424 Commits

Author	SHA1	Message	Date
xiaoxia	1a2f34e508	fix(cover): overlay title text on cover image after frame extraction ... Root cause: cover frame is extracted during preview video rendering (Step 3), but user selects title AFTER rendering (Step 4). The cover frame naturally has no title because it was captured before the title was chosen. Fix: In generation_cover.py, after finding cover_url_from_task, check if plan.config['title']['text'] has a title. If yes, download the cover image, use ffmpeg drawtext to overlay the title (white text + black shadow, centered near bottom), upload to OSS, and return the new URL. On failure, fallback to original cover_url. Changes: - Add _overlay_title_on_cover_image() helper in generation_cover.py - Add _escape_drawtext_text() for ffmpeg special character escaping - Modify cover route to check for title and overlay if present - 12 new unit tests (test_cover_title_overlay.py) - All 13603 tests pass Fixes: cover title missing issue in staging	2026-08-15 15:42:56 +08:00
xiaoxia	59faf37fc9	fix: 统一封面管道增加cover_url多步fallback查找，修复500错误 (#1373) ... Co-authored-by: xiaoxia <dev@xiaoxiajianji.com> Co-committed-by: xiaoxia <dev@xiaoxiajianji.com>	2026-08-15 11:41:00 +08:00
xiaoxia	206d517a91	fix: 删除封面生成多余logger调用+增加source_edit_plan_id查找兜底 (#1372)	2026-08-15 10:09:16 +08:00
xiaoxia	0301370dd8	refactor: 统一封面生成管道 — 从渲染后视频抽帧作为封面 (#1371) ... Co-authored-by: xiaoxia <dev@xiaoxiajianji.com> Co-committed-by: xiaoxia <dev@xiaoxiajianji.com>	2026-08-14 22:16:00 +08:00
xiaoxia	8e4a8a8184	fix: videoRatio aspect ratio parsed as resolution causing scale=9:1920 (#1364)	2026-08-14 00:56:54 +08:00
xiaoxia	21e84c71c4	perf: render-time cover frame pre-extraction + FFmpeg fallback, remove MediaKit (#1360) ... Co-authored-by: xiaoxia <dev@xiaoxiajianji.com> Co-committed-by: xiaoxia <dev@xiaoxiajianji.com>	2026-08-13 20:51:02 +08:00
xiaoxia	17174e2cf5	refactor(cover): 封面生成API从templates_editor解耦到generation路由 (#1359) ... Co-authored-by: xiaoxia <dev@xiaoxiajianji.com> Co-committed-by: xiaoxia <dev@xiaoxiajianji.com>	2026-08-13 18:56:00 +08:00
xiaoxia	68fa7fd163	fix(cover): 修复 OSS 路径双斜杠导致 MediaKit 404 (#1357) ... Co-authored-by: xiaoxia <dev@xiaoxiajianji.com> Co-committed-by: xiaoxia <dev@xiaoxiajianji.com>	2026-08-13 17:51:00 +08:00
xiaoxia	7b9e803603	fix(#1341): 增强封面生成错误日志和前端提示 + ESLint 修复 + prettier 格式化 (#1347)	2026-08-12 22:28:38 +08:00
xiaoxia	887e66b0f9	refactor(#1341): 删除 EditingPlanner 废弃的封面代码	2026-08-12 17:01:04 +08:00
CI Bot	947b3ed86e	fix(#1341): MediaKit 视频理解从 API 路由移到 Worker 任务 ... P0 Bug: analyze_videos() 在 HTTP 请求中同步调用耗时 30-60s， 前端 10s 超时。 修复： - 从 generation_preview.py 移除 MediaKit 同步调用 - 在 Worker generate_video 任务中、渲染之前调用 - 分析结果保存到 task.extra_meta['asset_analyses']	2026-08-12 12:36:13 +08:00
CI Bot	8ac6d1e6bc	fix(#1341): analyze_videos 添加缺失的 prompt 参数 ... generation_preview.py 调用 mk_client.analyze_videos() 缺少必传 prompt 参数， 导致 TypeError 被 try/except 吞掉，MediaKit 视频理解从未生效。	2026-08-12 09:48:16 +08:00
CI Bot	7d6c0cc67e	fix: cover.py step3 改用 user+template 查找预览任务；删除 generation.py 冗余 rendered_url ... P0 修复：预览任务创建时 plan_id 尚未存在，source_edit_plan_id 始终为空， 导致封面生成第三步查找永远匹配不到。改为按 user_id + template_id + is_preview + status=completed 查找最近预览任务。 技术债：generation.py 复用预览产物分支删除冗余 rendered_url 写入， 统一只写 rendered_storage_key。 新增 repository 方法：list_latest_completed_preview(user_id, template_id)	2026-08-12 00:07:06 +08:00
CI Bot	43fa4a5575	test: fix test_ai_tasks + format cover.py for #1341	2026-08-11 22:02:08 +08:00
CI Bot	308fbf2130	fix: #1341 cover generation full-chain fix + MediaKit integration	2026-08-11 21:01:35 +08:00
CI Bot	3d5438a3af	fix: cover generation full-chain fix — bare URL, remove stub fallback, clean dead code ... - cover.py: use get_url() (bare URL) instead of get_download_url() (signed URL) to avoid OSS sign_url encoding / as %2F causing 403 - ai_service.py: remove SVG stub fallback in _call_ai_cover_service(), raise RuntimeError with clear error message instead - cover.py: wrap run_generate_cover() in try/except RuntimeError → HTTPException - schemas.py: remove unused CoverGenerateResponse class - Note: rendered/* OSS public read must be configured for bare URL to work	2026-08-11 18:22:23 +08:00
CI Bot	441d61127d	refactor(cover): use rendered preview video for cover generation, remove dead endpoints ... BREAKING CHANGE: cover/extract and cover/smart endpoints removed (unused by frontend) Changes: - cover.py: editor_generate_cover() now reads rendered_storage_key from plan config instead of querying assets table via SessionLocal - Removed 4 dead endpoints: GET/PUT /cover, POST /cover/extract, POST /cover/smart - Removed 4 unused schemas: CoverConfigResponse, CoverUpdateRequest, CoverExtractRequest, CoverSmartRequest - Deleted stale fix_confirm.py and test-results artifact - rendered/* videos use signed URL (1h) since path is not publicly readable - Flow is now strictly serial: preview render -> cover extraction from preview	2026-08-11 17:45:04 +08:00
CI Bot	d23db7654e	fix(cover): use bare public URL for video since OSS uploads/* public read is enabled ... - Changed get_download_url(expires_seconds=3600) to get_url() in cover.py - OSS Bucket Policy now allows anonymous reads on uploads/* - Bare URL verified: HEAD returns 200 for video files - Eliminates signed URL encoding issues that caused MediaKit 403	2026-08-11 17:21:59 +08:00
CI Bot	7ad076243b	fix(cover): use signed URL with 1h expiry for video in cover generation ... Revert bare get_url() back to get_download_url(expires_seconds=3600) because uploads/* path is not publicly readable on OSS. MediaKit downloads video via GET, and signed URL works correctly. 1-hour expiry is more than sufficient for MediaKit frame extraction.	2026-08-11 16:43:13 +08:00
CI Bot	7b534c950b	fix(cover): increase MediaKit timeout + use public URL + extend frontend timeout ... Three fixes for cover generation: 1. cover.py: get_download_url -> get_url for video URL (OSS public read) 2. ai_service.py: extract_frames poll_interval=3.0, max_poll_attempts=60 (180s) 3. aiFeatures.ts: generateCover axios timeout 10s -> 180s	2026-08-11 16:09:07 +08:00
CI Bot	e79464b9ef	fix(cover): use SessionLocal() instead of get_db_session() as context manager ... get_db_session() is a FastAPI Depends generator (uses yield), not a context manager. Using "with get_db_session()" raises: "generator object does not support the context manager protocol" Replace with SessionLocal() + try/finally for proper session lifecycle management in non-FastAPI-dependency context.	2026-08-11 15:41:09 +08:00
CI Bot	d6f0eef929	fix: cover.py import路径修正 app.database → app.dependencies ... get_db_session 定义在 app.dependencies 而非 app.database， 错误 import 导致 MediaKit 抽帧异常被吞，封面返回 SVG 占位图。	2026-08-11 14:29:02 +08:00
CI Bot	142d4331cc	fix: AI Code Review 阻塞级问题修复 — 分辨率一致性校验+空指针防护	2026-08-11 06:44:32 +08:00
CI Bot	ed80fd7205	fix: 修复 _get_task_output_url 导入路径并更新测试	2026-08-10 23:54:47 +08:00
CI Bot	29994ffcbc	feat: 统一预览与确认生成渲染路径 — 预览=正式品质+确认复用 ... ## 渲染管线统一 - UnifiedRenderService: 移除 is_preview 参数，统一 CRF 23 + medium preset - RenderAdapter: 移除 is_preview 参数，统一执行校验和缩略图生成 - Worker generation.py: 移除 480p+1M 码率覆盖逻辑 - generation_preview.py: 删除 PREVIEW_RESOLUTION 常量和 _calc_preview_resolution() ## 确认生成复用预览产物 - Domain: 新增 GenerationTask.mark_confirmed() 方法 - confirm_generation: 预览已完成时直接复用产物（秒出），无需重新渲染 - Editor generate: 检查 plan 关联的预览任务，plan未修改时复用产物 - Schema: CreatePreviewGenerationTaskRequest 新增 source_edit_plan_id 字段 ## 测试更新 - test_1280_preview_speedup: 重写为验证统一品质参数 - test_confirm_generation: 重写为验证复用逻辑 ## 交付标准 - 预览渲染质量与确认生成一致（1080p, CRF 23, medium） - 预览后确认生成直接复用产物（秒出） - Worker 端无任何 is_preview 低质量渲染代码残留	2026-08-10 22:32:35 +08:00
CI	8d260f16f2	style: black + isort 格式化修复	2026-08-10 21:24:19 +08:00
CI Bot	e09fdda74c	fix: 确认生成兜底增强 — project_id 为空时通过 user_id 查找素材 ... 根因：模板编辑器直接创建的草稿 plan 没有 project_id 和 config.asset_ids， 导致 _auto_fallback_auto_material_mode 跳过素材分配， can_generate 报 "没有可渲染的就绪片段"。 修复： - asset_repository 新增 find_ready_videos_by_user 方法 - _auto_fallback_auto_material_mode 增加策略2：project_id 为空时 通过 uploaded_by_user_id 查找用户上传的就绪视频 - generation.py 传递 current_user.user.id 给兜底函数	2026-08-10 20:19:12 +08:00
xiaoxia	e2b3811277	fix: FFmpeg mjpeg 缩略图/封面生成添加 format=yuvj420p ... 修复 FFmpeg 新版中 mjpeg 编码器拒绝非全范围 YUV 输入的问题： - thumbnail_generator.py: scale filter 添加 format=yuvj420p - processor.py: 添加 pix_fmt=yuvj420p - cover_generator.py: 两处 vf filter 添加 format=yuvj420p - cover_service.py: vf filter + simple fallback 添加 pix_fmt - asset_analyzer.py: 添加 -pix_fmt yuvj420p 修复错误: Non full-range YUV is non-standard, set strict_std_compliance to at most unofficial to use it.	2026-08-10 11:37:22 +08:00
xiaoxia	0c6ec8995f	fix: 封面模板 API 错误处理加固 (#1329)	2026-08-10 10:40:11 +08:00
xiaoxia	153b38a62f	feat: 封面模板 CRUD API (#1323) ... Co-authored-by: xiaoxia <dev@xiaoxiajianji.com> Co-committed-by: xiaoxia <dev@xiaoxiajianji.com>	2026-08-09 18:26:00 +08:00
CI Bot	de30703d41	style: auto-format with black + isort + prettier [skip ci-format-check]	2026-08-09 03:23:53 +00:00
xiaoxia	68e41f6173	fix: 预览生成接口透传 voice_library_id 修复预览无声问题 ... 根因：预览生成接口（POST /preview）硬编码 voice_library_id=""， 导致用户选择的上传音频从未传递给 Worker，渲染时没有音频可混入。 修复： - CreatePreviewGenerationTaskRequest 增加 voice_library_id 字段 - 预览端点将 request.voice_library_id 透传给 CreateGenerationTaskCommand - 确认生成接口已自动继承源预览任务的 voice_library_id - 新增 4 个单元测试验证 schema 字段 下游链路（Worker 下载音频 → 渲染引擎混音）无需修改， 只需前端在创建预览任务时传入 voice_library_id 即可。	2026-08-09 11:21:08 +08:00
xiaoxia	7d4e1b1b2f	fix: 确认生成 API 改为复用 worker.generate_video 渲染路径（适配 develop 分支） ... 从 main 分支 PR #1308 手动适配到 develop 分支，保留 develop 现有功能： Schema 层： - 新增 ConfirmGenerationRequest schema - CreateGenerationTaskRequest 增加 6 个新字段（is_preview, source_task_id, output_width, output_height, cover_url, custom_title） - GenerationTaskResponse 增加对应字段 Domain 层： - GenerationTask 实体增加 source_task_id, output_width, output_height, cover_url, custom_title 字段 - create() 工厂方法增加对应参数 Application 层： - CreateGenerationTaskCommand 增加对应字段 - UseCase 执行时传递新字段 DB 层： - GenerationTaskModel 增加 5 个新列（is_preview 已存在于 develop） - 仓储 _to_domain/create/update 映射更新 API 路由： - 新增 POST /tasks/{task_id}/confirm 端点 - 使用 safe_enqueue_generation_task 适配 develop 任务入队机制 - 更新 _to_generation_task_response 和 retry 端点 Worker 层： - _load_task_info 返回新字段 - generate_video 支持动态分辨率（output_width/output_height 覆盖默认分辨率） Alembic 迁移 054： - generation_tasks 表增加 source_task_id, output_width, output_height, cover_url, custom_title 列 单元测试： - tests/unit/test_confirm_generation.py（7 个测试用例全部通过） - 适配 develop 的 safe_enqueue_generation_task 入队机制	2026-08-08 21:44:30 +08:00
xiaoxia	248fd5408c	feat: 预览数量 preview_count 上限校验 + Worker 并发提升至 4 (#1292)	2026-08-08 11:14:45 +08:00
xiaoxia	76f41529fe	fix: #1286 can_generate 增加最后防线自动修复，解决「没有可渲染的就绪片段」报错 (#1288)	2026-08-07 22:42:20 +08:00
xiaoxia	aff8f6dae8	fix: 生成时报错「没有可渲染的就绪片段」— mark_clips_ready 仅标记有素材的片段 (#1270) (#1276) ... Co-authored-by: xiaoxia <dev@xiaoxiajianji.com> Co-committed-by: xiaoxia <dev@xiaoxiajianji.com>	2026-08-07 20:46:00 +08:00
xiaoxia	316f01b3f0	feat: 预览生成支持随机选素材 + 随机截取片段 (#1256) (#1268)	2026-08-07 19:50:01 +08:00
xiaoxia	4f323b394b	fix: 音频上传422错误 — create_asset project_id改为可选，从library自动推导 (#1260)	2026-08-07 17:12:36 +08:00
xiaoxia	bb5bed0f4c	refactor: 统一智能选素材逻辑，迁移 generation_tasks smart 模式到 smart_match (#1243)	2026-08-05 11:27:57 +08:00
xiaoxia	e27a48f0e0	feat: implement POST /assets/smart-match endpoint (#1242) ... Co-authored-by: xiaoxia <dev@xiaoxiajianji.com> Co-committed-by: xiaoxia <dev@xiaoxiajianji.com>	2026-08-05 09:51:01 +08:00
xiaoxia	004de97394	fix(render): 画中画模式全面下线 + 标题ASS自动换行 (#1239)	2026-08-04 18:36:19 +08:00
xiaoxia	f396e3ecb4	fix(preview): 预览生成从模板读取 editing_mode 作为 strategy_id (#1235)	2026-08-04 14:43:49 +08:00
xiaoxia	a356639a34	fix: 修复模板查询方法名（repo.get替代repo.get_by_id） (#1231)	2026-08-04 01:35:23 +08:00
xiaoxia	405bce6e1b	fix: 预览分辨率自动从模板mode推断（Fix3补充） (#1230)	2026-08-04 01:08:22 +08:00
xiaoxia	9808a6eef9	fix: URL签名降级逻辑 — bucket未配置时返回原始URL而非构造URL	2026-08-03 23:25:56 +08:00
xiaoxia	a9d55dc2cf	fix: 预览视频URL签名 — 私有bucket返回预签名URL避免403黑屏	2026-08-03 22:45:17 +08:00
xiaoxia	8ea16aae74	fix: MediaKit 视频理解轮询限时至30s，修复 nginx 504 超时 ... - analyze_videos() 轮询参数从默认(3s*60=180s)改为(2s*15=30s) - 预留30s给 LLM 调用，总耗时控制在 nginx 60s 限制内 - 超时自动降级到无分析模式，不影响主流程 - 修复 URL→asset_id 映射改用并行列表，消除重复 URL 覆盖风险	2026-08-03 12:40:03 +08:00
xiaoxia	82bc193693	feat: #1209 AI智能选片段接入MediaKit视频理解 ... - MediaKitClient 新增 analyze_videos() 方法，调用视频理解智能策略 API - _call_ai_recommend_service 支持 asset_analyses 参数，将视频内容分析注入 prompt - API 路由层自动获取素材 URL 并调用 MediaKit 分析视频内容 - LLM 可根据视频实际内容做智能编排（叙事连贯、场景匹配等） - 降级策略：MediaKit 不可用时保持原有 fallback 行为 - 15 个新单元测试全部通过	2026-08-03 09:43:34 +08:00
xiaoxia	9ac15d8edc	fix: 预置音色 preview 改为顺序生成，避免 DashScope 429 限流 (#1223) ... fix: 预置音色 preview 改为顺序生成 (#1223)	2026-08-03 00:36:07 +08:00
xiaoxia	7ef0366de0	feat: 预置音色 preview_url 动态生成，解决签名 URL 过期问题 (#1222) ... feat: 预置音色 preview_url 动态生成 (#1222) - _resolve_preset_preview_url: 缓存优先+CosyVoice合成+降级fallback - _resolve_all_preset_preview_urls: ThreadPoolExecutor并行生成 - 修改list_preset_voices/list_voices_unified注入CosyVoiceService - 13个新单测，117个相关测试全绿	2026-08-02 22:14:12 +08:00