCardProps> = ({
         <div className="xx-voice-controls">
           <button
             className="xx-voice-play-btn"
+            disabled={isLoading}
             onClick={(e) => {
               e.stopPropagation()
               isPlaying ? onPause() : onPlay()
             }}
             title={isPlaying ? "暂停" : "试听"}
+            aria-label={isPlaying ? "暂停" : "试听"}
           >
-            {isPlaying ? <PauseCircleOutlined /> : <PlayCircleOutlined />}
+            {isLoading ? (
+              <LoadingOutlined spin />
+            ) : isPlaying ? (
+              <PauseCircleOutlined />
+            ) : (
+              <PlayCircleOutlined />
+            )}
           </button>
           <div ref={progressRef} className="xx-voice-progress" onClick={handleProgressClick}>
             <div className="xx-voice-progress-bar" style={{ width: `${progress}%` }} />
           </div>
           <span className="xx-voice-time">
-            {isPlaying ? formatTime(currentTime) : formatTime(duration)}
+            {isPlaying ? formatTime(currentTime) : duration > 0 ? formatTime(duration) : "试听"}
           </span>
         </div>
       )}

apps/web/src/pages/voices/components/VoiceModals.tsx

@@ -5,6 +5,7 @@ import React from "react"
 import type { ClonedVoiceDisplay, PresetVoiceDisplay } from "../types"
 import type { VoiceClone } from "@/api/voice-clone"
 import type { TtsStatus } from "./TtsModal"
+import type { TtsClonedVoiceOption } from "./tts-modal/VoiceSelector"
 import CloneModal from "@/components/voice/CloneModal"
 import CloneDetailModal from "./CloneDetailModal"
 import UploadVoiceModal from "./UploadVoiceModal"
@@ -45,6 +46,8 @@ export interface VoiceModalsProps {
   ttsAudioUrl: string | null
   ttsError: string | null
   presetVoices: PresetVoiceDisplay[]
+  /** 可用克隆音色（仅 ready） */
+  clonedVoices?: TtsClonedVoiceOption[]
   onTtsClose: () => void
   onTtsTextChange: (text: string) => void
   onTtsVoiceChange: (id: string) => void
@@ -81,6 +84,7 @@ export const VoiceModals: React.FC<VoiceModalsProps> = ({
   ttsAudioUrl,
   ttsError,
   presetVoices,
+  clonedVoices,
   onTtsClose,
   onTtsTextChange,
   onTtsVoiceChange,
@@ -129,6 +133,7 @@ export const VoiceModals: React.FC<VoiceModalsProps> = ({
         ttsAudioUrl={ttsAudioUrl}
         ttsError={ttsError}
         presetVoices={presetVoices}
+        clonedVoices={clonedVoices}
         onClose={onTtsClose}
         onTextChange={onTtsTextChange}
         onVoiceChange={onTtsVoiceChange}

apps/web/src/pages/voices/components/clone-voice-card/CardFooter.tsx

@@ -1,13 +1,21 @@
 import React from "react"
-import { PlayCircleOutlined, PauseCircleOutlined, ReloadOutlined } from "@ant-design/icons"
+import {
+  PlayCircleOutlined,
+  PauseCircleOutlined,
+  ReloadOutlined,
+  LoadingOutlined,
+} from "@ant-design/icons"
 import { type ClonedVoiceDisplay } from "@/pages/voices/types"
 
 interface CardFooterProps {
   voice: ClonedVoiceDisplay
   isPlaying: boolean
+  isLoading?: boolean
   currentTime: number
+  playDuration?: number
   onPlay: () => void
   onPause: () => void
+  onSeek?: (time: number) => void
   onUse: () => void
   onRetry: () => void
 }
@@ -15,14 +23,30 @@ interface CardFooterProps {
 const CardFooter: React.FC<CardFooterProps> = ({
   voice,
   isPlaying,
+  isLoading = false,
   currentTime,
+  playDuration,
   onPlay,
   onPause,
+  onSeek,
   onUse,
   onRetry,
 }) => {
   const isFailed = voice.status === "failed"
   const isProcessing = voice.status === "processing"
+  // 播放中以 audio 真实时长为准
+  const effectiveDuration = isPlaying ? playDuration || voice.duration : voice.duration
+
+  /** 播放中点击进度条 seek；非播放态点击触发播放 */
+  const handleProgressClick = (e: React.MouseEvent<HTMLDivElement>) => {
+    const rect = e.currentTarget.getBoundingClientRect()
+    const percent = Math.max(0, Math.min(1, (e.clientX - rect.left) / rect.width))
+    if (isPlaying && effectiveDuration > 0) {
+      onSeek?.(percent * effectiveDuration)
+    } else if (!isLoading) {
+      onPlay()
+    }
+  }
 
   return (
     <div className="xx-clone-footer">
@@ -31,21 +55,30 @@ const CardFooter: React.FC<CardFooterProps> = ({
           <button
             type="button"
             className="xx-clone-play-btn"
+            disabled={isLoading}
             onClick={(e) => {
               e.stopPropagation()
               isPlaying ? onPause() : onPlay()
             }}
             title={isPlaying ? "暂停" : "试听"}
+            aria-label={isPlaying ? "暂停" : "试听"}
           >
-            {isPlaying ? <PauseCircleOutlined /> : <PlayCircleOutlined />}
+            {isLoading ? (
+              <LoadingOutlined spin />
+            ) : isPlaying ? (
+              <PauseCircleOutlined />
+            ) : (
+              <PlayCircleOutlined />
+            )}
           </button>
-          <div className="xx-clone-progress">
+          <div className="xx-clone-progress" onClick={handleProgressClick}>
             <div
               className="xx-clone-progress-bar"
               style={{
-                width: isPlaying
-                  ? `${Math.min((currentTime / Math.max(voice.duration, 1)) * 100, 100)}%`
-                  : "0%",
+                width:
+                  isPlaying && effectiveDuration > 0
+                    ? `${Math.min((currentTime / effectiveDuration) * 100, 100)}%`
+                    : "0%",
               }}
             />
           </div>

apps/web/src/pages/voices/components/tts-modal/VoiceSelector.tsx

@@ -2,14 +2,28 @@ import React from "react"
 import { type PresetVoiceDisplay } from "@/pages/voices/types"
 import { genderLabel } from "@/pages/voices/utils/format"
 
+/** 克隆音色下拉选项（最小结构，新旧页面各自映射） */
+export interface TtsClonedVoiceOption {
+  /** 克隆音色 profile id（合成时直接作为 voice_id 传后端解析） */
+  id: string
+  name: string
+}
+
 interface VoiceSelectorProps {
   value: string
   onChange: (voiceId: string) => void
   presetVoices: PresetVoiceDisplay[]
+  /** 可用克隆音色（仅克隆完成/ready），为空时不显示「我的克隆音色」分组 */
+  clonedVoices?: TtsClonedVoiceOption[]
 }
 
-/** 音色选择下拉 */
-const VoiceSelector: React.FC<VoiceSelectorProps> = ({ value, onChange, presetVoices }) => {
+/** 音色选择下拉：预置音色 + 我的克隆音色 */
+const VoiceSelector: React.FC<VoiceSelectorProps> = ({
+  value,
+  onChange,
+  presetVoices,
+  clonedVoices = [],
+}) => {
   return (
     <div>
       <div
@@ -36,11 +50,22 @@ const VoiceSelector: React.FC<VoiceSelectorProps> = ({ value, onChange, presetVo
         }}
       >
         <option value="">默认音色</option>
-        {presetVoices.map((v) => (
-          <option key={v.id} value={v.id}>
-            {v.name} — {genderLabel(v.gender)}
-          </option>
-        ))}
+        <optgroup label="预置音色">
+          {presetVoices.map((v) => (
+            <option key={v.id} value={v.id}>
+              {v.name} — {genderLabel(v.gender)}
+            </option>
+          ))}
+        </optgroup>
+        {clonedVoices.length > 0 && (
+          <optgroup label="我的克隆音色">
+            {clonedVoices.map((v) => (
+              <option key={v.id} value={v.id}>
+                {v.name}
+              </option>
+            ))}
+          </optgroup>
+        )}
       </select>
     </div>
   )

apps/web/src/pages/voices/components/tts-modal/types.ts

@@ -1,4 +1,5 @@
 import { type PresetVoiceDisplay } from "@/pages/voices/types"
+import type { TtsClonedVoiceOption } from "./VoiceSelector"
 
 export type TtsStatus = "idle" | "synthesizing" | "done" | "error"
 
@@ -11,6 +12,8 @@ export interface TtsModalProps {
   ttsAudioUrl: string | null
   ttsError: string | null
   presetVoices: PresetVoiceDisplay[]
+  /** 可用克隆音色（仅 ready），为空时下拉不显示该分组 */
+  clonedVoices?: TtsClonedVoiceOption[]
   onClose: () => void
   onTextChange: (text: string) => void
   onVoiceChange: (voiceId: string) => void

apps/web/src/pages/voices/hooks/useAudioPlayer.ts

@@ -1,102 +1,200 @@
-import { useState, useRef, useCallback, useEffect } from "react"
-
 /**
- * 音频播放控制 Hook
- * 封装当前播放状态、播放/暂停/跳转控制，使用 setInterval 模拟进度更新
- * （适用于预置音色/克隆音色卡片的播放按钮交互）
+ * 配音库真实音频播放控制 Hook（预置音色 / 克隆音色 / 配音素材三个 tab 共用）
+ *
+ * - 配音素材：直接播放 asset.file_url（用户上传的真实音频）
+ * - 预置/克隆音色：调 POST /tts/preview 合成本示例文案，拿到 audio_url 后真实播放，
+ *   合成结果按 voiceId 内存缓存，同一音色二次试听不重复合成
+ * - 全库同一时刻只有一个 Audio 在响：切卡片 / 切 tab / 离开页面自动停止
+ * - timeupdate 驱动进度条，loadedmetadata 取真实时长，ended 自动复位
+ *
+ * 注意：合成失败的错误提示由 apiClient 拦截器统一 toast（含后端
+ * 「音色克隆尚未完成，请稍后再试」文案），hook 内不重复提示。
  */
+import { useState, useRef, useCallback, useEffect } from "react"
+import { previewTts } from "@/api/tts"
+import type { VoiceMaterial } from "@/pages/voice-materials/types"
+
+/** 卡片试听统一示例文案 */
+export const VOICE_PREVIEW_TEXT = "你好呀，欢迎使用小虾智剪，这是我的配音效果，希望你喜欢。"
+
+interface PreviewCacheEntry {
+  url: string
+  duration?: number
+}
+
 export function useAudioPlayer() {
   const [playingId, setPlayingId] = useState<string | null>(null)
   const [currentTime, setCurrentTime] = useState(0)
-  const intervalRef = useRef<number | null>(null)
+  const [duration, setDuration] = useState(0)
+  const [loadingId, setLoadingId] = useState<string | null>(null)
 
-  /** 开始播放指定音色（从 startTime 开始，默认从 0 开始） */
-  const handlePlay = useCallback(
-    (voiceId: string, duration: number, startTime: number = 0) => {
-      if (playingId === voiceId) return
-      if (intervalRef.current) {
-        clearInterval(intervalRef.current)
-      }
-      setPlayingId(voiceId)
-      setCurrentTime(startTime)
-      intervalRef.current = window.setInterval(() => {
-        setCurrentTime((prev) => {
-          if (prev >= duration) {
-            if (intervalRef.current) {
-              clearInterval(intervalRef.current)
-              intervalRef.current = null
-            }
-            setPlayingId(null)
-            return 0
-          }
-          return prev + 0.1
-        })
-      }, 100)
-    },
-    [playingId],
-  )
+  const audioRef = useRef<HTMLAudioElement | null>(null)
+  const pausedRef = useRef<{ id: string; url: string; duration: number } | null>(null)
+  const previewCacheRef = useRef<Map<string, PreviewCacheEntry>>(new Map())
+  /** 试听合成请求序号：旧请求返回时丢弃，防止竞态 */
+  const reqSeqRef = useRef(0)
 
-  /** 暂停播放 */
-  const handlePause = useCallback(() => {
-    if (intervalRef.current) {
-      clearInterval(intervalRef.current)
-      intervalRef.current = null
+  /** 停止当前播放并复位状态 */
+  const stopPlayback = useCallback(() => {
+    if (audioRef.current) {
+      audioRef.current.pause()
+      audioRef.current = null
     }
+    pausedRef.current = null
     setPlayingId(null)
+    setLoadingId(null)
+    setCurrentTime(0)
+    setDuration(0)
   }, [])
 
-  /** 跳转到指定时间 */
-  const handleSeek = useCallback(
-    (voiceId: string, time: number, duration: number) => {
-      if (playingId !== voiceId) {
-        // 不同音色：从指定时间开始播放
-        handlePlay(voiceId, duration, time)
-      } else {
-        // 同一音色：直接跳转
-        setCurrentTime(time)
+  /** 用指定 URL 创建 Audio 并播放 */
+  const startAudio = useCallback((id: string, url: string, knownDuration?: number) => {
+    if (audioRef.current) {
+      audioRef.current.pause()
+      audioRef.current = null
+    }
+    const audio = new Audio(url)
+    audioRef.current = audio
+
+    audio.addEventListener("timeupdate", () => {
+      setCurrentTime(audio.currentTime)
+    })
+    audio.addEventListener("loadedmetadata", () => {
+      if (Number.isFinite(audio.duration) && audio.duration > 0) {
+        setDuration(audio.duration)
       }
+    })
+    audio.addEventListener("ended", () => {
+      if (audioRef.current === audio) audioRef.current = null
+      pausedRef.current = null
+      setPlayingId(null)
+      setCurrentTime(0)
+    })
+
+    if (knownDuration && knownDuration > 0) setDuration(knownDuration)
+    setCurrentTime(0)
+    setPlayingId(id)
+    pausedRef.current = null
+
+    audio.play().catch(() => {
+      // 自动播放被拦截或 URL 失效：复位按钮，错误提示由拦截器/环境处理
+      if (audioRef.current === audio) audioRef.current = null
+      setPlayingId(null)
+      setLoadingId((cur) => (cur === id ? null : cur))
+    })
+  }, [])
+
+  /** 播放配音素材（file_url 直链） */
+  const playMaterial = useCallback(
+    (material: VoiceMaterial) => {
+      if (!material.fileUrl) return
+      startAudio(material.id, material.fileUrl, material.duration)
     },
-    [playingId, handlePlay],
+    [startAudio],
   )
 
-  /** 切换播放/暂停 */
-  const handleTogglePlay = useCallback(
-    (voiceId: string, duration: number) => {
+  /** 预置/克隆音色试听：先 TTS 合成（带缓存），再真实播放 */
+  const playVoice = useCallback(
+    async (voiceId: string) => {
+      // 暂停中恢复
+      if (pausedRef.current?.id === voiceId && audioRef.current) {
+        try {
+          await audioRef.current.play()
+        } catch {
+          return
+        }
+        setPlayingId(voiceId)
+        pausedRef.current = null
+        return
+      }
+
+      const cached = previewCacheRef.current.get(voiceId)
+      if (cached) {
+        startAudio(voiceId, cached.url, cached.duration)
+        return
+      }
+
+      const seq = ++reqSeqRef.current
+      setLoadingId(voiceId)
+      try {
+        const res = await previewTts({ text: VOICE_PREVIEW_TEXT, voice_id: voiceId, speed: 1.0 })
+        if (seq !== reqSeqRef.current) return // 已被更新的请求取代
+        previewCacheRef.current.set(voiceId, { url: res.audio_url, duration: res.duration })
+        setLoadingId(null)
+        startAudio(voiceId, res.audio_url, res.duration)
+      } catch {
+        if (seq !== reqSeqRef.current) return
+        // 错误文案（含「克隆尚未完成」）由 apiClient 拦截器统一 toast
+        setLoadingId(null)
+      }
+    },
+    [startAudio],
+  )
+
+  /** 暂停（记录暂停对象，供再次点击恢复） */
+  const handlePause = useCallback(() => {
+    const audio = audioRef.current
+    if (!audio) return
+    audio.pause()
+    pausedRef.current = { id: playingId ?? "", url: audio.src, duration }
+    setPlayingId(null)
+  }, [playingId, duration])
+
+  /** 音色卡片播放/暂停切换 */
+  const handleToggleVoice = useCallback(
+    (voiceId: string) => {
       if (playingId === voiceId) {
         handlePause()
       } else {
-        handlePlay(voiceId, duration)
+        stopPlayback()
+        void playVoice(voiceId)
       }
     },
-    [playingId, handlePlay, handlePause],
+    [playingId, handlePause, stopPlayback, playVoice],
   )
 
-  /** 停止所有播放（切换 Tab 时调用） */
-  const stopPlayback = useCallback(() => {
-    if (intervalRef.current) {
-      clearInterval(intervalRef.current)
-      intervalRef.current = null
-    }
-    setPlayingId(null)
-    setCurrentTime(0)
+  /** 素材卡片播放/暂停切换 */
+  const handleToggleMaterial = useCallback(
+    (material: VoiceMaterial) => {
+      if (playingId === material.id) {
+        handlePause()
+      } else {
+        stopPlayback()
+        playMaterial(material)
+      }
+    },
+    [playingId, handlePause, stopPlayback, playMaterial],
+  )
+
+  /** 进度条 seek（仅播放中有效；非播放态点击进度条由卡片改为触发播放） */
+  const handleSeek = useCallback((time: number) => {
+    const audio = audioRef.current
+    if (!audio) return
+    audio.currentTime = time
+    setCurrentTime(time)
   }, [])
 
-  // 组件卸载时清理
+  // 组件卸载时清理音频
   useEffect(() => {
     return () => {
-      if (intervalRef.current) {
-        clearInterval(intervalRef.current)
+      if (audioRef.current) {
+        audioRef.current.pause()
+        audioRef.current = null
       }
     }
   }, [])
 
   return {
     playingId,
+    loadingId,
     currentTime,
-    handlePlay,
+    duration,
+    playMaterial,
+    playVoice,
     handlePause,
+    handleToggleVoice,
+    handleToggleMaterial,
     handleSeek,
-    handleTogglePlay,
     stopPlayback,
   }
 }

apps/web/src/pages/voices/hooks/useMaterialBatchDelete.ts

@@ -0,0 +1,97 @@
+/**
+ * 配音库「配音素材」Tab 批量删除逻辑
+ *
+ * 选择状态本地维护；批量删除循环调 deleteAsset（无批量接口），
+ * 单个失败不中断；删除项含正在播放的素材时停止播放。
+ */
+import { useState, useCallback, useMemo } from "react"
+import { useQueryClient } from "@tanstack/react-query"
+import { Modal } from "@/components/ui"
+import { deleteAsset, type AssetItem } from "@/api/assets"
+import type { Toast } from "../components/VoiceToasts"
+
+interface UseMaterialBatchDeleteOptions {
+  materials: AssetItem[]
+  /** 停止播放回调（删除正在播放的素材时调用） */
+  stopPlayback: () => void
+  showToast: (message: string, type: Toast["type"]) => void
+}
+
+export function useMaterialBatchDelete({
+  materials,
+  stopPlayback,
+  showToast,
+}: UseMaterialBatchDeleteOptions) {
+  const queryClient = useQueryClient()
+  const [selectedIds, setSelectedIds] = useState<Set<string>>(new Set())
+  const [batchDeleting, setBatchDeleting] = useState(false)
+
+  /** 只统计当前列表中仍然存在的选中项（删除后自动收敛） */
+  const validSelected = useMemo(
+    () => materials.filter((m) => selectedIds.has(m.id)),
+    [materials, selectedIds],
+  )
+  const selectedCount = validSelected.length
+  const allSelected = materials.length > 0 && selectedCount === materials.length
+
+  const toggleSelect = useCallback((id: string) => {
+    setSelectedIds((prev) => {
+      const next = new Set(prev)
+      if (next.has(id)) next.delete(id)
+      else next.add(id)
+      return next
+    })
+  }, [])
+
+  const toggleSelectAll = useCallback(() => {
+    setSelectedIds(allSelected ? new Set() : new Set(materials.map((m) => m.id)))
+  }, [allSelected, materials])
+
+  const clearSelection = useCallback(() => setSelectedIds(new Set()), [])
+
+  const handleBatchDelete = useCallback(() => {
+    const targets = validSelected
+    if (targets.length === 0) return
+    Modal.confirm({
+      title: "确认批量删除",
+      content: `确定删除选中的 ${targets.length} 个素材？删除后不可恢复。`,
+      okText: "删除",
+      okButtonProps: { danger: true },
+      cancelText: "取消",
+      onOk: async () => {
+        setBatchDeleting(true)
+        let successCount = 0
+        for (const asset of targets) {
+          try {
+            await deleteAsset(asset.id)
+            successCount++
+          } catch {
+            /* 单个失败不中断 */
+          }
+        }
+        // 删除项含正在播放的素材（播放 id 与素材 id 一致）→ 停止播放
+        stopPlayback()
+        queryClient.invalidateQueries({ queryKey: ["voice-materials"] })
+        queryClient.invalidateQueries({ queryKey: ["assets", "voice"] })
+        setSelectedIds(new Set())
+        setBatchDeleting(false)
+        if (successCount === targets.length) {
+          showToast(`已批量删除 ${successCount} 个素材`, "success")
+        } else {
+          showToast(`已批量删除 ${successCount}/${targets.length} 个素材，部分失败`, "error")
+        }
+      },
+    })
+  }, [validSelected, stopPlayback, queryClient, showToast])
+
+  return {
+    selectedIds,
+    selectedCount,
+    allSelected,
+    batchDeleting,
+    toggleSelect,
+    toggleSelectAll,
+    clearSelection,
+    handleBatchDelete,
+  }
+}

apps/web/src/pages/voices/hooks/useMaterialDelete.ts

@@ -0,0 +1,56 @@
+/**
+ * 配音库「配音素材」Tab 单个删除逻辑
+ */
+import { useCallback } from "react"
+import { useMutation, useQueryClient } from "@tanstack/react-query"
+import { Modal } from "@/components/ui"
+import { deleteAsset, type AssetItem } from "@/api/assets"
+import type { Toast } from "../components/VoiceToasts"
+
+interface UseMaterialDeleteOptions {
+  materials: AssetItem[]
+  /** 停止播放回调（删除正在播放的素材时调用） */
+  stopPlayback: () => void
+  showToast: (message: string, type: Toast["type"]) => void
+}
+
+export function useMaterialDelete({
+  materials,
+  stopPlayback,
+  showToast,
+}: UseMaterialDeleteOptions) {
+  const queryClient = useQueryClient()
+
+  const deleteMutation = useMutation({
+    mutationFn: (assetId: string) => deleteAsset(assetId),
+    onSuccess: () => {
+      queryClient.invalidateQueries({ queryKey: ["voice-materials"] })
+      queryClient.invalidateQueries({ queryKey: ["assets", "voice"] })
+      // 删除成功后停止播放（若删的是正在播放的素材，播放 id 与素材 id 一致）
+      stopPlayback()
+      showToast("素材已删除", "success")
+    },
+    onError: (err: unknown) => {
+      const msg = err instanceof Error ? err.message : "未知错误"
+      showToast(`删除失败：${msg}`, "error")
+    },
+  })
+
+  const handleMaterialDelete = useCallback(
+    (asset: AssetItem) => {
+      const material = materials.find((m) => m.id === asset.id)
+      if (!material) return
+      Modal.confirm({
+        title: "确认删除",
+        content: `确定删除配音素材「${material.name}」吗？删除后不可恢复。`,
+        okText: "删除",
+        okButtonProps: { danger: true },
+        cancelText: "取消",
+        onOk: () => deleteMutation.mutate(material.id),
+      })
+    },
+    [materials, deleteMutation],
+  )
+
+  return { handleMaterialDelete, isDeleting: deleteMutation.isPending }
+}

apps/web/src/pages/voices/hooks/useTtsSynthesize.ts

@@ -3,6 +3,7 @@ import { useQueryClient } from "@tanstack/react-query"
 import { message } from "antd"
 import { synthesizeSpeech, getTTSJobStatus, saveTtsToLibrary } from "@/api/tts"
 import { type PresetVoiceDisplay } from "../types"
+import type { TtsClonedVoiceOption } from "../components/tts-modal/VoiceSelector"
 
 export type TtsStatus = "idle" | "synthesizing" | "done" | "error"
 
@@ -12,10 +13,16 @@ export type TtsStatus = "idle" | "synthesizing" | "done" | "error"
  */
 interface UseTtsSynthesizeProps {
   presetVoices: PresetVoiceDisplay[]
+  /** 可用克隆音色（仅 ready；合成时 voice_id 直接传克隆 profile UUID，后端解析） */
+  clonedVoices?: TtsClonedVoiceOption[]
   showToast: (message: string, type: "success" | "error") => void
 }
 
-export function useTtsSynthesize({ presetVoices, showToast }: UseTtsSynthesizeProps) {
+export function useTtsSynthesize({
+  presetVoices,
+  clonedVoices = [],
+  showToast,
+}: UseTtsSynthesizeProps) {
   const queryClient = useQueryClient()
 
   const [ttsOpen, setTtsOpen] = useState(false)
@@ -131,6 +138,7 @@ export function useTtsSynthesize({ presetVoices, showToast }: UseTtsSynthesizePr
     ttsError,
     // 可选音色列表
     ttsPresetVoices: presetVoices,
+    ttsClonedVoices: clonedVoices,
     // Setters
     setTtsText,
     setTtsVoiceId,

apps/web/src/pages/voices/hooks/useVoiceUpload.ts

@@ -32,24 +32,39 @@ export function useVoiceUpload({ showToast }: UseVoiceUploadProps) {
         if (!lib) throw new Error("配音库不存在，请先在配音库页面创建")
 
         /* 直传文件（后端会自动创建 ingest job） */
-        const { ingest_job_id } = await uploadAssetDirect({
+        const complete = await uploadAssetDirect({
           file: data.file,
           library_id: lib.id,
           onProgress: (p) => setUploadProgress(p),
         })
 
-        /* 轮询 ingest job 状态，等待 Worker 处理完成 */
-        let jobStatus = ""
+        /* 去重命中（同库已存在相同 file_hash 素材）：
+         * 后端返回 duplicated=true，ingest_job_id 为空，
+         * 不轮询、直接按上传成功处理（onSuccess 分支 toast + 刷新列表）。
+         * 注意：mutationFn 正常 return 即视为 resolve，useMutation 会自动调 onSuccess。
+         */
+        if (complete.duplicated === true) {
+          return
+        }
+        if (!complete.ingest_job_id) {
+          throw new Error("上传完成但未返回处理任务 ID，请重试")
+        }
+        const { ingest_job_id } = complete
+
+        /* 轮询 ingest job 状态，等待 Worker 处理完成；
+         * complete 后先立即查一次（后端通常不到 1s 处理完），未完成再每 5s 轮询。
+         * 成功状态为 IngestJobStatus.completed；"ready" 是 voice-clones 的状态，此处误用需避免。
+         */
+        let job = await getIngestJob(ingest_job_id)
         let retries = 0
         const maxRetries = 60 // 最多等待 5 分钟（60 * 5秒）
-        while (jobStatus !== "ready" && jobStatus !== "failed" && retries < maxRetries) {
+        while (job.status !== "completed" && job.status !== "failed" && retries < maxRetries) {
           await new Promise((r) => setTimeout(r, 5000))
-          const job = await getIngestJob(ingest_job_id)
-          jobStatus = job.status
+          job = await getIngestJob(ingest_job_id)
           retries++
         }
 
-        if (jobStatus === "failed") {
+        if (job.status === "failed") {
           throw new Error("音频处理失败，请重试")
         }
         if (retries >= maxRetries) {

apps/web/src/pages/voices/voices.css

@@ -972,76 +972,173 @@
 }
 
 /* ================================================================
-   配音素材卡片（与配音库Tab集成）
+   配音素材卡片（复用预置音色卡片 .xx-voice-card 布局与播放控件）
    ================================================================ */
 
-.vmat-card {
+/* 圆形图标：紫色系（素材统一配色，不用预置音色的性别色） */
+.vmat-card .vmat-avatar {
+  background: linear-gradient(135deg, #6366f1 0%, #8b5cf6 100%);
+}
+
+/* 左上角批量选择 checkbox */
+.vmat-card-checkbox {
+  position: absolute;
+  top: 10px;
+  left: 10px;
+  z-index: 3;
+  width: 20px;
+  height: 20px;
+  border-radius: var(--radius-sm);
+  border: 1.5px solid var(--border-color);
   background: var(--bg-primary);
-  border-radius: var(--radius-lg);
-  overflow: hidden;
+  color: #fff;
+  display: grid;
+  place-items: center;
+  font-size: 11px;
+  line-height: 1;
+  cursor: pointer;
+  opacity: 0;
+  transition: var(--transition-fast);
+}
+
+.vmat-card:hover .vmat-card-checkbox,
+.vmat-card-checkbox.checked {
+  opacity: 1;
+}
+
+.vmat-card-checkbox.checked {
+  background: var(--primary-color);
+  border-color: var(--primary-color);
+}
+
+/* 触摸屏无 hover：checkbox 常驻显示 */
+@media (hover: none) {
+  .vmat-card-checkbox {
+    opacity: 1;
+  }
+}
+
+/* 右上角删除按钮：常驻可见，hover 加深为 danger 色 */
+.vmat-card-delete {
+  position: absolute;
+  top: 8px;
+  right: 8px;
+  z-index: 3;
+  width: 28px;
+  height: 28px;
+  border: none;
+  border-radius: var(--radius-sm);
+  background: transparent;
+  color: var(--text-tertiary);
+  cursor: pointer;
+  font-size: 14px;
+  display: grid;
+  place-items: center;
+  transition: var(--transition-fast);
+}
+
+.vmat-card-delete:hover {
+  background: var(--error-50);
+  color: var(--error-600);
+}
+
+/* 选中态高亮（覆盖预置卡片的选中边框） */
+.vmat-card.selected {
+  border-color: var(--primary-color);
+  box-shadow: 0 0 0 1px var(--primary-color);
+}
+
+/* 信息区给 checkbox / 删除按钮留位 */
+.vmat-card .vmat-info {
+  padding-right: 24px;
+}
+
+/* 播放波形/控件继承 .xx-voice-card 网格，无需额外样式 */
+
+/* 无可播放音频时播放按钮置灰 */
+.vmat-card .xx-voice-play-btn:disabled {
+  background: var(--bg-tertiary);
+  color: var(--text-tertiary);
+  cursor: not-allowed;
+  transform: none;
+  box-shadow: none;
+}
+
+/* ── 批量操作栏 ─────────────────────────────────── */
+.vmat-batch-bar {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  padding: 8px 12px;
+  margin-bottom: var(--space-md);
+  background: var(--bg-secondary);
   border: 1px solid var(--border-color);
-  transition: all 0.2s;
+  border-radius: var(--radius-md);
+}
+
+.vmat-batch-bar-left {
+  display: flex;
+  align-items: center;
+  gap: var(--space-sm);
+}
+
+.vmat-batch-bar-right {
+  display: flex;
+  align-items: center;
+  gap: var(--space-sm);
+}
+
+.vmat-checkbox {
+  width: 18px;
+  height: 18px;
+  border-radius: var(--radius-sm);
+  border: 1.5px solid var(--border-color);
+  background: var(--bg-primary);
+  color: #fff;
+  display: grid;
+  place-items: center;
+  font-size: 10px;
+  line-height: 1;
+  cursor: pointer;
+  padding: 0;
+  transition: var(--transition-fast);
+}
+
+.vmat-checkbox.checked {
+  background: var(--primary-color);
+  border-color: var(--primary-color);
+}
+
+.vmat-select-all {
+  background: none;
+  border: none;
+  padding: 0;
+  font-size: var(--font-size-sm);
+  color: var(--text-secondary);
   cursor: pointer;
 }
 
-.vmat-card:hover {
-  transform: translateY(-2px);
-  box-shadow: 0 8px 20px rgba(0, 0, 0, 0.08);
-  border-color: var(--primary-300);
+.vmat-select-all:hover {
+  color: var(--primary-color);
 }
 
-.vmat-thumb {
-  position: relative;
-  aspect-ratio: 16 / 9;
-  background: linear-gradient(135deg, #6366f1 0%, #8b5cf6 100%);
-  display: flex;
-  align-items: center;
-  justify-content: center;
+.vmat-batch-count {
+  font-size: var(--font-size-sm);
+  color: var(--primary-color);
+  font-weight: var(--font-weight-medium);
 }
 
-.vmat-thumb-icon {
-  font-size: 32px;
-  color: rgba(255, 255, 255, 0.9);
-}
-
-.vmat-duration {
-  position: absolute;
-  bottom: 8px;
-  right: 8px;
-  padding: 2px 8px;
-  background: rgba(0, 0, 0, 0.6);
-  color: #fff;
-  font-size: 11px;
-  border-radius: 4px;
-  font-family: monospace;
-}
-
-.vmat-info {
-  padding: 12px 14px;
-}
-
-.vmat-name {
-  font-size: 13px;
-  font-weight: 500;
-  color: var(--text-primary);
-  overflow: hidden;
-  text-overflow: ellipsis;
-  white-space: nowrap;
-  margin-bottom: 4px;
-}
-
-.vmat-meta {
-  font-size: 12px;
-  color: var(--text-tertiary);
-}
-
-/* 骨架屏 */
+/* ── 骨架屏 ─────────────────────────────────────── */
 .vmat-card--skeleton {
   pointer-events: none;
 }
 
-.vmat-card--skeleton .vmat-thumb {
+.vmat-skeleton-avatar {
+  width: 48px;
+  height: 48px;
+  border-radius: var(--radius-full);
   background: var(--bg-tertiary);
+  grid-row: 1 / 3;
 }
 
 .vmat-skeleton-line {

apps/web/src/test/api/asset-usage.test.ts

@@ -0,0 +1,24 @@
+import { describe, expect, it } from "vitest"
+import { isAssetUsable } from "@/api/assets"
+
+describe("isAssetUsable", () => {
+  it("usable 字段缺失时降级为可用（后端字段未上线零影响）", () => {
+    expect(isAssetUsable({})).toBe(true)
+    expect(isAssetUsable({ usable: undefined })).toBe(true)
+    expect(isAssetUsable({ usable: null })).toBe(true)
+  })
+
+  it("usable === true 时可用", () => {
+    expect(isAssetUsable({ usable: true })).toBe(true)
+  })
+
+  it("usable === false 时不可用（已彻底用尽）", () => {
+    expect(isAssetUsable({ usable: false })).toBe(false)
+    expect(isAssetUsable({ usable: false, used_ratio: 1 })).toBe(false)
+  })
+
+  it("used_ratio 不影响可用性判断（只影响角标展示）", () => {
+    expect(isAssetUsable({ used_ratio: 0.99 })).toBe(true)
+    expect(isAssetUsable({ used_ratio: 0 })).toBe(true)
+  })
+})

apps/web/src/test/components/voice/CloneModal.test.tsx

@@ -1,42 +1,82 @@
 import React from "react"
-// 重构：useCloneModal Hook 已拆分为 useCloneFormState + useCloneSubmit 子 Hook
 import { describe, expect, it, vi } from "vitest"
-import { render } from "@testing-library/react"
+import { render, screen } from "@testing-library/react"
 import CloneModal from "@/components/voice/CloneModal"
 
+vi.mock("react-router-dom", () => ({
+  useNavigate: () => vi.fn(),
+}))
+
 vi.mock("@/api/voice-clone", () => ({
   createVoiceClone: vi.fn(),
-  toVoiceClone: vi.fn(),
+  toVoiceClone: vi.fn((x) => x),
 }))
 
+const mockGetAssetsByKind = vi.fn()
 vi.mock("@/api/assets", () => ({
   uploadAssetDirect: vi
     .fn()
     .mockResolvedValue({ storage_key: "test", ingest_job_id: "test", url: "http://test" }),
+  ensureDefaultLibrary: vi.fn().mockResolvedValue({ id: "lib-1" }),
+  getAssetsByKind: (...args: unknown[]) => mockGetAssetsByKind(...args),
 }))
 
+vi.mock("@/api/projects", () => ({
+  getOrCreateDefaultProject: vi.fn().mockResolvedValue({ id: "proj-1" }),
+}))
+
+vi.mock("@tanstack/react-query", () => ({
+  useQuery: ({ enabled, queryFn }: { enabled: boolean; queryFn: () => unknown }) => {
+    // enabled=false 时不发请求（模拟弹窗关闭）
+    if (!enabled) return { data: undefined, isLoading: false }
+    return { data: mockQueryData, isLoading: mockLoading }
+  },
+}))
+
+let mockQueryData: unknown = undefined
+let mockLoading = false
+
 vi.mock("@/components/ui", () => ({
-  Modal: ({ open, children, onCancel, onOk, title }: any) =>
+  Modal: ({ open, children, onCancel, title }: any) =>
     open ? React.createElement("div", { role: "dialog", "data-title": title }, children) : null,
   Button: ({ children, onClick, disabled, buttonType }: any) =>
     React.createElement("button", { onClick, disabled, "data-type": buttonType }, children),
 }))
 
 describe("CloneModal", () => {
-  it("should render when closed", () => {
+  it("关闭时不渲染", () => {
+    mockQueryData = undefined
     const { container } = render(<CloneModal open={false} onClose={vi.fn()} />)
-    expect(container).toBeTruthy()
+    expect(container.querySelector('[role="dialog"]')).toBeNull()
   })
 
-  it("should render input phase when open", () => {
-    const { container } = render(<CloneModal open={true} onClose={vi.fn()} />)
-    expect(container).toBeTruthy()
+  it("打开时显示「从配音素材选择」和「直接录制」，不再有文件上传入口", () => {
+    mockQueryData = []
+    mockLoading = false
+    render(<CloneModal open={true} onClose={vi.fn()} />)
+    expect(screen.getByText("从配音素材选择")).toBeTruthy()
+    expect(screen.getByText("直接录制")).toBeTruthy()
+    // 文件上传入口已删除
+    expect(screen.queryByText(/拖拽音频文件/)).toBeNull()
+    expect(screen.queryByText("上传音频")).toBeNull()
   })
 
-  it("should call onClose when cancel", () => {
-    const onClose = vi.fn()
-    render(<CloneModal open={true} onClose={onClose} />)
-    // just verify render doesn't crash
-    expect(onClose).toBeDefined()
+  it("素材为空时提示先上传素材并给跳转入口", () => {
+    mockQueryData = []
+    render(<CloneModal open={true} onClose={vi.fn()} />)
+    expect(screen.getByText("请先在配音库上传素材")).toBeTruthy()
+    expect(screen.getByText("去配音库上传")).toBeTruthy()
+  })
+
+  it("有素材时下拉展示素材名和时长", () => {
+    mockQueryData = [
+      { id: "a1", name: "旁白录音.m4a", duration: 65 },
+      { id: "a2", name: "访谈.mp3", duration: undefined },
+    ]
+    render(<CloneModal open={true} onClose={vi.fn()} />)
+    expect(screen.getByText("旁白录音.m4a（01:05）")).toBeTruthy()
+    expect(screen.getByText("访谈.mp3（--:--）")).toBeTruthy()
+    // 空态提示不出现
+    expect(screen.queryByText("请先在配音库上传素材")).toBeNull()
   })
 })

apps/web/src/test/components/voice/CloneModal/utils.test.ts

@@ -1,11 +0,0 @@
-/**
- * Smoke test for utils
- */
-import { describe, it, expect } from "vitest"
-import "@/components/voice/CloneModal/utils"
-
-describe("utils smoke", () => {
-  it("should load module successfully", () => {
-    expect(true).toBe(true)
-  })
-})

apps/web/src/test/pages/assets/smoke.test.tsx

@@ -20,7 +20,7 @@ import "@/pages/assets/components/CreateLibraryModal"
 import "@/pages/assets/components/LibrarySidebar"
 import "@/pages/assets/components/PlayModal"
 import "@/pages/assets/components/ResultDrawer"
-import "@/pages/assets/components/UploadProgressModal"
+import "@/pages/assets/components/UploadQueuePanel"
 
 // 类型与常量
 import "@/pages/assets/types"

apps/web/src/test/pages/assets/usage-badge.test.ts

@@ -0,0 +1,53 @@
+import { describe, expect, it } from "vitest"
+import { getUsageBadge } from "@/pages/assets/types"
+
+describe("getUsageBadge", () => {
+  it("非视频素材不显示角标", () => {
+    expect(getUsageBadge({ kind: "voice", usable: false })).toBeNull()
+    expect(getUsageBadge({ kind: "image", usable: false })).toBeNull()
+  })
+
+  it("字段缺失时不显示角标（降级零影响）", () => {
+    expect(getUsageBadge({ kind: "video" })).toBeNull()
+    expect(getUsageBadge({ kind: "video", usable: true, usedRatio: undefined })).toBeNull()
+  })
+
+  it("usable === false 显示红色实心「已用尽」", () => {
+    expect(getUsageBadge({ kind: "video", usable: false, usedRatio: 1 })).toEqual({
+      label: "已用尽",
+      variant: "exhausted",
+    })
+    // usable === false 优先级最高，即使 usedRatio 字段缺失
+    expect(getUsageBadge({ kind: "video", usable: false })).toEqual({
+      label: "已用尽",
+      variant: "exhausted",
+    })
+  })
+
+  it("used_ratio >= 0.85 显示红色「即将用尽」", () => {
+    expect(getUsageBadge({ kind: "video", usable: true, usedRatio: 0.85 })).toEqual({
+      label: "即将用尽",
+      variant: "warning",
+    })
+    expect(getUsageBadge({ kind: "video", usable: true, usedRatio: 0.97 })).toEqual({
+      label: "即将用尽",
+      variant: "warning",
+    })
+  })
+
+  it("used_ratio >= 0.5 显示橙色「已用 xx%」", () => {
+    expect(getUsageBadge({ kind: "video", usable: true, usedRatio: 0.5 })).toEqual({
+      label: "已用 50%",
+      variant: "ratio",
+    })
+    expect(getUsageBadge({ kind: "video", usable: true, usedRatio: 0.84 })).toEqual({
+      label: "已用 84%",
+      variant: "ratio",
+    })
+  })
+
+  it("used_ratio < 0.5 不显示角标", () => {
+    expect(getUsageBadge({ kind: "video", usable: true, usedRatio: 0.49 })).toBeNull()
+    expect(getUsageBadge({ kind: "video", usable: true, usedRatio: 0 })).toBeNull()
+  })
+})

apps/web/src/test/pages/assets/useAssetUpload.test.tsx

@@ -0,0 +1,226 @@
+/**
+ * useAssetUpload 队列批量上传测试
+ * - 并发直传不超过 MAX_CONCURRENT(3)
+ * - prepare 返回 asset_id 后 invalidate 列表
+ * - 状态流转 uploading→ingesting→done；失败可重试；duplicated 命中
+ */
+import { describe, it, expect, vi, beforeEach } from "vitest"
+import { renderHook, waitFor, act } from "@testing-library/react"
+import { QueryClient, QueryClientProvider } from "@tanstack/react-query"
+import type { ReactNode } from "react"
+
+const invalidateQueries = vi.fn()
+vi.mock("@/api/assets", () => ({
+  prepareDirectUploadHandle: vi.fn(),
+}))
+
+const { prepareDirectUploadHandle } = await import("@/api/assets")
+const { useAssetUpload } = await import("@/pages/assets/hooks/useAssetUpload")
+
+interface FakeHandle {
+  prepared: {
+    upload_url: string
+    method: string
+    storage_key: string
+    expires_at: string
+    fields: Record<string, string>
+    max_size_bytes: number
+    asset_id: string
+  }
+  transfer: ReturnType<typeof vi.fn>
+  complete: ReturnType<typeof vi.fn>
+  /** 手动结束传输（transfer 被调用后挂载）；finish(true) 以失败结束 */
+  finish: (fail?: boolean) => void
+}
+
+let activeTransfers = 0
+let maxConcurrent = 0
+
+/**
+ * 创建一个假 handle：transfer 返回挂起的 promise，
+ * finish 槽位在 transfer executor 同步执行时挂载，测试中调用 finish() 控制成败
+ */
+const makeFakeHandle = (opts: { id: string; duplicated?: boolean; failTransfer?: boolean }) => {
+  const h = {
+    prepared: {
+      upload_url: "https://oss.example.com/u",
+      method: "POST",
+      storage_key: `uploads/${opts.id}/y.mp4`,
+      expires_at: "2099-01-01",
+      fields: {},
+      max_size_bytes: 2_000_000_000,
+      asset_id: opts.id,
+    },
+    transfer: vi.fn(),
+    complete: vi.fn().mockResolvedValue({
+      storage_key: "uploads/x/y.mp4",
+      ingest_job_id: opts.duplicated ? "" : "job-1",
+      url: "https://oss.example.com/u",
+      duplicated: opts.duplicated,
+      asset_id: opts.id,
+    }),
+    finish: (() => {}) as (fail?: boolean) => void,
+  }
+  h.transfer.mockImplementation(
+    () =>
+      new Promise<void>((_resolve, reject) => {
+        activeTransfers += 1
+        maxConcurrent = Math.max(maxConcurrent, activeTransfers)
+        h.finish = (fail = false) => {
+          activeTransfers -= 1
+          if (fail || opts.failTransfer) reject(new Error("OSS boom"))
+          else _resolve()
+        }
+      }),
+  )
+  return h
+}
+
+type FakeHandleLike = ReturnType<typeof makeFakeHandle>
+
+/** prepare mock：调用序号生成稳定 id，立即把 handle（含 finish 槽位）推入数组 */
+const installPrepareMock = (
+  handles: FakeHandleLike[],
+  optOverrides?: (id: string) => { duplicated?: boolean; failTransfer?: boolean },
+) => {
+  let callNo = 0
+  ;(prepareDirectUploadHandle as unknown as ReturnType<typeof vi.fn>).mockImplementation(
+    async () => {
+      const id = `asset-${callNo++}`
+      const overrides = optOverrides?.(id) ?? {}
+      const h = makeFakeHandle({ id, ...overrides })
+      handles.push(h)
+      await new Promise((r) => setTimeout(r, 10))
+      return h
+    },
+  )
+}
+
+const createWrapper = () => {
+  const qc = new QueryClient({
+    defaultOptions: { queries: { retry: false }, mutations: { retry: false } },
+  })
+  // 监听 invalidate 调用
+  const orig = qc.invalidateQueries.bind(qc)
+  qc.invalidateQueries = ((...args: unknown[]) => {
+    invalidateQueries()
+    return orig(...(args as never))
+  }) as never
+  return ({ children }: { children: ReactNode }) => (
+    <QueryClientProvider client={qc}>{children}</QueryClientProvider>
+  )
+}
+
+const mp4 = (name: string) => new File([new Uint8Array(10)], name, { type: "video/mp4" })
+
+beforeEach(() => {
+  vi.clearAllMocks()
+  activeTransfers = 0
+  maxConcurrent = 0
+})
+
+describe("useAssetUpload", () => {
+  it("5 个文件批量入队：同时直传不超过 3 个，全部完成且刷新列表", async () => {
+    const handles: FakeHandleLike[] = []
+    installPrepareMock(handles)
+
+    const { result } = renderHook(() => useAssetUpload({ effectiveLibId: "lib-1" }), {
+      wrapper: createWrapper(),
+    })
+
+    await act(async () => {
+      result.current.enqueueUploads(Array.from({ length: 5 }, (_, i) => mp4(`v${i}.mp4`)))
+    })
+
+    // 3 个进入 uploading（transfer 被挂起），2 个排队
+    await waitFor(() => {
+      expect(handles.length).toBe(3)
+      expect(result.current.uploadItems.filter((it) => it.status === "uploading").length).toBe(3)
+      expect(result.current.uploadItems.filter((it) => it.status === "preparing").length).toBe(2)
+    })
+    expect(maxConcurrent).toBe(3)
+
+    // 完成前 3 个 → 队列拉起后 2 个
+    await act(async () => {
+      handles[0].finish()
+      handles[1].finish()
+      handles[2].finish()
+    })
+    await waitFor(() => {
+      expect(handles.length).toBe(5)
+      expect(result.current.uploadItems.filter((it) => it.status === "uploading").length).toBe(2)
+    })
+    expect(maxConcurrent).toBeLessThanOrEqual(3)
+    await waitFor(() => expect(handles[4].transfer).toHaveBeenCalled())
+
+    // 完成剩余 2 个
+    await act(async () => {
+      handles[3].finish()
+      handles[4].finish()
+    })
+    await waitFor(() => {
+      expect(result.current.uploadItems.filter((it) => it.status === "done").length).toBe(5)
+    })
+    expect(invalidateQueries).toHaveBeenCalled()
+  })
+
+  it("传输失败标记 error，重试后成功", async () => {
+    const handles: FakeHandleLike[] = []
+    let firstCall = true
+    installPrepareMock(handles, () => {
+      const fail = firstCall
+      firstCall = false
+      return { failTransfer: fail }
+    })
+
+    const { result } = renderHook(() => useAssetUpload({ effectiveLibId: "lib-1" }), {
+      wrapper: createWrapper(),
+    })
+
+    await act(async () => {
+      result.current.enqueueUploads([mp4("bad.mp4")])
+    })
+    await waitFor(() => expect(handles.length).toBe(1))
+    await waitFor(() => expect(handles[0].transfer).toHaveBeenCalled())
+    await act(async () => {
+      handles[0].finish()
+    })
+    await waitFor(() => expect(result.current.uploadItems[0].status).toBe("error"))
+
+    // 重试：重新 prepare（handles[1] 成功）
+    const tempId = result.current.uploadItems[0].tempId
+    await act(async () => {
+      result.current.retryUpload(tempId)
+    })
+    await waitFor(() => expect(handles.length).toBe(2))
+    await waitFor(() => expect(handles[1].transfer).toHaveBeenCalled())
+    await act(async () => {
+      handles[1].finish()
+    })
+    await waitFor(() => {
+      expect(result.current.uploadItems.find((it) => it.tempId === tempId)?.status).toBe("done")
+    })
+  })
+
+  it("complete 返回 duplicated 时标记去重完成", async () => {
+    const handles: FakeHandleLike[] = []
+    installPrepareMock(handles, () => ({ duplicated: true }))
+
+    const { result } = renderHook(() => useAssetUpload({ effectiveLibId: "lib-1" }), {
+      wrapper: createWrapper(),
+    })
+
+    await act(async () => {
+      result.current.enqueueUploads([mp4("dup.mp4")])
+    })
+    await waitFor(() => expect(handles.length).toBe(1))
+    await waitFor(() => expect(handles[0].transfer).toHaveBeenCalled())
+    await act(async () => {
+      handles[0].finish()
+    })
+    await waitFor(() => {
+      expect(result.current.uploadItems[0].duplicated).toBe(true)
+      expect(result.current.uploadItems[0].status).toBe("done")
+    })
+  })
+})

apps/web/src/test/pages/generate/smoke.test.tsx

@@ -32,7 +32,6 @@ import "@/pages/generate/components/material/MaterialModeTabs"
 import "@/pages/generate/components/material/ManualMaterialList"
 import "@/pages/generate/components/material/SmartMatchInput"
 import "@/pages/generate/components/material/SmartMatchResults"
-import "@/pages/generate/components/material/SmartMatchCard"
 import "@/pages/generate/components/title/AiTitleGenerator"
 import "@/pages/generate/components/title/AiTitleCard"
 import "@/pages/generate/components/title/TitleStylePanel"

apps/web/src/test/pages/voices/hooks/useAudioPlayer.test.ts

@@ -1,232 +1,242 @@
 /**
- * useAudioPlayer hook 测试 — VoiceLibrary 版本
+ * useAudioPlayer hook 测试 — VoiceLibrary 版本（真实 Audio + TTS 试听）
  *
- * 该 Hook 使用 setInterval 模拟音频播放进度，纯逻辑可测。
- * 参考 voice-materials/hooks/useAudioPlayer.test.ts 的测试结构。
+ * mock HTMLAudioElement 与 previewTts，验证三 tab 共用的播放状态逻辑。
  */
 import { describe, it, expect, beforeEach, vi, afterEach } from "vitest"
 import { renderHook, act } from "@testing-library/react"
 import { useAudioPlayer } from "@/pages/voices/hooks/useAudioPlayer"
+import type { VoiceMaterial } from "@/pages/voice-materials/types"
 
-describe("useAudioPlayer (voices)", () => {
-  beforeEach(() => {
-    vi.useFakeTimers()
+const mockAudioPlay = vi.fn()
+const mockAudioPause = vi.fn()
+const listeners: Record<string, (() => void) | ((ev: unknown) => void)> = {}
+let mockAudioInstance: {
+  play: ReturnType<typeof vi.fn>
+  pause: ReturnType<typeof vi.fn>
+  addEventListener: (ev: string, cb: () => void) => void
+  currentTime: number
+  duration: number
+  src: string
+  volume: number
+  paused: boolean
+}
+
+const previewTtsMock = vi.fn()
+
+vi.mock("@/api/tts", () => ({
+  previewTts: (...args: unknown[]) => previewTtsMock(...args),
+}))
+
+const mockMaterial: VoiceMaterial = {
+  id: "asset-1",
+  name: "测试素材",
+  description: "",
+  gender: "neutral",
+  tagIds: [],
+  fileName: "test.mp3",
+  fileSize: 1024,
+  duration: 30,
+  mimeType: "audio/mpeg",
+  createdAt: "2024-01-01T00:00:00Z",
+  fileUrl: "https://example.com/test.mp3",
+}
+
+beforeEach(() => {
+  vi.clearAllMocks()
+  for (const k of Object.keys(listeners)) delete listeners[k]
+
+  mockAudioInstance = {
+    play: mockAudioPlay.mockResolvedValue(undefined),
+    pause: mockAudioPause,
+    addEventListener: (ev: string, cb: () => void) => {
+      listeners[ev] = cb
+    },
+    currentTime: 0,
+    duration: 30,
+    src: "",
+    volume: 1,
+    paused: true,
+  }
+
+  global.Audio = vi.fn().mockImplementation((url: string) => {
+    mockAudioInstance.src = url
+    return mockAudioInstance
+  }) as unknown as typeof Audio
+})
+
+afterEach(() => {
+  vi.clearAllMocks()
+})
+
+describe("useAudioPlayer (voices, 真实播放)", () => {
+  it("初始状态为空", () => {
+    const { result } = renderHook(() => useAudioPlayer())
+    expect(result.current.playingId).toBeNull()
+    expect(result.current.loadingId).toBeNull()
+    expect(result.current.currentTime).toBe(0)
   })
 
-  afterEach(() => {
-    vi.useRealTimers()
-    vi.clearAllMocks()
-  })
-
-  it("应该使用初始状态初始化", () => {
+  it("playMaterial 用 file_url 真实播放素材", () => {
     const { result } = renderHook(() => useAudioPlayer())
 
+    act(() => {
+      result.current.playMaterial(mockMaterial)
+    })
+
+    expect(result.current.playingId).toBe("asset-1")
+    expect(mockAudioPlay).toHaveBeenCalledTimes(1)
+    expect(mockAudioInstance.src).toBe("https://example.com/test.mp3")
+  })
+
+  it("file_url 缺失时不播放", () => {
+    const { result } = renderHook(() => useAudioPlayer())
+
+    act(() => {
+      result.current.playMaterial({ ...mockMaterial, fileUrl: undefined })
+    })
+
+    expect(result.current.playingId).toBeNull()
+    expect(mockAudioPlay).not.toHaveBeenCalled()
+  })
+
+  it("handleToggleMaterial 播放中再点为暂停", () => {
+    const { result } = renderHook(() => useAudioPlayer())
+
+    act(() => {
+      result.current.handleToggleMaterial(mockMaterial)
+    })
+    expect(result.current.playingId).toBe("asset-1")
+
+    act(() => {
+      result.current.handleToggleMaterial(mockMaterial)
+    })
+    expect(result.current.playingId).toBeNull()
+    expect(mockAudioPause).toHaveBeenCalled()
+  })
+
+  it("暂停后再次点击恢复播放", () => {
+    const { result } = renderHook(() => useAudioPlayer())
+
+    act(() => {
+      result.current.handleToggleMaterial(mockMaterial)
+    })
+    act(() => {
+      result.current.handleToggleMaterial(mockMaterial) // 暂停
+    })
+    mockAudioPlay.mockClear()
+    act(() => {
+      result.current.handleToggleMaterial(mockMaterial) // 恢复
+    })
+    expect(result.current.playingId).toBe("asset-1")
+    expect(mockAudioPlay).toHaveBeenCalledTimes(1)
+  })
+
+  it("playVoice 合成中显示 loading，成功后播放且二次点用缓存不重复合成", async () => {
+    previewTtsMock.mockResolvedValue({ audio_url: "https://example.com/tts.mp3", duration: 5 })
+    const { result } = renderHook(() => useAudioPlayer())
+
+    await act(async () => {
+      await result.current.playVoice("voice-1")
+    })
+
+    expect(previewTtsMock).toHaveBeenCalledWith(
+      expect.objectContaining({ voice_id: "voice-1", speed: 1.0 }),
+    )
+    expect(result.current.playingId).toBe("voice-1")
+    expect(mockAudioInstance.src).toBe("https://example.com/tts.mp3")
+
+    // 停止后再次试听同一音色 → 走缓存
+    act(() => {
+      result.current.stopPlayback()
+    })
+    mockAudioPlay.mockClear()
+    previewTtsMock.mockClear()
+
+    await act(async () => {
+      await result.current.playVoice("voice-1")
+    })
+    expect(previewTtsMock).not.toHaveBeenCalled()
+    expect(result.current.playingId).toBe("voice-1")
+    expect(mockAudioPlay).toHaveBeenCalledTimes(1)
+  })
+
+  it("合成失败时复位 loading 且不播放", async () => {
+    previewTtsMock.mockRejectedValue(new Error("克隆尚未完成"))
+    const { result } = renderHook(() => useAudioPlayer())
+
+    await act(async () => {
+      await result.current.playVoice("voice-x")
+    })
+
+    expect(result.current.loadingId).toBeNull()
+    expect(result.current.playingId).toBeNull()
+    expect(mockAudioPlay).not.toHaveBeenCalled()
+  })
+
+  it("互斥：播放新素材时停止上一个", () => {
+    const { result } = renderHook(() => useAudioPlayer())
+
+    act(() => {
+      result.current.playMaterial(mockMaterial)
+    })
+    act(() => {
+      result.current.playMaterial({
+        ...mockMaterial,
+        id: "asset-2",
+        fileUrl: "https://e.com/2.mp3",
+      })
+    })
+
+    expect(result.current.playingId).toBe("asset-2")
+    expect(mockAudioPause).toHaveBeenCalledTimes(1)
+  })
+
+  it("handleSeek 播放中调整 currentTime", () => {
+    const { result } = renderHook(() => useAudioPlayer())
+
+    act(() => {
+      result.current.playMaterial(mockMaterial)
+    })
+    act(() => {
+      result.current.handleSeek(12)
+    })
+
+    expect(mockAudioInstance.currentTime).toBe(12)
+  })
+
+  it("timeupdate 更新进度、ended 复位", () => {
+    const { result } = renderHook(() => useAudioPlayer())
+
+    act(() => {
+      result.current.playMaterial(mockMaterial)
+    })
+
+    act(() => {
+      mockAudioInstance.currentTime = 7
+      ;(listeners["timeupdate"] as () => void)()
+    })
+    expect(result.current.currentTime).toBe(7)
+
+    act(() => {
+      ;(listeners["ended"] as () => void)()
+    })
     expect(result.current.playingId).toBeNull()
     expect(result.current.currentTime).toBe(0)
   })
 
-  it("handlePlay 应该开始播放指定音色", () => {
+  it("stopPlayback 复位全部状态", () => {
     const { result } = renderHook(() => useAudioPlayer())
 
     act(() => {
-      result.current.handlePlay("voice-1", 10)
+      result.current.playMaterial(mockMaterial)
     })
-
-    expect(result.current.playingId).toBe("voice-1")
-    expect(result.current.currentTime).toBe(0)
-  })
-
-  it("handlePlay 对同一个音色不应重复启动播放", () => {
-    const { result } = renderHook(() => useAudioPlayer())
-
-    act(() => {
-      result.current.handlePlay("voice-1", 10)
-    })
-
-    const initialTime = result.current.currentTime
-
-    // 推进一些时间让进度走动
-    act(() => {
-      vi.advanceTimersByTime(200)
-    })
-
-    const timeAfterAdvance = result.current.currentTime
-    expect(timeAfterAdvance).toBeGreaterThan(initialTime)
-
-    // 对同一个音色再次调用 handlePlay 不应重置
-    act(() => {
-      result.current.handlePlay("voice-1", 10)
-    })
-
-    expect(result.current.playingId).toBe("voice-1")
-    expect(result.current.currentTime).toBe(timeAfterAdvance)
-  })
-
-  it("handlePlay 切换音色时应停止上一个并从头开始", () => {
-    const { result } = renderHook(() => useAudioPlayer())
-
-    act(() => {
-      result.current.handlePlay("voice-1", 10)
-    })
-
-    act(() => {
-      vi.advanceTimersByTime(500)
-    })
-
-    expect(result.current.playingId).toBe("voice-1")
-    expect(result.current.currentTime).toBeGreaterThan(0)
-
-    act(() => {
-      result.current.handlePlay("voice-2", 15)
-    })
-
-    expect(result.current.playingId).toBe("voice-2")
-    expect(result.current.currentTime).toBe(0)
-  })
-
-  it("播放进度应该随时间递增", () => {
-    const { result } = renderHook(() => useAudioPlayer())
-
-    act(() => {
-      result.current.handlePlay("voice-1", 10)
-    })
-
-    // 每 100ms 增加 0.1
-    act(() => {
-      vi.advanceTimersByTime(300)
-    })
-
-    expect(result.current.currentTime).toBeCloseTo(0.3, 1)
-    expect(result.current.playingId).toBe("voice-1")
-  })
-
-  it("播放到结尾应自动停止并重置", () => {
-    const { result } = renderHook(() => useAudioPlayer())
-
-    act(() => {
-      result.current.handlePlay("voice-1", 0.5) // 0.5 秒的短音频
-    })
-
-    act(() => {
-      vi.advanceTimersByTime(600) // 超过 0.5 秒
-    })
-
-    expect(result.current.playingId).toBeNull()
-    expect(result.current.currentTime).toBe(0)
-  })
-
-  it("handlePause 应该暂停播放", () => {
-    const { result } = renderHook(() => useAudioPlayer())
-
-    act(() => {
-      result.current.handlePlay("voice-1", 10)
-    })
-
-    act(() => {
-      vi.advanceTimersByTime(200)
-    })
-
-    const timeBeforePause = result.current.currentTime
-
-    act(() => {
-      result.current.handlePause()
-    })
-
-    expect(result.current.playingId).toBeNull()
-
-    // 暂停后时间不应再变化
-    act(() => {
-      vi.advanceTimersByTime(500)
-    })
-
-    expect(result.current.currentTime).toBe(timeBeforePause)
-  })
-
-  it("handleSeek 应该跳转到指定时间", () => {
-    const { result } = renderHook(() => useAudioPlayer())
-
-    act(() => {
-      result.current.handlePlay("voice-1", 10)
-    })
-
-    act(() => {
-      result.current.handleSeek("voice-1", 5, 10)
-    })
-
-    expect(result.current.currentTime).toBe(5)
-  })
-
-  it("handleSeek 对不同音色应该开始播放该音色", () => {
-    const { result } = renderHook(() => useAudioPlayer())
-
-    act(() => {
-      result.current.handlePlay("voice-1", 10)
-    })
-
-    act(() => {
-      result.current.handleSeek("voice-2", 3, 15)
-    })
-
-    expect(result.current.playingId).toBe("voice-2")
-    expect(result.current.currentTime).toBe(3)
-  })
-
-  it("handleTogglePlay 应该在播放和暂停之间切换", () => {
-    const { result } = renderHook(() => useAudioPlayer())
-
-    // 初始为暂停，调用应开始播放
-    act(() => {
-      result.current.handleTogglePlay("voice-1", 10)
-    })
-
-    expect(result.current.playingId).toBe("voice-1")
-
-    // 再次调用应暂停
-    act(() => {
-      result.current.handleTogglePlay("voice-1", 10)
-    })
-
-    expect(result.current.playingId).toBeNull()
-  })
-
-  it("stopPlayback 应该重置所有播放状态", () => {
-    const { result } = renderHook(() => useAudioPlayer())
-
-    act(() => {
-      result.current.handlePlay("voice-1", 10)
-    })
-
-    act(() => {
-      vi.advanceTimersByTime(300)
-    })
-
-    expect(result.current.playingId).toBe("voice-1")
-    expect(result.current.currentTime).toBeGreaterThan(0)
-
     act(() => {
       result.current.stopPlayback()
     })
 
     expect(result.current.playingId).toBeNull()
     expect(result.current.currentTime).toBe(0)
-
-    // 停止后定时器不应再触发
-    const timeAfterStop = result.current.currentTime
-    act(() => {
-      vi.advanceTimersByTime(500)
-    })
-    expect(result.current.currentTime).toBe(timeAfterStop)
-  })
-
-  it("返回值应该包含所有必要的方法和状态", () => {
-    const { result } = renderHook(() => useAudioPlayer())
-
-    expect(typeof result.current.handlePlay).toBe("function")
-    expect(typeof result.current.handlePause).toBe("function")
-    expect(typeof result.current.handleSeek).toBe("function")
-    expect(typeof result.current.handleTogglePlay).toBe("function")
-    expect(typeof result.current.stopPlayback).toBe("function")
-    expect(typeof result.current.playingId).toBe("object") // string | null
-    expect(typeof result.current.currentTime).toBe("number")
+    expect(mockAudioPause).toHaveBeenCalled()
   })
 })

apps/web/src/test/pages/voices/smoke.test.tsx

@@ -29,7 +29,7 @@ import "@/pages/voices/components/tts-modal/ErrorAlert"
 import "@/pages/voices/components/tts-modal/ResultPanel"
 import "@/pages/voices/components/tts-modal/types"
 import "@/pages/voices/components/VoiceFilterBar"
-import "@/pages/voices/components/MaterialVoiceCard"
+import "@/pages/voices/components/MaterialVoiceTab"
 
 // 类型与常量
 import "@/pages/voices/types"

apps/worker/worker_app/tasks/ingest.py

@@ -1,3 +1,4 @@
+import json
 import shutil
 import subprocess
 import tempfile
@@ -150,6 +151,214 @@ def extract_media_metadata(file_url: str, media_type: str) -> tuple[dict, bool]:
     return metadata, success
 
 
+# ── HEVC 自动转码辅助函数（模块级，便于单元测试）─────────────────────────
+HEVC_CODECS = ("hevc", "h265", "hvh1")
+# 转码目标：长边封顶 1920（只缩不放，与 validate 的 max_long_edge 一致），
+# 竖屏/横屏/超宽屏统一按长边等比缩放，短边自动按比例（-2 保证偶数）。
+TRANSCODE_MAX_LONG_EDGE = 1920
+TRANSCODE_TIMEOUT_SECONDS = 900
+# ffmpeg scale 滤镜中 if(...) 表达式内的逗号必须用 \, 转义，
+# 否则逗号被当作 filter 分隔符解析，报 "No such filter" / Invalid size。
+# subprocess list 传参不经 shell，\ 在 Python 字符串里直接写一个字面反斜杠即可。
+# 横屏（iw>=ih）限宽 min(1920,iw)、高 -2 自适应；竖屏（ih>iw）限高、宽自适应；
+# min() 保证小视频不放大。与 validate_transcode_output 的"长边 <= 1920"规则对齐，
+# 超宽屏（如 4000x1000）短边不触发旧的短边缩放、长边超限被误降级的问题由此消除。
+_TRANSCODE_VF = (
+    rf"scale=w=if(gte(iw\,ih)\,min({TRANSCODE_MAX_LONG_EDGE}\,iw)\,-2):"
+    rf"h=if(gt(ih\,iw)\,min({TRANSCODE_MAX_LONG_EDGE}\,ih)\,-2),format=yuv420p"
+)
+
+
+def is_hevc_codec(codec: str | None) -> bool:
+    """判断编码是否为 HEVC（不区分大小写）。"""
+    return (codec or "").lower() in HEVC_CODECS
+
+
+def probe_rotation(path: str) -> int | None:
+    """ffprobe 读取视频旋转角度（display matrix side data）。
+
+    返回 0/90/-90/180 等整数；无 side data 或探测失败返回 None。
+
+    注意：旧实现同时请求 side_data 和 stream_tags 且取输出第一行，
+    iOS 文件会输出两行（如 "270\\n90"）导致取到错误值，现仅读 side_data。
+    """
+    try:
+        result = subprocess.run(
+            [
+                "ffprobe",
+                "-v",
+                "error",
+                "-select_streams",
+                "v:0",
+                "-show_entries",
+                "side_data=rotation",
+                "-of",
+                "default=noprint_wrappers=1:nokey=1",
+                str(path),
+            ],
+            stdout=subprocess.PIPE,
+            stderr=subprocess.DEVNULL,
+            text=True,
+            timeout=60,
+        )
+        first_line = (result.stdout or "").strip().split("\n")[0].strip()
+        if not first_line:
+            return None
+        return int(float(first_line))
+    except (subprocess.TimeoutExpired, ValueError, OSError):
+        return None
+
+
+def is_portrait_rotation(rotation: int | None) -> bool:
+    """rotation side data 为 ±90/270 时表示竖屏拍摄。
+
+    注意：这只覆盖"存储横屏 + display matrix 旋转"的 iOS 风格视频；
+    物理竖屏视频（Android 常见，存储即 h>w、rotation=None/0）不会命中，
+    方向判定请用 is_portrait_video()。
+    """
+    return rotation in (90, 270, -90)
+
+
+def is_portrait_video(
+    stored_width: int | None,
+    stored_height: int | None,
+    rotation: int | None,
+) -> bool:
+    """按显示方向判断是否竖屏（显示高度 > 显示宽度）。
+
+    - rotation 为 90/270/-90 时，显示方向的宽高相对存储维度互换；
+    - rotation 为 0/180/None 时，显示方向即存储维度。
+
+    这样两类竖屏都能正确识别：
+    - iOS：存储 1920x1080 + rotation=90 → 显示 1080x1920 竖屏
+    - Android/物理竖屏：存储 1080x1920、无 rotation → 显示 1080x1920 竖屏
+    探测失败（维度为 None）时退回仅看 rotation，保证调用链不中断。
+    """
+    if not stored_width or not stored_height:
+        return is_portrait_rotation(rotation)
+    if is_portrait_rotation(rotation):
+        return stored_width > stored_height
+    return stored_height > stored_width
+
+
+def build_transcode_vf() -> str:
+    """构建转码视频滤镜（竖屏/横屏统一，按显示长边封顶 1920、只缩不放）。
+
+    依赖 ffmpeg 内置 autorotate（默认开启）按 display matrix 物理旋转画面，
+    输出自动剥离 rotation side data；滤镜只做等比缩放，方向无关：
+    横屏限宽、竖屏限高，短边 -2 自适应偶数，min() 保证小视频不放大。
+
+    旧实现的问题：
+    - 显式 transpose=1 与 autorotate 叠加，竖屏被二次旋转成横屏；
+    - 竖屏沿用按高缩放表达式，1080x1920 被错误缩成 608x1080；
+    - 仅按短边 1080 触发缩放，超宽屏（如 4000x1000）长边超 1920 会被
+      validate 拦截误降级，用户拿到浏览器无法播放的 HEVC 原文件。
+    """
+    return _TRANSCODE_VF
+
+
+def probe_dimensions(path: str) -> tuple[int | None, int | None]:
+    """ffprobe 读取视频宽高（像素维度）。"""
+    try:
+        result = subprocess.run(
+            [
+                "ffprobe",
+                "-v",
+                "error",
+                "-select_streams",
+                "v:0",
+                "-show_entries",
+                "stream=width,height",
+                "-of",
+                "csv=p=0:s=x",
+                str(path),
+            ],
+            stdout=subprocess.PIPE,
+            stderr=subprocess.DEVNULL,
+            text=True,
+            timeout=60,
+        )
+        text = (result.stdout or "").strip().split("\n")[0].strip()
+        width_str, height_str = text.split("x")
+        return int(width_str), int(height_str)
+    except (subprocess.TimeoutExpired, ValueError, OSError):
+        return None, None
+
+
+def probe_video_info(path: str) -> tuple[int | None, int | None, int | None]:
+    """一次 ffprobe 同时读取视频宽高与旋转角度（display matrix side data）。
+
+    返回 (width, height, rotation)；探测失败对应位置为 None。
+    合并维度/角度两次探测，减少大文件、高并发下的 ffprobe 进程开销。
+    rotation 仅取 stream side_data_list 的 Display Matrix（不读 tags.rotate，
+    避免 iOS 文件 tag 值与 side data 双来源取错）。用 -show_streams 全量 JSON
+    输出解析，兼容 ffmpeg 4.x/7.x（show_entries 嵌套 section 名跨版本不一致）。
+    """
+    try:
+        result = subprocess.run(
+            [
+                "ffprobe",
+                "-v",
+                "error",
+                "-select_streams",
+                "v:0",
+                "-show_streams",
+                "-of",
+                "json",
+                str(path),
+            ],
+            stdout=subprocess.PIPE,
+            stderr=subprocess.DEVNULL,
+            text=True,
+            timeout=60,
+        )
+        data = json.loads(result.stdout or "{}")
+        streams = data.get("streams") or []
+        if not streams:
+            return None, None, None
+        stream = streams[0]
+        width = int(stream["width"]) if stream.get("width") else None
+        height = int(stream["height"]) if stream.get("height") else None
+        rotation = None
+        for side in stream.get("side_data_list") or []:
+            if side.get("side_data_type") == "Display Matrix" and side.get("rotation") is not None:
+                deg = int(round(float(side["rotation"]))) % 360
+                # ffprobe：顺时针 90 拍摄输出 90，逆时针 90 输出 -90（归一为 270）
+                rotation = {0: 0, 90: 90, 180: 180, 270: -90}.get(deg, deg if deg in (90, 180) else None)
+                break
+        return width, height, rotation
+    except (subprocess.TimeoutExpired, ValueError, OSError, json.JSONDecodeError, KeyError, TypeError):
+        return None, None, None
+
+
+def validate_transcode_output(
+    output_path: str,
+    expected_portrait: bool,
+    max_long_edge: int = TRANSCODE_MAX_LONG_EDGE,
+) -> bool:
+    """校验转码产物方向与维度。
+
+    - 竖屏源：产物必须 height > width，且仍有 rotation side data 视为失败
+      （播放器会二次旋转成横屏）
+    - 横屏源：产物必须 width >= height
+    - 长边不得超过 max_long_edge（只缩不放）
+    校验失败时调用方应降级使用原始文件，不允许产出方向错误的文件覆盖。
+    """
+    width, height = probe_dimensions(output_path)
+    if not width or not height:
+        return False
+    if expected_portrait and height <= width:
+        return False
+    if not expected_portrait and width < height:
+        return False
+    if max(width, height) > max_long_edge:
+        return False
+    # 产物仍带 rotation side data 说明方向没有物理固化，播放器会再次旋转
+    if probe_rotation(output_path) is not None:
+        return False
+    return True
+
+
 @celery_app.task(name="worker.ingest_asset")
 def ingest_asset(job_id: str) -> dict:
     """
@@ -245,15 +454,13 @@ def ingest_asset(job_id: str) -> dict:
             # 浏览器 WebCodecs 硬件解码 HEVC 输出黑帧，上传时自动转码
             # 失败时降级使用原始文件，不阻塞上传流程
             if media_type == "video" and local_file and local_file.exists():
-                codec = (metadata.get("codec") or "").lower()
-                if codec in ("hevc", "h265", "hvh1"):
+                if is_hevc_codec(metadata.get("codec")):
                     logger.info(
                         "检测到 HEVC 编码 (codec=%s)，启动转码: job_id=%s",
-                        codec,
+                        metadata.get("codec"),
                         job_id,
                     )
                     _tc_tmp = None
-                    _needs_rotation = False
 
                     # ── Step 1: 磁盘空间检查（独立 try/except，失败仍尝试转码）──
                     try:
@@ -264,50 +471,21 @@ def ingest_asset(job_id: str) -> dict:
                     except Exception as _disk_err:
                         logger.warning("磁盘检查失败，仍尝试转码: job_id=%s err=%s", job_id, _disk_err)
 
-                    # ── Step 2: ffprobe 旋转检测（独立 try/except，失败不阻塞转码）──
-                    try:
-                        _probe_cmd = [
-                            "ffprobe",
-                            "-v",
-                            "error",
-                            "-select_streams",
-                            "v:0",
-                            "-show_entries",
-                            "side_data=rotation",
-                            "-show_entries",
-                            "stream_tags=rotate",
-                            "-of",
-                            "default=noprint_wrappers=1:nokey=1",
-                            str(local_file),
-                        ]
-                        _probe_result = subprocess.run(
-                            _probe_cmd,
-                            stdout=subprocess.PIPE,
-                            stderr=subprocess.DEVNULL,
-                            text=True,
-                            timeout=60,  # 大文件在容器 overlay 文件系统上解析可能较慢
-                        )
-                        _rotation_str = (_probe_result.stdout or "").strip().split("\n")[0]
-                        if _rotation_str in ("90", "270", "-90"):
-                            _needs_rotation = True
-                            logger.info(
-                                "检测到竖屏视频 (rotation=%s)，将物理旋转画面: job_id=%s",
-                                _rotation_str,
-                                job_id,
-                            )
-                    except subprocess.TimeoutExpired:
-                        logger.warning(
-                            "ffprobe 旋转检测超时(60s)，跳过旋转继续转码: job_id=%s",
-                            job_id,
-                        )
-                        _needs_rotation = False
-                    except Exception as _probe_err:
-                        logger.warning(
-                            "ffprobe 旋转检测异常，跳过旋转继续转码: job_id=%s err=%s",
-                            job_id,
-                            _probe_err,
-                        )
-                        _needs_rotation = False
+                    # ── Step 2: 方向检测（按显示方向判定竖/横屏）──────────────
+                    # 不能只看 rotation side data：Android 等设备的物理竖屏视频
+                    # 存储维度已是 h>w 且 rotation=0/None，只看 rotation 会误判横屏、
+                    # 套用横屏滤镜把 1080x1920 压成 608x1080，转码产物校验失败降级，
+                    # 用户拿到 HEVC 原文件浏览器仍黑帧。
+                    _src_w, _src_h, _rotation = probe_video_info(str(local_file))
+                    _is_portrait = is_portrait_video(_src_w, _src_h, _rotation)
+                    logger.info(
+                        "视频方向检测: stored=%sx%s rotation=%s portrait=%s: job_id=%s",
+                        _src_w,
+                        _src_h,
+                        _rotation,
+                        _is_portrait,
+                        job_id,
+                    )
 
                     # ── Step 3: ffmpeg 转码（独立 try/except）──
                     try:
@@ -315,11 +493,11 @@ def ingest_asset(job_id: str) -> dict:
                         _tc_tmp = Path(_tc_tmp_file.name)
                         _tc_tmp_file.close()  # 关闭文件描述符，ffmpeg 会自己打开
 
-                        # 构建 video filter：竖屏先旋转再缩放
-                        if _needs_rotation:
-                            _vf = "transpose=1,scale='if(gt(ih,1080),-2,iw)':'if(gt(ih,1080),1080,ih)'"
-                        else:
-                            _vf = "scale='if(gt(ih,1080),-2,iw)':'if(gt(ih,1080),1080,ih)'"
+                        # 旋转交给 ffmpeg 内置 autorotate（按 display matrix 物理旋转，
+                        # 输出自动剥离 side data）；滤镜只做 1080p 等比"只缩不放"。
+                        # 注意不能再加 transpose：旧逻辑 autorotate + transpose 双重旋转，
+                        # 竖屏被转成横屏；scale 表达式内逗号必须 \, 转义（见 build_transcode_vf）。
+                        _vf = build_transcode_vf()
 
                         _cmd = [
                             "ffmpeg",
@@ -333,7 +511,7 @@ def ingest_asset(job_id: str) -> dict:
                             "-crf",
                             "18",
                             "-vf",
-                            _vf + ",format=yuv420p",
+                            _vf,
                             "-colorspace",
                             "bt709",
                             "-color_primaries",
@@ -344,54 +522,62 @@ def ingest_asset(job_id: str) -> dict:
                             "yuv420p",
                             "-level",
                             "4.2",
+                            "-c:a",
+                            "aac",
+                            "-b:a",
+                            "128k",
+                            "-movflags",
+                            "+faststart",
+                            str(_tc_tmp),
                         ]
-                        # 竖屏视频：清除旋转元数据
-                        if _needs_rotation:
-                            _cmd.extend(["-metadata:s:v:0", "rotate=0"])
-                        _cmd.extend(
-                            [
-                                "-c:a",
-                                "aac",
-                                "-b:a",
-                                "128k",
-                                "-movflags",
-                                "+faststart",
-                                str(_tc_tmp),
-                            ]
-                        )
                         _proc = subprocess.run(
                             _cmd,
                             stdout=subprocess.DEVNULL,
                             stderr=subprocess.PIPE,
                             text=True,
-                            timeout=900,
+                            timeout=TRANSCODE_TIMEOUT_SECONDS,
                         )
                         if _proc.returncode == 0 and _tc_tmp.exists() and _tc_tmp.stat().st_size > 0:
-                            from video_processing.oss_helpers import upload_to_oss
-
-                            _p = Path(job.storage_key)
-                            _new_key = str(_p.parent / (_p.stem + "_h264" + _p.suffix))
-                            _url = upload_to_oss(_tc_tmp, _new_key)
-                            if _url:
-                                # 先提取元数据，确认成功后再更新 storage_key（避免脏数据）
-                                _new_metadata, _new_extract_success = extract_media_metadata(
-                                    str(_tc_tmp),
-                                    media_type,
-                                )
-                                if _new_extract_success:
-                                    job.storage_key = _new_key
-                                    metadata = _new_metadata
-                                    extract_success = _new_extract_success
-                                logger.info(
-                                    "HEVC→H.264 转码完成: job_id=%s key=%s",
+                            # ── Step 4: 方向/维度校验，不符则降级，杜绝横屏文件覆盖 ──
+                            if not validate_transcode_output(str(_tc_tmp), _is_portrait):
+                                _w, _h = probe_dimensions(str(_tc_tmp))
+                                _rot = probe_rotation(str(_tc_tmp))
+                                logger.error(
+                                    "转码产物方向/维度校验失败，降级使用原始文件: "
+                                    "job_id=%s source_rotation=%s portrait=%s out=%sx%s out_rotation=%s",
                                     job_id,
-                                    _new_key[:80],
+                                    _rotation,
+                                    _is_portrait,
+                                    _w,
+                                    _h,
+                                    _rot,
                                 )
                             else:
-                                logger.warning(
-                                    "转码文件上传 OSS 失败，使用原始文件: job_id=%s",
-                                    job_id,
-                                )
+                                from video_processing.oss_helpers import upload_to_oss
+
+                                _p = Path(job.storage_key)
+                                _new_key = str(_p.parent / (_p.stem + "_h264" + _p.suffix))
+                                _url = upload_to_oss(_tc_tmp, _new_key)
+                                if _url:
+                                    # 先提取元数据，确认成功后再更新 storage_key（避免脏数据）
+                                    _new_metadata, _new_extract_success = extract_media_metadata(
+                                        str(_tc_tmp),
+                                        media_type,
+                                    )
+                                    if _new_extract_success:
+                                        job.storage_key = _new_key
+                                        metadata = _new_metadata
+                                        extract_success = _new_extract_success
+                                    logger.info(
+                                        "HEVC→H.264 转码完成: job_id=%s key=%s",
+                                        job_id,
+                                        _new_key[:80],
+                                    )
+                                else:
+                                    logger.warning(
+                                        "转码文件上传 OSS 失败，使用原始文件: job_id=%s",
+                                        job_id,
+                                    )
                         else:
                             _tail = _proc.stderr[-300:] if _proc.stderr else ""
                             logger.warning(

packages/adapters/sqlalchemy_impl/models.py

@@ -55,6 +55,7 @@ class ProjectModel(Base):
 
 class AssetLibraryModel(Base):
     __tablename__ = "asset_libraries"
+    __table_args__ = (UniqueConstraint("project_id", "kind", name="uq_asset_libraries_project_kind"),)
 
     id = Column(String(36), primary_key=True)
     project_id = Column(String(36), nullable=True, index=True)