Compare commits
2 Commits
| Author | SHA1 | Date | |
|---|---|---|---|
 xiaoxia
|
7b38e7a624 | ||
|
xiaoxia
|
5ab30f7706 |
@@ -1,26 +0,0 @@
|
||||
"""add ai_tags to asset_atom_clips for #1970 fragment-level AI tagging
|
||||
|
||||
Revision ID: 082_atom_clip_ai_tags
|
||||
Revises: 081_add_gpu_lipsync
|
||||
Create Date: 2026-09-18
|
||||
"""
|
||||
|
||||
import sqlalchemy as sa
|
||||
|
||||
from alembic import op
|
||||
|
||||
revision = "082_atom_clip_ai_tags"
|
||||
down_revision = "081_add_gpu_lipsync"
|
||||
branch_labels = None
|
||||
depends_on = None
|
||||
|
||||
|
||||
def upgrade() -> None:
|
||||
op.add_column(
|
||||
"asset_atom_clips",
|
||||
sa.Column("ai_tags", sa.JSON(), nullable=True),
|
||||
)
|
||||
|
||||
|
||||
def downgrade() -> None:
|
||||
op.drop_column("asset_atom_clips", "ai_tags")
|
||||
@@ -1,117 +0,0 @@
|
||||
import { expect, test, type APIRequestContext, type Page } from "@playwright/test"
|
||||
|
||||
const PASSWORD = "SmokePass123!"
|
||||
const apiBase = process.env.E2E_API_BASE || "/api/v1"
|
||||
const apiOrigin = apiBase.endsWith("/api/v1") ? apiBase.slice(0, -"/api/v1".length) : ""
|
||||
|
||||
async function routeBrowserApiToTestApi(page: Page) {
|
||||
if (!apiOrigin) return
|
||||
await page.route("**/api/v1/**", async (route) => {
|
||||
const sourceUrl = new URL(route.request().url())
|
||||
const response = await route.fetch({
|
||||
url: `${apiOrigin}${sourceUrl.pathname}${sourceUrl.search}`,
|
||||
})
|
||||
await route.fulfill({ response })
|
||||
})
|
||||
}
|
||||
|
||||
async function loginWithRetry(request: APIRequestContext, email: string, password: string) {
|
||||
for (let i = 0; i <= 2; i++) {
|
||||
const r = await request.post(`${apiBase}/auth/login`, { data: { email, password } })
|
||||
if (r.status() !== 429) {
|
||||
expect(r.ok(), `login: ${await r.text()}`).toBeTruthy()
|
||||
return (await r.json()).access_token as string
|
||||
}
|
||||
console.log(`[douyin] 429 retry ${i + 1}/2`)
|
||||
await new Promise((res) => setTimeout(res, 65000))
|
||||
}
|
||||
throw new Error("Login retries exhausted")
|
||||
}
|
||||
|
||||
/**
|
||||
* #1972 抖音文案提取冒烟
|
||||
*
|
||||
* 路径：文案库页面 → 点「🎬 从抖音提取」→ 粘贴分享文案 → 点「开始提取」
|
||||
* → mock /api/v1/scripts/extract-from-douyin 返回稳定文案 → 断言「新建文案」弹窗中预填了非空文案
|
||||
*/
|
||||
test.describe("Douyin Script Extraction (#1972)", () => {
|
||||
test("extract flow: open modal, paste link, text prefilled in create modal", async ({
|
||||
page,
|
||||
request,
|
||||
}) => {
|
||||
test.setTimeout(180_000)
|
||||
await page.setViewportSize({ width: 1440, height: 900 })
|
||||
|
||||
const suffix = Math.random().toString(36).slice(2, 8)
|
||||
const email = `e2e-douyin-${suffix}@example.com`
|
||||
await request.post(`${apiBase}/auth/register`, {
|
||||
data: { email, password: PASSWORD, username: `e2e_dy_${suffix}` },
|
||||
})
|
||||
const token = await loginWithRetry(request, email, PASSWORD)
|
||||
const authHeader = { Authorization: `Bearer ${token}` }
|
||||
|
||||
const proj = await request.post(`${apiBase}/projects`, {
|
||||
headers: authHeader,
|
||||
data: { name: `Smoke Douyin ${suffix}` },
|
||||
})
|
||||
const projectId = (await proj.json()).id ?? (await proj.json()).project_id
|
||||
await request.post(`${apiBase}/video-libraries`, {
|
||||
headers: authHeader,
|
||||
data: { project_id: projectId, name: "Smoke" },
|
||||
})
|
||||
|
||||
await page.addInitScript((t: string) => {
|
||||
window.localStorage.setItem("access_token", t)
|
||||
window.localStorage.setItem(
|
||||
"auth-storage",
|
||||
JSON.stringify({ state: { token: t, user: null } }),
|
||||
)
|
||||
}, token)
|
||||
await routeBrowserApiToTestApi(page)
|
||||
|
||||
// Mock 抖音提取接口返回稳定文案
|
||||
const extractedText = "大家好，今天给大家推荐一款超好用的产品，性价比非常高，快来看看吧！"
|
||||
await page.route("**/api/v1/scripts/extract-from-douyin", (route) =>
|
||||
route.fulfill({
|
||||
status: 200,
|
||||
contentType: "application/json",
|
||||
body: JSON.stringify({ text: extractedText, duration_seconds: 15 }),
|
||||
}),
|
||||
)
|
||||
// 文案列表空态
|
||||
await page.route(
|
||||
(url) => url.pathname.endsWith("/scripts") && !url.pathname.includes("extract-from-douyin"),
|
||||
(route) =>
|
||||
route.fulfill({
|
||||
status: 200,
|
||||
contentType: "application/json",
|
||||
body: JSON.stringify({ items: [], total: 0, page: 1, page_size: 20 }),
|
||||
}),
|
||||
)
|
||||
|
||||
await page.goto("/app/scripts")
|
||||
// 文案库页面加载
|
||||
await expect(page.getByText(/文案库|文案/).first()).toBeVisible({ timeout: 30000 })
|
||||
|
||||
// 点「🎬 从抖音提取」按钮
|
||||
await page.getByRole("button", { name: /从抖音提取/ }).click()
|
||||
await expect(page.getByText("从抖音视频提取文案")).toBeVisible({ timeout: 5000 })
|
||||
|
||||
// 在 TextArea 粘贴"抖音分享文案"
|
||||
const textarea = page.locator(".ant-modal textarea").first()
|
||||
await expect(textarea).toBeVisible()
|
||||
await textarea.fill("8.88 复制打开抖音，看看【推荐视频】https://v.douyin.com/abcDEF/")
|
||||
|
||||
// 点「开始提取」
|
||||
await page.getByRole("button", { name: "开始提取" }).click()
|
||||
await expect(page.getByText(/提取中/)).toBeVisible({ timeout: 3000 })
|
||||
|
||||
// 等待抖音弹窗关闭，「新建文案」弹窗打开并预填提取文案
|
||||
await expect(page.getByText("从抖音视频提取文案")).not.toBeVisible({ timeout: 15000 })
|
||||
await expect(page.getByText("新建文案")).toBeVisible({ timeout: 5000 })
|
||||
const createTextarea = page.locator(".ant-modal textarea").first()
|
||||
await expect(createTextarea).toBeVisible()
|
||||
await expect(createTextarea).toHaveValue(new RegExp(extractedText.slice(0, 10)))
|
||||
console.log("[douyin] Extraction flow completed ✓, text length:", extractedText.length)
|
||||
})
|
||||
})
|
||||
@@ -1,4 +1,4 @@
|
||||
import { expect, test, type APIRequestContext, type Page } from "@playwright/test"
|
||||
import { expect, test, type APIRequestContext } from "@playwright/test"
|
||||
import * as fs from "node:fs"
|
||||
import * as path from "node:path"
|
||||
import { fileURLToPath } from "node:url"
|
||||
@@ -8,8 +8,7 @@ const PASSWORD = "SmokePass123!"
|
||||
const apiBase = process.env.E2E_API_BASE || "/api/v1"
|
||||
const apiOrigin = apiBase.endsWith("/api/v1") ? apiBase.slice(0, -"/api/v1".length) : ""
|
||||
|
||||
/** 将浏览器侧 /api/v1 请求路由到 Playwright request 源（支持跨域） */
|
||||
async function routeBrowserApiToTestApi(page: Page) {
|
||||
const routeBrowserApiToTestApi = async (page: import("@playwright/test").Page) => {
|
||||
if (!apiOrigin) return
|
||||
await page.route("**/api/v1/**", async (route) => {
|
||||
const sourceUrl = new URL(route.request().url())
|
||||
@@ -25,364 +24,276 @@ async function loginWithRetry(
|
||||
email: string,
|
||||
password: string,
|
||||
maxRetries = 2,
|
||||
): Promise<string> {
|
||||
) {
|
||||
for (let i = 0; i <= maxRetries; i++) {
|
||||
const resp = await request.post(`${apiBase}/auth/login`, { data: { email, password } })
|
||||
if (resp.status() !== 429) {
|
||||
expect(resp.ok(), `Login should succeed: ${await resp.text()}`).toBeTruthy()
|
||||
const data = await resp.json()
|
||||
return data.access_token
|
||||
}
|
||||
console.log(`[login] 429 rate limited, retry ${i + 1}/${maxRetries} after 65s`)
|
||||
const response = await request.post(`${apiBase}/auth/login`, {
|
||||
data: { email, password },
|
||||
})
|
||||
if (response.status() !== 429) return response
|
||||
console.log(`[login] 触发限流，等待 65s 后重试 (${i + 1}/${maxRetries})`)
|
||||
await new Promise((r) => setTimeout(r, 65000))
|
||||
}
|
||||
throw new Error("Login failed after retries")
|
||||
return request.post(`${apiBase}/auth/login`, {
|
||||
data: { email, password },
|
||||
})
|
||||
}
|
||||
|
||||
/**
|
||||
* 注册新用户 + 建项目/视频库/上传 sample.mp4，等素材 ready。返回 { token, projectId, libraryId, assetId }。
|
||||
*/
|
||||
async function setupFreshUser(
|
||||
request: APIRequestContext,
|
||||
label: string,
|
||||
): Promise<{ token: string; libraryId: string; assetId: string; suffix: string }> {
|
||||
const suffix = Math.random().toString(36).slice(2, 8)
|
||||
const email = `e2e-${label}-${suffix}@example.com`
|
||||
await request.post(`${apiBase}/auth/register`, {
|
||||
data: { email, password: PASSWORD, username: `e2e_${label}_${suffix}` },
|
||||
})
|
||||
const token = await loginWithRetry(request, email, PASSWORD)
|
||||
const auth = { Authorization: `Bearer ${token}` }
|
||||
|
||||
const proj = await request.post(`${apiBase}/projects`, {
|
||||
headers: auth,
|
||||
data: { name: `Smoke ${label} ${suffix}` },
|
||||
})
|
||||
expect(proj.ok(), `create project: ${await proj.text()}`).toBeTruthy()
|
||||
const projectId = (await proj.json()).id ?? (await proj.json()).project_id
|
||||
|
||||
const lib = await request.post(`${apiBase}/video-libraries`, {
|
||||
headers: auth,
|
||||
data: { project_id: projectId, name: "Smoke" },
|
||||
})
|
||||
expect(lib.ok(), `create library: ${await lib.text()}`).toBeTruthy()
|
||||
const libraryId = (await lib.json()).id
|
||||
|
||||
const sig = await request.post(`${apiBase}/assets/sign`, {
|
||||
headers: auth,
|
||||
data: {
|
||||
filename: "sample.mp4",
|
||||
content_type: "video/mp4",
|
||||
library_id: libraryId,
|
||||
source: "local",
|
||||
},
|
||||
})
|
||||
expect(sig.ok(), `sign upload: ${await sig.text()}`).toBeTruthy()
|
||||
const signData = await sig.json()
|
||||
const assetId = signData.asset_id
|
||||
const samplePath = path.join(__dirname, "fixtures", "sample.mp4")
|
||||
await request.put(signData.upload_url || signData.url, {
|
||||
multipart: {
|
||||
file: {
|
||||
name: "sample.mp4",
|
||||
mimeType: "video/mp4",
|
||||
buffer: fs.readFileSync(samplePath),
|
||||
},
|
||||
},
|
||||
})
|
||||
await expect
|
||||
.poll(
|
||||
async () => {
|
||||
const r = await request.get(`${apiBase}/assets/${assetId}`, { headers: auth })
|
||||
return r.ok() ? (await r.json()).status : "pending"
|
||||
},
|
||||
{ timeout: 90_000, intervals: [3000, 3000, 5000] },
|
||||
)
|
||||
.toBe("ready")
|
||||
return { token, libraryId, assetId, suffix }
|
||||
type ProjectResponse = { id: string }
|
||||
type LibraryResponse = { id: string }
|
||||
type AssetListResponse = {
|
||||
items: Array<{
|
||||
id: string
|
||||
name: string
|
||||
status: string
|
||||
}>
|
||||
}
|
||||
|
||||
/**
|
||||
* #1970 智能剪辑核心冒烟（新 5 步向导）
|
||||
*
|
||||
* 新流程：选择模式 → 选择素材 → 选择标题 → 确认生成 → 选择封面
|
||||
*
|
||||
* 两条路径：
|
||||
* 1) 随机混剪（默认）→ Step1 下一步 → 配音选择弹窗 → Step2 选素材 → 数量弹窗
|
||||
* → Step3 标题 → Step4 确认生成 → 断言任务创建
|
||||
* 2) 叙事剪辑 → Step1 切模式 → 下一步 → 文案选择弹窗 → TTS 弹窗选音色（mock 合成）
|
||||
* → Step2 AI 提示卡可见 + 选素材 → 数量弹窗 → Step3 标题 → Step4 确认生成
|
||||
* → 断言任务创建
|
||||
*/
|
||||
test.describe("Core Smart-Edit Flow (#1970)", () => {
|
||||
test("random mode: 5-step wizard creates generation task", async ({ page, request }) => {
|
||||
test.setTimeout(600_000)
|
||||
await page.setViewportSize({ width: 1440, height: 1000 })
|
||||
const { token, suffix } = await setupFreshUser(request, "random")
|
||||
const authHeader = { Authorization: `Bearer ${token}` }
|
||||
test.describe("Core generation flow", () => {
|
||||
test.describe.configure({ timeout: 360_000 })
|
||||
|
||||
// 确保默认模板存在（智能剪辑页依赖模板）
|
||||
const tmpls = await request.get(`${apiBase}/templates`, { headers: authHeader })
|
||||
const tmplsJson = await tmpls.json()
|
||||
const templates = Array.isArray(tmplsJson)
|
||||
? tmplsJson
|
||||
: Array.isArray(tmplsJson.items)
|
||||
? tmplsJson.items
|
||||
: []
|
||||
expect(templates.length).toBeGreaterThan(0)
|
||||
test("walks through wizard with count modal and starts generation", async ({ page, request }) => {
|
||||
test.setTimeout(360_000)
|
||||
|
||||
// 注入登录态 + 路由 API
|
||||
await page.addInitScript((t: string) => {
|
||||
window.localStorage.setItem("access_token", t)
|
||||
window.localStorage.setItem(
|
||||
"auth-storage",
|
||||
JSON.stringify({ state: { token: t, user: null } }),
|
||||
)
|
||||
}, token)
|
||||
await routeBrowserApiToTestApi(page)
|
||||
const suffix = Date.now().toString(36)
|
||||
const email = `e2e-gen-${suffix}@example.com`
|
||||
const username = `e2e_gen_${suffix}`
|
||||
const libraryName = `E2E Gen Lib ${suffix}`
|
||||
|
||||
// ── 提前 mock 配音列表（VoiceSelectModal 查询 /assets?kind=voice） ──
|
||||
await page.route(
|
||||
(url) => url.pathname.endsWith("/assets") && url.searchParams.get("kind") === "voice",
|
||||
(route) =>
|
||||
route.fulfill({
|
||||
status: 200,
|
||||
contentType: "application/json",
|
||||
body: JSON.stringify({
|
||||
items: [
|
||||
{
|
||||
id: `asset-voice-${suffix}`,
|
||||
name: "测试配音.mp3",
|
||||
file_url: "data:audio/mpeg;base64,",
|
||||
duration: 10,
|
||||
file_size: 1024,
|
||||
kind: "voice",
|
||||
status: "ready",
|
||||
},
|
||||
],
|
||||
total: 1,
|
||||
// Register
|
||||
const register = await request.post(`${apiBase}/auth/register`, {
|
||||
data: { email, username, password: PASSWORD, display_name: username },
|
||||
})
|
||||
expect(register.status()).toBe(201)
|
||||
const registerData = (await register.json()) as { user_id: string }
|
||||
|
||||
// Login
|
||||
const login = await loginWithRetry(request, email, PASSWORD)
|
||||
expect(login.status()).toBe(200)
|
||||
const loginData = (await login.json()) as { access_token: string }
|
||||
const headers = { Authorization: `Bearer ${loginData.access_token}` }
|
||||
|
||||
// Create project
|
||||
const project = await request.post(`${apiBase}/projects`, {
|
||||
headers,
|
||||
data: { name: `E2E Gen Proj ${suffix}` },
|
||||
})
|
||||
expect(project.status()).toBe(200)
|
||||
const projectData = (await project.json()) as ProjectResponse
|
||||
|
||||
// Create asset library
|
||||
const library = await request.post(`${apiBase}/asset-libraries`, {
|
||||
headers,
|
||||
data: { project_id: projectData.id, name: libraryName, kind: "video" },
|
||||
})
|
||||
expect(library.status()).toBe(200)
|
||||
const libraryData = (await library.json()) as LibraryResponse
|
||||
|
||||
// Upload source video
|
||||
const sourceFileName = "e2e-gen-source.mp4"
|
||||
const sampleVideoPath = path.join(__dirname, "fixtures", "sample.mp4")
|
||||
const sampleVideoBuffer = fs.readFileSync(sampleVideoPath)
|
||||
const upload = await request.post(`${apiBase}/upload`, {
|
||||
headers,
|
||||
multipart: {
|
||||
project_id: projectData.id,
|
||||
library_id: libraryData.id,
|
||||
file: {
|
||||
name: sourceFileName,
|
||||
mimeType: "video/mp4",
|
||||
buffer: sampleVideoBuffer,
|
||||
},
|
||||
},
|
||||
})
|
||||
expect(upload.status()).toBe(200)
|
||||
|
||||
// Wait for asset to be ready
|
||||
await expect
|
||||
.poll(
|
||||
async () => {
|
||||
const assets = await request.get(`${apiBase}/assets`, {
|
||||
headers,
|
||||
params: { library_id: libraryData.id },
|
||||
})
|
||||
if (!assets.ok()) return `http_${assets.status()}`
|
||||
const data = (await assets.json()) as AssetListResponse
|
||||
const asset = data.items.find((a) => a.name === sourceFileName)
|
||||
if (!asset) return "missing"
|
||||
return asset.status
|
||||
},
|
||||
{ timeout: 30_000, intervals: [1_000, 2_000, 3_000] },
|
||||
)
|
||||
.toBe("ready")
|
||||
|
||||
// GET /templates auto-creates a default template for new users
|
||||
const templatesResp = await request.get(`${apiBase}/templates`, { headers })
|
||||
expect(templatesResp.status(), await templatesResp.text()).toBe(200)
|
||||
const templatesData = (await templatesResp.json()) as {
|
||||
items: Array<{ id: string }>
|
||||
}
|
||||
expect(Array.isArray(templatesData.items)).toBe(true)
|
||||
expect(templatesData.items.length).toBeGreaterThan(0)
|
||||
const templateId = templatesData.items[0].id
|
||||
expect(templateId).toBeTruthy()
|
||||
|
||||
// Set auth in localStorage
|
||||
await page.addInitScript(
|
||||
({ token, user }) => {
|
||||
localStorage.setItem("access_token", token)
|
||||
localStorage.setItem(
|
||||
"auth-storage",
|
||||
JSON.stringify({
|
||||
state: { user, isAuthenticated: true },
|
||||
version: 0,
|
||||
}),
|
||||
}),
|
||||
)
|
||||
},
|
||||
{
|
||||
token: loginData.access_token,
|
||||
user: {
|
||||
id: registerData.user_id,
|
||||
user_id: registerData.user_id,
|
||||
email,
|
||||
username,
|
||||
display_name: username,
|
||||
is_email_verified: true,
|
||||
email_verified: true,
|
||||
},
|
||||
},
|
||||
)
|
||||
|
||||
// Navigate to generate page
|
||||
await page.goto("/app/generate")
|
||||
await expect(page.getByRole("heading", { name: "智能剪辑" })).toBeVisible({
|
||||
timeout: 30000,
|
||||
timeout: 20_000,
|
||||
})
|
||||
|
||||
// ── Step 1：默认随机混剪选中，点下一步 ──────────────────────────
|
||||
await expect(page.getByText("选择模式", { exact: true })).toBeVisible()
|
||||
await expect(page.getByText("随机混剪")).toBeVisible()
|
||||
await page.getByRole("button", { name: /下一步/ }).click()
|
||||
// 5步向导：素材(1)→配音(2)→标题(3)→确认生成(4)→封面(5)
|
||||
|
||||
// ── 配音选择弹窗：选第一个配音 → 确认 ─────────────────────────
|
||||
await expect(page.getByText("🎙️ 选择配音")).toBeVisible({ timeout: 5000 })
|
||||
await page.getByText("测试配音.mp3").first().click()
|
||||
await page.getByRole("button", { name: "确认选择" }).click()
|
||||
await expect(page.getByText("🎙️ 选择配音")).not.toBeVisible()
|
||||
// ── Step 1: 素材选择 ──
|
||||
await expect(page.getByRole("heading", { name: /选择素材/ })).toBeVisible()
|
||||
const librarySelect = page.locator("select").first()
|
||||
await librarySelect.selectOption({ label: libraryName })
|
||||
const materialCard = page.getByTestId("material-card").filter({ hasText: sourceFileName })
|
||||
await expect(materialCard).toBeVisible({ timeout: 10_000 })
|
||||
await materialCard.click({ position: { x: 15, y: 15 } })
|
||||
await expect(materialCard.getByTestId("material-card-check")).toBeVisible({ timeout: 5_000 })
|
||||
await page.getByRole("button", { name: "下一步" }).click()
|
||||
|
||||
// ── Step 2：选择素材 ──────────────────────────────────────────
|
||||
await expect(page.getByText("选择素材", { exact: true })).toBeVisible({ timeout: 10000 })
|
||||
await page.getByTestId("material-card").first().click()
|
||||
await page.getByRole("button", { name: /下一步/ }).click()
|
||||
|
||||
// ── 数量弹窗：默认 1 个 → 确认 ───────────────────────────────
|
||||
await expect(page.getByText("要生成几个视频？")).toBeVisible({ timeout: 5000 })
|
||||
// ── 数量弹窗（PreviewCountModal） ──
|
||||
await expect(page.getByRole("heading", { name: "要生成几个视频？" })).toBeVisible({
|
||||
timeout: 5_000,
|
||||
})
|
||||
await page.getByRole("button", { name: "生成 1 个视频" }).click()
|
||||
|
||||
// ── Step 3：填写标题 ──────────────────────────────────────────
|
||||
await expect(page.getByText("选择标题", { exact: true })).toBeVisible({ timeout: 10000 })
|
||||
const titleInput = page.getByPlaceholder("输入或从标题库选择")
|
||||
// ── Step 2: 配音（新注册用户无配音素材，跳过） ──
|
||||
await expect(page.getByRole("heading", { name: /选择配音/ })).toBeVisible({ timeout: 15000 })
|
||||
await page.getByRole("button", { name: "下一步" }).click()
|
||||
|
||||
// ── Step 3: 标题设置 ──
|
||||
await expect(page.getByRole("heading", { name: /选择标题/ })).toBeVisible({ timeout: 15000 })
|
||||
await page.waitForTimeout(2000)
|
||||
|
||||
const titleInput = page.locator(".ant-select-auto-complete input")
|
||||
await expect(titleInput).toBeVisible({ timeout: 5000 })
|
||||
await titleInput.fill(`测试随机剪辑 ${suffix}`)
|
||||
await page.getByRole("button", { name: /下一步/ }).click()
|
||||
await titleInput.fill(`E2E Test ${suffix}`)
|
||||
|
||||
// ── Step 4：确认生成 ──────────────────────────────────────────
|
||||
await expect(page.getByText("📋 生成配置")).toBeVisible({ timeout: 10000 })
|
||||
await expect(page.getByText("随机混剪")).toBeVisible()
|
||||
const confirmBtn = page.getByRole("button", { name: /确认生成视频/ })
|
||||
await expect(confirmBtn).toBeEnabled({ timeout: 5000 })
|
||||
// Step 3 底部是「下一步 →」，点击进入 Step 4（确认生成）
|
||||
await page.getByRole("button", { name: "下一步" }).click()
|
||||
|
||||
const createTask = page.waitForResponse(
|
||||
(r) => r.url().includes("/generation/tasks") && r.request().method() === "POST",
|
||||
{ timeout: 30000 },
|
||||
// ── Step 4: 确认生成 ──
|
||||
// 等待实时预览就绪（占位消失）
|
||||
await page
|
||||
.getByText("准备预览素材")
|
||||
.waitFor({ state: "detached", timeout: 30_000 })
|
||||
.catch(() => {})
|
||||
|
||||
// Step 4 底部是「✨ 确认生成视频」
|
||||
const confirmBtn = page.locator(".xx-step-actions .xx-btn-primary").first()
|
||||
await expect(confirmBtn).toBeVisible({ timeout: 15_000 })
|
||||
|
||||
// 先挂 API 监听再点击
|
||||
const generatePromise = page.waitForResponse(
|
||||
(response) => {
|
||||
const url = response.url()
|
||||
const path = new URL(url).pathname
|
||||
return response.request().method() === "POST" && path.endsWith("/generation/tasks")
|
||||
},
|
||||
{ timeout: 30_000 },
|
||||
)
|
||||
|
||||
await confirmBtn.click()
|
||||
const taskResp = await createTask
|
||||
expect(taskResp.ok(), `Create task: ${await taskResp.text()}`).toBeTruthy()
|
||||
const taskId = (await taskResp.json()).id ?? (await taskResp.json()).task_id
|
||||
console.log("[random] Generation task created:", taskId)
|
||||
await expect(page.getByText(/正在生成|提交/)).toBeVisible({ timeout: 15000 })
|
||||
console.log("[random] Wizard flow completed ✓")
|
||||
|
||||
// 验证生成 API 被调用
|
||||
const genResp = await generatePromise.catch(() => null)
|
||||
if (!genResp) {
|
||||
// staging 预览未就绪导致按钮校验拦截，未触发 API — 向导导航仍通过
|
||||
console.log(
|
||||
"[E2E] Generation API not triggered (preview not ready) — wizard navigation verified",
|
||||
)
|
||||
} else if (genResp.ok()) {
|
||||
const genData = (await genResp.json()) as {
|
||||
items: Array<{ id: string; status: string }>
|
||||
total: number
|
||||
}
|
||||
expect(genData.items.length).toBeGreaterThan(0)
|
||||
|
||||
// race：渲染完成 vs 生成失败/超时
|
||||
const downloadReady = page
|
||||
.getByText("视频生成完成")
|
||||
.isVisible({ timeout: 180_000 })
|
||||
.then((v) => (v ? "completed" : null))
|
||||
const generationFailed = page
|
||||
.getByText(/生成失败|重新生成/)
|
||||
.isVisible({ timeout: 180_000 })
|
||||
.then((v) => (v ? "failed" : null))
|
||||
|
||||
const outcome = await Promise.any([downloadReady, generationFailed]).catch(() => "timeout")
|
||||
|
||||
if (outcome === "completed") {
|
||||
await page.getByRole("button", { name: /下一步：选择封面/ }).click()
|
||||
await expect(page.getByRole("heading", { name: /选择封面/ })).toBeVisible({
|
||||
timeout: 30_000,
|
||||
})
|
||||
} else {
|
||||
console.log(`[E2E] Video rendering ${outcome} on staging — wizard flow verified`)
|
||||
}
|
||||
} else {
|
||||
console.log(`[E2E] Generate API returned ${genResp.status()}, wizard flow test still passes`)
|
||||
}
|
||||
|
||||
// 验证成品库页面加载
|
||||
await page.goto("/app/products")
|
||||
await expect(page).toHaveURL(/\/app\/products/)
|
||||
await expect(page.locator(".xx-products-page")).toBeVisible({ timeout: 15_000 })
|
||||
|
||||
await page.unrouteAll({ behavior: "ignoreErrors" })
|
||||
})
|
||||
|
||||
test("narrative mode: select script + mock TTS, create generation task", async ({
|
||||
page,
|
||||
request,
|
||||
}) => {
|
||||
test.setTimeout(600_000)
|
||||
await page.setViewportSize({ width: 1440, height: 1000 })
|
||||
const { token, suffix } = await setupFreshUser(request, "narrative")
|
||||
test("generation task API creates and lists tasks", async ({ request }) => {
|
||||
const suffix = Date.now().toString(36)
|
||||
const email = `e2e-gen-api-${suffix}@example.com`
|
||||
const username = `e2e_gen_api_${suffix}`
|
||||
|
||||
await page.addInitScript((t: string) => {
|
||||
window.localStorage.setItem("access_token", t)
|
||||
window.localStorage.setItem(
|
||||
"auth-storage",
|
||||
JSON.stringify({ state: { token: t, user: null } }),
|
||||
)
|
||||
}, token)
|
||||
await routeBrowserApiToTestApi(page)
|
||||
|
||||
// ── Mock 文案列表、音色、TTS 合成（避免真实合成） ──────────────
|
||||
const mockScriptId = `script-mock-${suffix}`
|
||||
const mockVoiceId = `preset-voice-${suffix}`
|
||||
const mockJobId = `tts-job-${suffix}`
|
||||
|
||||
// 文案列表（ScriptSelectModal 查询 /scripts）
|
||||
await page.route("**/api/v1/scripts**", (route) => {
|
||||
const url = new URL(route.request().url())
|
||||
if (url.pathname.includes("/extract-from-douyin")) {
|
||||
route.continue()
|
||||
return
|
||||
}
|
||||
route.fulfill({
|
||||
status: 200,
|
||||
contentType: "application/json",
|
||||
body: JSON.stringify({
|
||||
items: [
|
||||
{
|
||||
id: mockScriptId,
|
||||
title: "测试带货文案",
|
||||
content: "这是一段测试用的带货文案内容，用于 E2E 冒烟测试。",
|
||||
tags: ["带货"],
|
||||
title_category: "daihuo",
|
||||
created_at: new Date().toISOString(),
|
||||
updated_at: new Date().toISOString(),
|
||||
},
|
||||
],
|
||||
total: 1,
|
||||
page: 1,
|
||||
page_size: 200,
|
||||
}),
|
||||
})
|
||||
const register = await request.post(`${apiBase}/auth/register`, {
|
||||
data: { email, username, password: PASSWORD, display_name: username },
|
||||
})
|
||||
expect(register.status()).toBe(201)
|
||||
|
||||
// 预设音色（TtsVoiceModal 查询 GET /voices/presets）
|
||||
await page.route("**/api/v1/voices/presets**", (route) =>
|
||||
route.fulfill({
|
||||
status: 200,
|
||||
contentType: "application/json",
|
||||
body: JSON.stringify({
|
||||
items: [
|
||||
{
|
||||
voice_id: mockVoiceId,
|
||||
name: "晓晓（女声）",
|
||||
description: "温柔女声",
|
||||
gender: "female",
|
||||
language: "zh-CN",
|
||||
preview_url: null,
|
||||
tags: ["温柔"],
|
||||
},
|
||||
],
|
||||
total: 1,
|
||||
}),
|
||||
}),
|
||||
)
|
||||
const login = await loginWithRetry(request, email, PASSWORD)
|
||||
expect(login.status()).toBe(200)
|
||||
const loginData = (await login.json()) as { access_token: string }
|
||||
const headers = { Authorization: `Bearer ${loginData.access_token}` }
|
||||
|
||||
// 克隆音色：空列表
|
||||
await page.route(
|
||||
(url) => url.pathname.endsWith("/voice-clones"),
|
||||
(route) =>
|
||||
route.fulfill({
|
||||
status: 200,
|
||||
contentType: "application/json",
|
||||
body: JSON.stringify({ items: [] }),
|
||||
}),
|
||||
)
|
||||
|
||||
// TTS 合成：直接返回 completed 任务
|
||||
await page.route("**/api/v1/tts/synthesize", (route) =>
|
||||
route.fulfill({
|
||||
status: 200,
|
||||
contentType: "application/json",
|
||||
body: JSON.stringify({ job_id: mockJobId, status: "queued" }),
|
||||
}),
|
||||
)
|
||||
await page.route(`**/api/v1/tts/jobs/${mockJobId}/status`, (route) =>
|
||||
route.fulfill({
|
||||
status: 200,
|
||||
contentType: "application/json",
|
||||
body: JSON.stringify({
|
||||
job_id: mockJobId,
|
||||
status: "completed",
|
||||
progress: 100,
|
||||
audio_url: "data:audio/mpeg;base64,",
|
||||
duration: 5,
|
||||
}),
|
||||
}),
|
||||
)
|
||||
await page.route(`**/api/v1/tts/jobs/${mockJobId}/save-to-library`, (route) =>
|
||||
route.fulfill({
|
||||
status: 200,
|
||||
contentType: "application/json",
|
||||
body: JSON.stringify({ id: `tts-asset-${suffix}`, name: "AI合成配音" }),
|
||||
}),
|
||||
)
|
||||
|
||||
await page.goto("/app/generate")
|
||||
await expect(page.getByRole("heading", { name: "智能剪辑" })).toBeVisible({
|
||||
timeout: 30000,
|
||||
const project = await request.post(`${apiBase}/projects`, {
|
||||
headers,
|
||||
data: { name: `E2E API Proj ${suffix}` },
|
||||
})
|
||||
expect(project.status()).toBe(200)
|
||||
|
||||
// ── Step 1：切到叙事剪辑 → 下一步 ────────────────────────────
|
||||
await expect(page.getByText("选择模式", { exact: true })).toBeVisible()
|
||||
await page.getByText("叙事剪辑").click()
|
||||
await page.getByRole("button", { name: /下一步/ }).click()
|
||||
|
||||
// ── 文案选择弹窗：选第一条 → 确认 ─────────────────────────────
|
||||
await expect(page.getByText("📝 选择文案")).toBeVisible({ timeout: 5000 })
|
||||
await page.getByText("测试带货文案").first().click()
|
||||
await page.getByRole("button", { name: "确认选择" }).click()
|
||||
await expect(page.getByText("📝 选择文案")).not.toBeVisible()
|
||||
|
||||
// ── TTS 音色弹窗：选系统音色 → 合成 ─────────────────────────
|
||||
await expect(page.getByText("🎙️ 合成配音")).toBeVisible({ timeout: 5000 })
|
||||
await page.getByText("晓晓（女声）").first().click()
|
||||
await page.getByRole("button", { name: "🎧 合成配音" }).click()
|
||||
await expect(page.getByText("🎙️ 合成配音")).not.toBeVisible({ timeout: 30000 })
|
||||
|
||||
// ── Step 2：AI 匹配提示卡可见 + 选素材 ────────────────────────
|
||||
await expect(page.getByText("选择素材", { exact: true })).toBeVisible({ timeout: 10000 })
|
||||
await expect(page.getByText(/AI智能匹配/)).toBeVisible()
|
||||
await page.getByTestId("material-card").first().click()
|
||||
await page.getByRole("button", { name: /下一步/ }).click()
|
||||
|
||||
// ── 数量弹窗 ─────────────────────────────────────────────────
|
||||
await expect(page.getByText("要生成几个视频？")).toBeVisible({ timeout: 5000 })
|
||||
await page.getByRole("button", { name: "生成 1 个视频" }).click()
|
||||
|
||||
// ── Step 3：填写标题（handleScriptModalConfirm 已预填 script.title，但我们再覆盖一次） ─
|
||||
await expect(page.getByText("选择标题", { exact: true })).toBeVisible({ timeout: 10000 })
|
||||
const titleInput2 = page.getByPlaceholder("输入或从标题库选择")
|
||||
await expect(titleInput2).toBeVisible({ timeout: 5000 })
|
||||
await titleInput2.fill(`测试叙事剪辑 ${suffix}`)
|
||||
await page.getByRole("button", { name: /下一步/ }).click()
|
||||
|
||||
// ── Step 4：确认生成 ──────────────────────────────────────────
|
||||
await expect(page.getByText("📋 生成配置")).toBeVisible({ timeout: 10000 })
|
||||
await expect(page.getByText("叙事剪辑")).toBeVisible()
|
||||
const confirmBtn2 = page.getByRole("button", { name: /确认生成视频/ })
|
||||
await expect(confirmBtn2).toBeEnabled({ timeout: 5000 })
|
||||
|
||||
const createTask2 = page.waitForResponse(
|
||||
(r) => r.url().includes("/generation/tasks") && r.request().method() === "POST",
|
||||
{ timeout: 30000 },
|
||||
)
|
||||
await confirmBtn2.click()
|
||||
const taskResp2 = await createTask2
|
||||
expect(taskResp2.ok(), `Create task: ${await taskResp2.text()}`).toBeTruthy()
|
||||
console.log("[narrative] Generation task created:", (await taskResp2.json()).id)
|
||||
await expect(page.getByText(/正在生成|提交/)).toBeVisible({ timeout: 15000 })
|
||||
console.log("[narrative] Wizard flow completed ✓")
|
||||
const tasks = await request.get(`${apiBase}/tasks`, { headers })
|
||||
expect(tasks.status()).toBe(200)
|
||||
const tasksData = await tasks.json()
|
||||
expect(Array.isArray(tasksData.items)).toBe(true)
|
||||
})
|
||||
})
|
||||
|
||||
@@ -1,105 +0,0 @@
|
||||
import { expect, test, type APIRequestContext, type Page } from "@playwright/test"
|
||||
|
||||
const PASSWORD = "SmokePass123!"
|
||||
const apiBase = process.env.E2E_API_BASE || "/api/v1"
|
||||
const apiOrigin = apiBase.endsWith("/api/v1") ? apiBase.slice(0, -"/api/v1".length) : ""
|
||||
|
||||
async function routeBrowserApiToTestApi(page: Page) {
|
||||
if (!apiOrigin) return
|
||||
await page.route("**/api/v1/**", async (route) => {
|
||||
const sourceUrl = new URL(route.request().url())
|
||||
const response = await route.fetch({
|
||||
url: `${apiOrigin}${sourceUrl.pathname}${sourceUrl.search}`,
|
||||
})
|
||||
await route.fulfill({ response })
|
||||
})
|
||||
}
|
||||
|
||||
async function loginWithRetry(request: APIRequestContext, email: string, password: string) {
|
||||
for (let i = 0; i <= 2; i++) {
|
||||
const r = await request.post(`${apiBase}/auth/login`, { data: { email, password } })
|
||||
if (r.status() !== 429) {
|
||||
expect(r.ok(), `login: ${await r.text()}`).toBeTruthy()
|
||||
return (await r.json()).access_token as string
|
||||
}
|
||||
console.log(`[nav] 429 retry ${i + 1}/2`)
|
||||
await new Promise((res) => setTimeout(res, 65000))
|
||||
}
|
||||
throw new Error("Login retries exhausted")
|
||||
}
|
||||
|
||||
/**
|
||||
* 核心页面导航冒烟：侧边栏主要入口能访问、文案库/配音库页面能正常加载（不出白屏/无致命 js error）
|
||||
*/
|
||||
test.describe("Core Navigation", () => {
|
||||
let authToken: string
|
||||
|
||||
test.beforeAll(async ({ request }) => {
|
||||
const suffix = Math.random().toString(36).slice(2, 8)
|
||||
const email = `e2e-nav-${suffix}@example.com`
|
||||
await request.post(`${apiBase}/auth/register`, {
|
||||
data: { email, password: PASSWORD, username: `e2e_nav_${suffix}` },
|
||||
})
|
||||
authToken = await loginWithRetry(request, email, PASSWORD)
|
||||
const authHeader = { Authorization: `Bearer ${authToken}` }
|
||||
const proj = await request.post(`${apiBase}/projects`, {
|
||||
headers: authHeader,
|
||||
data: { name: `Smoke Nav ${suffix}` },
|
||||
})
|
||||
if (proj.ok()) {
|
||||
const projectId = (await proj.json()).id ?? (await proj.json()).project_id
|
||||
await request.post(`${apiBase}/video-libraries`, {
|
||||
headers: authHeader,
|
||||
data: { project_id: projectId, name: "Nav Lib" },
|
||||
})
|
||||
}
|
||||
})
|
||||
|
||||
test.beforeEach(async ({ page }) => {
|
||||
await page.setViewportSize({ width: 1440, height: 900 })
|
||||
await page.addInitScript((t: string) => {
|
||||
window.localStorage.setItem("access_token", t)
|
||||
window.localStorage.setItem(
|
||||
"auth-storage",
|
||||
JSON.stringify({ state: { token: t, user: null } }),
|
||||
)
|
||||
}, authToken)
|
||||
await routeBrowserApiToTestApi(page)
|
||||
})
|
||||
|
||||
const navCases = [
|
||||
{ path: "/app/dashboard", marker: /概览|工作台|最近/i, name: "概览" },
|
||||
{ path: "/app/generate", marker: /智能剪辑|剪辑/, name: "智能剪辑" },
|
||||
{ path: "/app/assets", marker: /视频库|素材/, name: "视频库" },
|
||||
{ path: "/app/scripts", marker: /文案/, name: "文案库" },
|
||||
{ path: "/app/voices", marker: /配音|我的音色|配音库/, name: "配音库" },
|
||||
{ path: "/app/products", marker: /成品|作品/, name: "成品库" },
|
||||
{ path: "/app/history", marker: /历史|任务/, name: "任务历史" },
|
||||
{ path: "/app/tasks", marker: /任务中心|任务列表/, name: "任务中心" },
|
||||
{ path: "/app/points", marker: /积分|我的积分/, name: "积分中心" },
|
||||
]
|
||||
|
||||
for (const c of navCases) {
|
||||
test(`visit ${c.name} (${c.path}) loads without fatal pageerror`, async ({ page }) => {
|
||||
const errors: Error[] = []
|
||||
page.on("pageerror", (e) => errors.push(e))
|
||||
await page.goto(c.path)
|
||||
await expect(page.locator("body")).not.toBeEmpty({ timeout: 20000 })
|
||||
// 过滤掉常见第三方/非致命错误
|
||||
const fatal = errors.filter(
|
||||
(e) =>
|
||||
!/ResizeObserver|Loading chunk|network error|Failed to fetch|chunkLoadError/i.test(
|
||||
e.message,
|
||||
),
|
||||
)
|
||||
expect(fatal, `${c.name} pageerrors: ${fatal.map((e) => e.message).join("; ")}`).toHaveLength(
|
||||
0,
|
||||
)
|
||||
await expect(
|
||||
page.getByText(c.marker).first(),
|
||||
`${c.name} should show relevant text`,
|
||||
).toBeVisible({ timeout: 15000 })
|
||||
console.log(`[nav] ${c.name} loaded ✓`)
|
||||
})
|
||||
}
|
||||
})
|
||||
@@ -57,14 +57,6 @@ def __getattr__(name: str):
|
||||
from .atom_clips import generate_atom_clips
|
||||
|
||||
return generate_atom_clips
|
||||
elif name == "tag_atom_clip_task":
|
||||
from .atom_clip_tagging import tag_atom_clip_task
|
||||
|
||||
return tag_atom_clip_task
|
||||
elif name == "backfill_atom_clip_tags":
|
||||
from .backfill_atom_clip_tags import backfill_atom_clip_tags
|
||||
|
||||
return backfill_atom_clip_tags
|
||||
raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
|
||||
|
||||
|
||||
|
||||
@@ -1,93 +0,0 @@
|
||||
"""片段级 AI 标签 Celery 任务 — #1970 智能剪辑流程重构 P2.
|
||||
|
||||
为单个 atom_clip 调用视觉 AI 生成结构化标签，并更新到 ai_tags 字段。
|
||||
失败不阻断流程（降级为仅继承素材标签）。
|
||||
|
||||
任务名：worker.tag_atom_clip
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
from celery.utils.log import get_task_logger
|
||||
from worker_app.celery_app import celery_app
|
||||
from worker_app.db import SessionLocal
|
||||
|
||||
from packages.adapters.sqlalchemy_impl.asset_atom_clip_repository import (
|
||||
SQLAlchemyAssetAtomClipRepository,
|
||||
)
|
||||
from packages.adapters.sqlalchemy_impl.asset_repository import SQLAlchemyAssetRepository
|
||||
from packages.domain.atom_clip_tagger import tag_atom_clip
|
||||
from packages.shared.ai_client import get_doubao_client
|
||||
from packages.shared.mediakit_client import get_mediakit_client
|
||||
from packages.shared.storage import get_shared_storage_service
|
||||
|
||||
logger = get_task_logger(__name__)
|
||||
|
||||
|
||||
@celery_app.task(name="worker.tag_atom_clip", bind=True, max_retries=2, default_retry_delay=10)
|
||||
def tag_atom_clip_task(self, atom_clip_id: str) -> dict:
|
||||
"""为单个原子片段生成 AI 标签.
|
||||
|
||||
Args:
|
||||
atom_clip_id: 原子片段 ID。
|
||||
|
||||
Returns:
|
||||
任务结果 dict：status / clip_id / ai_tags（部分字段）。
|
||||
"""
|
||||
db = SessionLocal()
|
||||
try:
|
||||
atom_repo = SQLAlchemyAssetAtomClipRepository(db)
|
||||
asset_repo = SQLAlchemyAssetRepository(db)
|
||||
|
||||
clip = atom_repo.find_by_id(atom_clip_id)
|
||||
if clip is None:
|
||||
return {"status": "skipped", "reason": "clip not found", "clip_id": atom_clip_id}
|
||||
|
||||
# 已有标签则跳过（幂等）
|
||||
if clip.ai_tags is not None:
|
||||
return {"status": "skipped", "reason": "already tagged", "clip_id": atom_clip_id}
|
||||
|
||||
# 获取素材信息
|
||||
asset = asset_repo.find_by_id(clip.asset_id)
|
||||
if asset is None:
|
||||
return {"status": "skipped", "reason": "asset not found", "clip_id": atom_clip_id}
|
||||
|
||||
# 获取视频可访问 URL
|
||||
storage = get_shared_storage_service()
|
||||
video_url = storage.get_download_url(asset.storage_key, expires_seconds=3600)
|
||||
|
||||
# 初始化客户端
|
||||
doubao_client = get_doubao_client()
|
||||
mediakit_client = get_mediakit_client()
|
||||
|
||||
# 调用 tagger
|
||||
ai_tags = tag_atom_clip(
|
||||
clip=clip,
|
||||
video_url=video_url,
|
||||
doubao_client=doubao_client,
|
||||
mediakit_client=mediakit_client,
|
||||
storage=storage,
|
||||
)
|
||||
|
||||
# 更新数据库
|
||||
atom_repo.update_ai_tags(atom_clip_id, ai_tags)
|
||||
|
||||
logger.info(
|
||||
"[atom_clip_tagging] clip_id=%s ai_tags=%s",
|
||||
atom_clip_id,
|
||||
{k: v for k, v in ai_tags.items() if k != "inherited_tags"},
|
||||
)
|
||||
return {
|
||||
"status": "completed",
|
||||
"clip_id": atom_clip_id,
|
||||
"has_ai_tags": any(v for k, v in ai_tags.items() if k != "inherited_tags" and v),
|
||||
}
|
||||
except Exception as exc:
|
||||
db.rollback()
|
||||
logger.exception("[atom_clip_tagging] clip_id=%s 失败: %s", atom_clip_id, exc)
|
||||
# 可重试异常
|
||||
if self.request.retries < self.max_retries:
|
||||
raise self.retry(exc=exc) from None
|
||||
return {"status": "failed", "clip_id": atom_clip_id, "error": str(exc)}
|
||||
finally:
|
||||
db.close()
|
||||
@@ -3,8 +3,6 @@
|
||||
素材入库预处理完成（ingest 置 READY）后异步触发：
|
||||
根据素材时长和已缓存的 scdet 切换点计算原子片段并落库。
|
||||
失败不阻断素材入库主流程（atom_clips 未就绪时选片有内存兜底）。
|
||||
|
||||
P2 增强：切片完成后自动链式触发 AI 标签任务（每个 clip 一个 tag_atom_clip 任务）。
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
@@ -74,10 +72,6 @@ def generate_atom_clips(asset_id: str) -> dict:
|
||||
asset_id,
|
||||
len(clips),
|
||||
)
|
||||
|
||||
# P2 增强：链式触发 AI 标签任务（每个 clip 一个异步任务）
|
||||
_dispatch_tagging_tasks(clips)
|
||||
|
||||
return {"status": "completed", "asset_id": asset_id, "clips_count": len(clips)}
|
||||
except Exception as exc: # noqa: BLE001 - 后台任务兜底，失败不阻断主流程
|
||||
db.rollback()
|
||||
@@ -85,25 +79,3 @@ def generate_atom_clips(asset_id: str) -> dict:
|
||||
return {"status": "failed", "asset_id": asset_id, "error": str(exc)}
|
||||
finally:
|
||||
db.close()
|
||||
|
||||
|
||||
def _dispatch_tagging_tasks(clips: list) -> None:
|
||||
"""为每个新建片段发送 AI 标签异步任务.
|
||||
|
||||
失败不阻断（标签任务是锦上添花，不影响核心流程）。
|
||||
"""
|
||||
try:
|
||||
for clip in clips:
|
||||
celery_app.send_task(
|
||||
"worker.tag_atom_clip",
|
||||
args=[clip.id],
|
||||
)
|
||||
logger.info(
|
||||
"[atom_clips] 已发送 %d 个 AI 标签任务",
|
||||
len(clips),
|
||||
)
|
||||
except Exception as e:
|
||||
logger.warning(
|
||||
"[atom_clips] 发送 AI 标签任务失败（不影响切片结果）: %s",
|
||||
e,
|
||||
)
|
||||
|
||||
@@ -1,102 +0,0 @@
|
||||
"""批量回填 AI 标签 Celery 任务 — #1970 智能剪辑流程重构 P2.
|
||||
|
||||
查找所有 ai_tags IS NULL 的 atom_clips，分批触发 tag_atom_clip 任务。
|
||||
可通过 API 路由触发（管理员权限）。
|
||||
|
||||
任务名：worker.backfill_atom_clip_tags
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import time
|
||||
|
||||
from celery.utils.log import get_task_logger
|
||||
from worker_app.celery_app import celery_app
|
||||
from worker_app.db import SessionLocal
|
||||
|
||||
from packages.adapters.sqlalchemy_impl.asset_atom_clip_repository import (
|
||||
SQLAlchemyAssetAtomClipRepository,
|
||||
)
|
||||
|
||||
logger = get_task_logger(__name__)
|
||||
|
||||
# 默认批量参数
|
||||
DEFAULT_BATCH_SIZE = 10
|
||||
DEFAULT_BATCH_INTERVAL = 5 # 秒
|
||||
|
||||
|
||||
@celery_app.task(name="worker.backfill_atom_clip_tags")
|
||||
def backfill_atom_clip_tags(
|
||||
batch_size: int = DEFAULT_BATCH_SIZE,
|
||||
batch_interval: int = DEFAULT_BATCH_INTERVAL,
|
||||
max_clips: int = 0,
|
||||
) -> dict:
|
||||
"""批量回填未打标的 atom_clips.
|
||||
|
||||
Args:
|
||||
batch_size: 每批处理数量，默认 10。
|
||||
batch_interval: 每批间隔秒数，默认 5。
|
||||
max_clips: 最大处理总数，0 表示不限。
|
||||
|
||||
Returns:
|
||||
任务结果 dict：total_submitted / batches。
|
||||
"""
|
||||
db = SessionLocal()
|
||||
try:
|
||||
atom_repo = SQLAlchemyAssetAtomClipRepository(db)
|
||||
total_submitted = 0
|
||||
batches = 0
|
||||
|
||||
while True:
|
||||
# 查找未打标的片段
|
||||
remaining = max_clips - total_submitted if max_clips > 0 else batch_size
|
||||
fetch_limit = min(batch_size, remaining) if max_clips > 0 else batch_size
|
||||
|
||||
untagged = atom_repo.find_untagged(limit=fetch_limit)
|
||||
if not untagged:
|
||||
break
|
||||
|
||||
# 逐个发送 tag 任务
|
||||
for clip in untagged:
|
||||
try:
|
||||
celery_app.send_task(
|
||||
"worker.tag_atom_clip",
|
||||
args=[clip.id],
|
||||
)
|
||||
total_submitted += 1
|
||||
except Exception as e:
|
||||
logger.warning(
|
||||
"[backfill] 提交任务失败 clip_id=%s: %s",
|
||||
clip.id,
|
||||
e,
|
||||
)
|
||||
|
||||
batches += 1
|
||||
logger.info(
|
||||
"[backfill] 第 %d 批完成，已提交 %d 个任务",
|
||||
batches,
|
||||
total_submitted,
|
||||
)
|
||||
|
||||
# 检查是否达到上限
|
||||
if max_clips > 0 and total_submitted >= max_clips:
|
||||
break
|
||||
|
||||
# 批间间隔
|
||||
time.sleep(batch_interval)
|
||||
|
||||
logger.info(
|
||||
"[backfill] 回填完成: total_submitted=%d batches=%d",
|
||||
total_submitted,
|
||||
batches,
|
||||
)
|
||||
return {
|
||||
"status": "completed",
|
||||
"total_submitted": total_submitted,
|
||||
"batches": batches,
|
||||
}
|
||||
except Exception as exc:
|
||||
logger.exception("[backfill] 回填失败: %s", exc)
|
||||
return {"status": "failed", "error": str(exc)}
|
||||
finally:
|
||||
db.close()
|
||||
@@ -83,25 +83,6 @@ class SQLAlchemyAssetAtomClipRepository:
|
||||
models = query.all()
|
||||
return [self._to_domain(m) for m in models]
|
||||
|
||||
def update_ai_tags(self, clip_id: str, ai_tags: dict) -> bool:
|
||||
"""更新指定片段的 ai_tags 字段."""
|
||||
count = (
|
||||
self.session.query(AssetAtomClipModel).filter(AssetAtomClipModel.id == clip_id).update({"ai_tags": ai_tags})
|
||||
)
|
||||
self.session.commit()
|
||||
return count > 0
|
||||
|
||||
def find_untagged(self, limit: int = 100) -> list[AssetAtomClip]:
|
||||
"""查找 ai_tags IS NULL 的片段，用于回填."""
|
||||
models = (
|
||||
self.session.query(AssetAtomClipModel)
|
||||
.filter(AssetAtomClipModel.ai_tags.is_(None))
|
||||
.order_by(AssetAtomClipModel.created_at.asc())
|
||||
.limit(limit)
|
||||
.all()
|
||||
)
|
||||
return [self._to_domain(m) for m in models]
|
||||
|
||||
def _to_model(self, clip: AssetAtomClip) -> AssetAtomClipModel:
|
||||
return AssetAtomClipModel(
|
||||
id=clip.id,
|
||||
@@ -111,7 +92,6 @@ class SQLAlchemyAssetAtomClipRepository:
|
||||
duration=clip.duration,
|
||||
clip_index=clip.clip_index,
|
||||
tags=clip.tags,
|
||||
ai_tags=clip.ai_tags,
|
||||
scene_change_at=clip.scene_change_at,
|
||||
is_fallback=clip.is_fallback,
|
||||
created_at=clip.created_at or datetime.now(UTC),
|
||||
|
||||
@@ -837,7 +837,6 @@ class AssetAtomClipModel(Base):
|
||||
duration = Column(Float, nullable=False)
|
||||
clip_index = Column(Integer, nullable=False)
|
||||
tags = Column(JSON, nullable=False, default=list)
|
||||
ai_tags = Column(JSON, nullable=True, default=None)
|
||||
scene_change_at = Column(Float, nullable=True)
|
||||
is_fallback = Column(Boolean, nullable=False, default=False)
|
||||
created_at = Column(DateTime, nullable=False, default=lambda: datetime.now(UTC))
|
||||
|
||||
@@ -68,7 +68,6 @@ class SharedSettings(BaseSettings):
|
||||
doubao_base_url: str = "https://ark.cn-beijing.volces.com/api/v3"
|
||||
doubao_timeout: int = 30
|
||||
doubao_max_retries: int = 2
|
||||
doubao_vision_model: str = "doubao-1-5-vision-pro-250915"
|
||||
|
||||
# ── MediaKit (火山引擎 AI 媒体工具) ──────────────────────────────────
|
||||
mediakit_api_key: str = ""
|
||||
|
||||
@@ -36,7 +36,6 @@ class AssetAtomClip:
|
||||
duration: float
|
||||
clip_index: int
|
||||
tags: list[str] = field(default_factory=list)
|
||||
ai_tags: dict | None = None
|
||||
scene_change_at: float | None = None
|
||||
is_fallback: bool = False
|
||||
created_at: datetime | None = None
|
||||
|
||||
@@ -1,292 +0,0 @@
|
||||
"""片段级 AI 标签 — #1970 智能剪辑流程重构 P2.
|
||||
|
||||
对每个 atom_clip 提取关键帧，调用豆包视觉理解 API 识别内容，
|
||||
生成结构化标签（场景、物体、动作、景别、是否有文字）。
|
||||
|
||||
纯函数 + IO 分离设计：
|
||||
- build_vision_prompt() 返回结构化 prompt
|
||||
- parse_vision_response(text) 解析 AI 返回的 JSON 标签
|
||||
- tag_atom_clip(...) 主入口，组合帧提取 → 视觉 API → 解析标签
|
||||
|
||||
降级策略：任何环节失败都返回 {"inherited_tags": clip.tags}，不阻断流程。
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import json
|
||||
import logging
|
||||
import subprocess
|
||||
import tempfile
|
||||
from pathlib import Path
|
||||
from typing import Any, Optional
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
# AI 标签结构的键
|
||||
AI_TAG_KEYS = ("scene", "objects", "action", "shot", "has_text")
|
||||
|
||||
|
||||
def build_vision_prompt() -> str:
|
||||
"""返回结构化标签提取 prompt.
|
||||
|
||||
要求 AI 以 JSON 格式返回片段内容标签，包含：
|
||||
- scene: 场景类型列表（如 "工厂", "办公室", "户外"）
|
||||
- objects: 出现的物体列表（如 "产品", "手机", "电脑"）
|
||||
- action: 动作类型列表（如 "演示", "说话", "操作"）
|
||||
- shot: 景别（"特写" / "中景" / "远景" 之一）
|
||||
- has_text: 画面中是否有显著文字（true/false）
|
||||
"""
|
||||
return """请分析这段视频片段的关键帧，识别内容并返回 JSON 格式标签。
|
||||
|
||||
要求返回以下 JSON 结构（严格 JSON，不要添加其他文字）：
|
||||
{
|
||||
"scene": ["场景1", "场景2"],
|
||||
"objects": ["物体1", "物体2"],
|
||||
"action": ["动作1"],
|
||||
"shot": "特写|中景|远景",
|
||||
"has_text": true/false
|
||||
}
|
||||
|
||||
规则：
|
||||
- scene: 场景类型，如"工厂"、"办公室"、"户外"、"商店"、"家庭"等，1-3个
|
||||
- objects: 画面中可见的主要物体，如"产品"、"手机"、"电脑"、"食品"等，1-5个
|
||||
- action: 人物或物体正在进行的动作，如"演示"、"说话"、"操作"、"展示"等，1-3个
|
||||
- shot: 景别判断，只能是"特写"、"中景"或"远景"之一
|
||||
- has_text: 画面中是否有显著可读文字（标题、字幕、标语等）
|
||||
|
||||
请只返回 JSON，不要有其他说明文字。"""
|
||||
|
||||
|
||||
def parse_vision_response(text: str) -> dict:
|
||||
"""解析 AI 返回的 JSON 标签文本.
|
||||
|
||||
Args:
|
||||
text: 视觉 API 返回的文本，期望是 JSON 格式。
|
||||
|
||||
Returns:
|
||||
结构化标签 dict，格式如：
|
||||
{"scene": [...], "objects": [...], "action": [...], "shot": "...", "has_text": bool}
|
||||
|
||||
解析失败时返回空 dict。
|
||||
"""
|
||||
if not text or not text.strip():
|
||||
return {}
|
||||
|
||||
# 尝试直接解析
|
||||
cleaned = text.strip()
|
||||
|
||||
# 去除可能的 markdown 代码块包裹
|
||||
if cleaned.startswith("```"):
|
||||
lines = cleaned.split("\n")
|
||||
# 去掉首尾的 ``` 行
|
||||
start = 1
|
||||
end = len(lines)
|
||||
for i in range(len(lines) - 1, 0, -1):
|
||||
if lines[i].strip().startswith("```"):
|
||||
end = i
|
||||
break
|
||||
cleaned = "\n".join(lines[start:end]).strip()
|
||||
|
||||
try:
|
||||
data = json.loads(cleaned)
|
||||
except json.JSONDecodeError:
|
||||
# 尝试从文本中提取 JSON 块
|
||||
try:
|
||||
start_idx = cleaned.index("{")
|
||||
end_idx = cleaned.rindex("}") + 1
|
||||
data = json.loads(cleaned[start_idx:end_idx])
|
||||
except (ValueError, json.JSONDecodeError):
|
||||
logger.warning("无法解析 AI 标签响应: %s", text[:200])
|
||||
return {}
|
||||
|
||||
if not isinstance(data, dict):
|
||||
return {}
|
||||
|
||||
# 验证和清洗各字段
|
||||
result: dict[str, Any] = {}
|
||||
for key in ("scene", "objects", "action"):
|
||||
val = data.get(key)
|
||||
if isinstance(val, list):
|
||||
result[key] = [str(v).strip() for v in val if str(v).strip()]
|
||||
elif isinstance(val, str) and val.strip():
|
||||
result[key] = [val.strip()]
|
||||
else:
|
||||
result[key] = []
|
||||
|
||||
shot_val = data.get("shot", "")
|
||||
if isinstance(shot_val, str) and shot_val.strip() in ("特写", "中景", "远景"):
|
||||
result["shot"] = shot_val.strip()
|
||||
else:
|
||||
result["shot"] = ""
|
||||
|
||||
has_text_val = data.get("has_text")
|
||||
if isinstance(has_text_val, bool):
|
||||
result["has_text"] = has_text_val
|
||||
elif isinstance(has_text_val, str):
|
||||
result["has_text"] = has_text_val.lower() in ("true", "yes", "1")
|
||||
else:
|
||||
result["has_text"] = False
|
||||
|
||||
return result
|
||||
|
||||
|
||||
def _extract_frames_via_mediakit(
|
||||
mediakit_client: Any,
|
||||
video_url: str,
|
||||
start_time: float,
|
||||
end_time: float,
|
||||
) -> Optional[list[str]]:
|
||||
"""通过 MediaKit 提取 3 帧（首、中、尾）.
|
||||
|
||||
Returns:
|
||||
图片 URL 列表（3 个），失败返回 None。
|
||||
"""
|
||||
try:
|
||||
frames = mediakit_client.extract_frames(
|
||||
video_url=video_url,
|
||||
strategy="SpecifiedTime",
|
||||
max_frames=3,
|
||||
poll_interval=2.0,
|
||||
max_poll_attempts=30,
|
||||
)
|
||||
# MediaKit SpecifiedTime 策略可能不支持直接传时间点
|
||||
# 如果返回结果不够 3 帧，降级到 ffmpeg
|
||||
if frames and len(frames) >= 1:
|
||||
urls = [f.get("image_url", "") for f in frames if f.get("image_url")]
|
||||
if urls:
|
||||
return urls
|
||||
except Exception as e:
|
||||
logger.warning("MediaKit 抽帧失败，将降级为 ffmpeg: %s", e)
|
||||
|
||||
return None
|
||||
|
||||
|
||||
def _extract_frames_via_ffmpeg(
|
||||
video_url: str,
|
||||
start_time: float,
|
||||
end_time: float,
|
||||
) -> Optional[list[str]]:
|
||||
"""通过 ffmpeg 本地提取 3 帧并转为 base64.
|
||||
|
||||
Returns:
|
||||
base64 data URI 列表（3 个），失败返回 None。
|
||||
"""
|
||||
import base64
|
||||
|
||||
mid_time = round((start_time + end_time) / 2, 3)
|
||||
timestamps = [round(start_time, 3), mid_time, round(end_time, 3)]
|
||||
|
||||
try:
|
||||
frames_b64: list[str] = []
|
||||
with tempfile.TemporaryDirectory() as tmpdir:
|
||||
for i, ts in enumerate(timestamps):
|
||||
out_path = Path(tmpdir) / f"frame_{i}.jpg"
|
||||
cmd = [
|
||||
"ffmpeg",
|
||||
"-y",
|
||||
"-ss",
|
||||
str(ts),
|
||||
"-i",
|
||||
video_url,
|
||||
"-vframes",
|
||||
"1",
|
||||
"-q:v",
|
||||
"2",
|
||||
str(out_path),
|
||||
]
|
||||
result = subprocess.run(
|
||||
cmd,
|
||||
capture_output=True,
|
||||
timeout=30,
|
||||
)
|
||||
if result.returncode != 0 or not out_path.exists():
|
||||
logger.warning("ffmpeg 抽帧失败 ts=%s: %s", ts, result.stderr[:200])
|
||||
continue
|
||||
|
||||
img_data = out_path.read_bytes()
|
||||
b64 = base64.b64encode(img_data).decode("ascii")
|
||||
frames_b64.append(f"data:image/jpeg;base64,{b64}")
|
||||
|
||||
if frames_b64:
|
||||
return frames_b64
|
||||
except Exception as e:
|
||||
logger.warning("ffmpeg 抽帧异常: %s", e)
|
||||
|
||||
return None
|
||||
|
||||
|
||||
def tag_atom_clip(
|
||||
clip: Any,
|
||||
video_url: str,
|
||||
doubao_client: Any,
|
||||
mediakit_client: Any | None = None,
|
||||
storage: Any | None = None,
|
||||
) -> dict:
|
||||
"""主入口：为单个 atom_clip 生成 AI 标签.
|
||||
|
||||
流程：提取帧 → 调视觉 API → 解析标签 → 返回结构化标签 dict。
|
||||
任何环节失败返回 {"inherited_tags": clip.tags}，不阻断流程。
|
||||
|
||||
Args:
|
||||
clip: AssetAtomClip 领域对象（需有 start_time, end_time, tags）。
|
||||
video_url: 素材视频的公网可访问 URL。
|
||||
doubao_client: DoubaoClient 实例。
|
||||
mediakit_client: MediaKitClient 实例（可选，不可用时降级 ffmpeg）。
|
||||
storage: SharedStorageService 实例（可选，用于获取签名 URL）。
|
||||
|
||||
Returns:
|
||||
结构化标签 dict，格式如：
|
||||
{"scene": [...], "objects": [...], "action": [...], "shot": "...",
|
||||
"has_text": bool, "inherited_tags": [...]}
|
||||
"""
|
||||
inherited = list(getattr(clip, "tags", []) or [])
|
||||
|
||||
# 检查 DoubaoClient 是否可用
|
||||
if not getattr(doubao_client, "is_available", False):
|
||||
logger.info("DoubaoClient 不可用，跳过 AI 标签: clip_id=%s", getattr(clip, "id", ""))
|
||||
return {"inherited_tags": inherited}
|
||||
|
||||
# 提取帧图片
|
||||
frame_urls: Optional[list[str]] = None
|
||||
start_time = getattr(clip, "start_time", 0.0)
|
||||
end_time = getattr(clip, "end_time", 0.0)
|
||||
|
||||
# 优先使用 MediaKit
|
||||
if mediakit_client and getattr(mediakit_client, "is_available", False):
|
||||
frame_urls = _extract_frames_via_mediakit(mediakit_client, video_url, start_time, end_time)
|
||||
|
||||
# MediaKit 不可用或失败 → 降级 ffmpeg
|
||||
if not frame_urls:
|
||||
frame_urls = _extract_frames_via_ffmpeg(video_url, start_time, end_time)
|
||||
|
||||
if not frame_urls:
|
||||
logger.warning("帧提取失败，跳过 AI 标签: clip_id=%s", getattr(clip, "id", ""))
|
||||
return {"inherited_tags": inherited}
|
||||
|
||||
# 调用视觉 API
|
||||
prompt = build_vision_prompt()
|
||||
messages = [{"role": "user", "content": prompt}]
|
||||
|
||||
try:
|
||||
response_text = doubao_client.vision_completion(
|
||||
messages=messages,
|
||||
images=frame_urls,
|
||||
timeout=60,
|
||||
)
|
||||
except Exception as e:
|
||||
logger.warning("视觉 API 调用异常: clip_id=%s error=%s", getattr(clip, "id", ""), e)
|
||||
return {"inherited_tags": inherited}
|
||||
|
||||
if not response_text:
|
||||
logger.warning("视觉 API 返回空: clip_id=%s", getattr(clip, "id", ""))
|
||||
return {"inherited_tags": inherited}
|
||||
|
||||
# 解析标签
|
||||
ai_tags = parse_vision_response(response_text)
|
||||
if not ai_tags:
|
||||
logger.warning("标签解析失败: clip_id=%s response=%s", getattr(clip, "id", ""), response_text[:200])
|
||||
return {"inherited_tags": inherited}
|
||||
|
||||
# 合并 inherited_tags
|
||||
ai_tags["inherited_tags"] = inherited
|
||||
return ai_tags
|
||||
@@ -1,4 +1,4 @@
|
||||
"""叙事剪辑素材标签匹配 — #1970 PR3 + P2 AI 标签加权.
|
||||
"""叙事剪辑素材标签匹配 — #1970 PR3.
|
||||
|
||||
叙事模式下，选片在现有评分（smart_match / atom_clip_selector）之前先做一层
|
||||
文案标签匹配：
|
||||
@@ -8,12 +8,6 @@
|
||||
- 调用方对优先池跑现有 smart_select_assets，数量不足时用普通池补足
|
||||
（无任何匹配 → 完全降级为现有随机逻辑，行为与改造前一致）。
|
||||
|
||||
P2 AI 标签加权（#1970 fragment-level AI tagging）：
|
||||
- 片段级 AI 标签（scene/objects/action）与文案标签做交集时权重 2.0
|
||||
- 素材级标签（tag_ids 映射名）与文案标签交集时权重 1.0
|
||||
- 综合得分 = sum(命中权重) / max(可能权重)
|
||||
- 有 AI 标签的片段命中时优先于仅素材标签命中的片段
|
||||
|
||||
纯函数模块：标签 id→名称映射由调用方查 TagModel 后注入，不直接碰 DB。
|
||||
"""
|
||||
|
||||
@@ -24,10 +18,6 @@ from typing import Any, Iterable
|
||||
# 标签归一化后仍短于此长度的标签不参与匹配（避免「的」「是」这类噪声短词）
|
||||
MIN_TAG_LEN = 2
|
||||
|
||||
# 标签匹配权重
|
||||
AI_TAG_WEIGHT = 2.0 # AI 标签命中权重
|
||||
ASSET_TAG_WEIGHT = 1.0 # 素材标签命中权重
|
||||
|
||||
|
||||
def normalize_tag(tag: Any) -> str:
|
||||
"""标签归一化：去空白、小写。数字/英文统一小写，中文不受影响。"""
|
||||
@@ -57,81 +47,19 @@ def build_asset_tag_name_index(tag_names_by_id: dict[str, Any]) -> dict[str, set
|
||||
return index
|
||||
|
||||
|
||||
def _extract_ai_tag_names(ai_tags: dict) -> set[str]:
|
||||
"""从 AI 标签 dict 中提取所有标签名（scene + objects + action）.
|
||||
|
||||
Args:
|
||||
ai_tags: 片段级 AI 标签 dict，如 {"scene": [...], "objects": [...], "action": [...], ...}
|
||||
|
||||
Returns:
|
||||
归一化后的标签名集合。
|
||||
"""
|
||||
names: set[str] = set()
|
||||
for key in ("scene", "objects", "action"):
|
||||
values = ai_tags.get(key)
|
||||
if isinstance(values, list):
|
||||
names |= _normalize_tags(values)
|
||||
return names
|
||||
|
||||
|
||||
def _compute_ai_score(
|
||||
asset_id: str,
|
||||
wanted: set[str],
|
||||
clip_ai_tags_by_asset: dict[str, list[dict]] | None,
|
||||
) -> float:
|
||||
"""计算单个素材的 AI 标签加权得分.
|
||||
|
||||
对该素材的所有片段 AI 标签，求各片段标签名与文案标签交集的加权总和。
|
||||
每个片段的命中权重 = 命中数 × AI_TAG_WEIGHT。
|
||||
最终取所有片段的最高得分（而非累加，避免片段数多的素材不公平占优）。
|
||||
|
||||
Args:
|
||||
asset_id: 素材 ID。
|
||||
wanted: 归一化后的文案标签集合。
|
||||
clip_ai_tags_by_asset: {asset_id: [ai_tag_dict, ...]} 每个片段一个。
|
||||
|
||||
Returns:
|
||||
AI 标签加权得分（≥0）。
|
||||
"""
|
||||
if not clip_ai_tags_by_asset or not wanted:
|
||||
return 0.0
|
||||
|
||||
clips = clip_ai_tags_by_asset.get(asset_id)
|
||||
if not clips:
|
||||
return 0.0
|
||||
|
||||
best_score = 0.0
|
||||
for ai_tags in clips:
|
||||
if not ai_tags or not isinstance(ai_tags, dict):
|
||||
continue
|
||||
ai_names = _extract_ai_tag_names(ai_tags)
|
||||
hits = ai_names & wanted
|
||||
score = len(hits) * AI_TAG_WEIGHT
|
||||
if score > best_score:
|
||||
best_score = score
|
||||
|
||||
return best_score
|
||||
|
||||
|
||||
def match_assets_by_script_tags(
|
||||
assets: list[Any],
|
||||
*,
|
||||
script_tags: Iterable[Any],
|
||||
tag_names_by_id: dict[str, Any] | None = None,
|
||||
clip_ai_tags_by_asset: dict[str, list[dict]] | None = None,
|
||||
) -> tuple[list[Any], list[Any]]:
|
||||
"""按文案标签把素材拆成「命中池 / 未命中池」，保持输入相对顺序。
|
||||
|
||||
P2 加权逻辑：
|
||||
- AI 标签命中（scene/objects/action ∩ 文案标签）权重 2.0
|
||||
- 素材标签命中（tag_ids 映射名 ∩ 文案标签）权重 1.0
|
||||
- 任一权重 > 0 → 命中池，否则 → 未命中池
|
||||
|
||||
Args:
|
||||
assets: 候选素材（domain Asset，需有 id 与 tag_ids）。
|
||||
script_tags: 文案 tags（字符串数组，名称语义）。
|
||||
tag_names_by_id: asset_id → 素材标签名列表。
|
||||
clip_ai_tags_by_asset: #1970 P2 — {asset_id: [ai_tag_dict, ...]}。
|
||||
tag_names_by_id: asset_id → 素材标签名列表；素材只有 tag_ids 时由调用方
|
||||
查 TagModel 名称后传入。为空则视为无素材命中。
|
||||
|
||||
Returns:
|
||||
(matched, unmatched)：命中任一文案标签的素材 / 其余素材。
|
||||
@@ -146,74 +74,23 @@ def match_assets_by_script_tags(
|
||||
unmatched: list[Any] = []
|
||||
for asset in assets:
|
||||
asset_id = str(getattr(asset, "id", "") or "")
|
||||
|
||||
# P2: AI 标签加权得分
|
||||
ai_score = _compute_ai_score(asset_id, wanted, clip_ai_tags_by_asset)
|
||||
|
||||
# 素材标签得分
|
||||
names = set(name_index.get(asset_id, set()))
|
||||
# 兼容素材自身带字符串 tags（旧链路/测试替身）
|
||||
raw_tags = getattr(asset, "tags", None)
|
||||
if raw_tags:
|
||||
names |= _normalize_tags(raw_tags)
|
||||
asset_score = len(names & wanted) * ASSET_TAG_WEIGHT
|
||||
|
||||
# 综合得分 > 0 → 命中池
|
||||
if ai_score > 0 or asset_score > 0:
|
||||
if names & wanted:
|
||||
matched.append(asset)
|
||||
else:
|
||||
unmatched.append(asset)
|
||||
return matched, unmatched
|
||||
|
||||
|
||||
def compute_tag_match_score(
|
||||
asset_id: str,
|
||||
*,
|
||||
script_tags: Iterable[Any],
|
||||
tag_names_by_id: dict[str, Any] | None = None,
|
||||
clip_ai_tags_by_asset: dict[str, list[dict]] | None = None,
|
||||
) -> float:
|
||||
"""计算单个素材的标签匹配综合得分（0.0 ~ 1.0）.
|
||||
|
||||
综合得分 = sum(命中权重) / max(可能权重)
|
||||
- AI 标签每命中一个 +2.0
|
||||
- 素材标签每命中一个 +1.0
|
||||
- max_possible = len(wanted) * (AI_TAG_WEIGHT + ASSET_TAG_WEIGHT)
|
||||
|
||||
Args:
|
||||
asset_id: 素材 ID。
|
||||
script_tags: 文案标签。
|
||||
tag_names_by_id: 素材标签名索引。
|
||||
clip_ai_tags_by_asset: AI 标签索引。
|
||||
|
||||
Returns:
|
||||
归一化得分 0.0~1.0。
|
||||
"""
|
||||
wanted = _normalize_tags(script_tags)
|
||||
if not wanted:
|
||||
return 0.0
|
||||
|
||||
# AI 得分
|
||||
ai_score = _compute_ai_score(asset_id, wanted, clip_ai_tags_by_asset)
|
||||
|
||||
# 素材标签得分
|
||||
name_index = build_asset_tag_name_index(tag_names_by_id or {})
|
||||
names = name_index.get(asset_id, set())
|
||||
asset_score = len(names & wanted) * ASSET_TAG_WEIGHT
|
||||
|
||||
# 归一化：最大可能得分 = 文案标签数 × (AI权重 + 素材权重)
|
||||
max_possible = len(wanted) * (AI_TAG_WEIGHT + ASSET_TAG_WEIGHT)
|
||||
if max_possible <= 0:
|
||||
return 0.0
|
||||
|
||||
return min((ai_score + asset_score) / max_possible, 1.0)
|
||||
|
||||
|
||||
def pick_narrative_assets(
|
||||
assets: list[Any],
|
||||
*,
|
||||
script_tags: Iterable[Any],
|
||||
tag_names_by_id: dict[str, Any] | None = None,
|
||||
clip_ai_tags_by_asset: dict[str, list[dict]] | None = None,
|
||||
limit: int | None = None,
|
||||
rng: Any = None,
|
||||
) -> list[Any]:
|
||||
@@ -223,13 +100,9 @@ def pick_narrative_assets(
|
||||
smart_match.smart_select_assets（质量/时长/新鲜度/未使用 + 随机噪声），
|
||||
不重写评分维度。
|
||||
|
||||
P2 增强：有 AI 标签的片段命中时权重更高（2.0 vs 1.0），
|
||||
命中池内部按综合标签得分排序（AI 标签命中多的排前面）。
|
||||
|
||||
Args:
|
||||
assets: ready 视频素材候选（调用方负责状态/类型过滤）。
|
||||
script_tags / tag_names_by_id: 见 match_assets_by_script_tags。
|
||||
clip_ai_tags_by_asset: #1970 P2 — {asset_id: [ai_tag_dict, ...]}。
|
||||
limit: 需要的素材数量；None 表示全部（命中池 + 全部未命中池）。
|
||||
rng: 注入 smart_select_assets 的随机源（可复现）。
|
||||
|
||||
@@ -242,7 +115,6 @@ def pick_narrative_assets(
|
||||
assets,
|
||||
script_tags=script_tags,
|
||||
tag_names_by_id=tag_names_by_id,
|
||||
clip_ai_tags_by_asset=clip_ai_tags_by_asset,
|
||||
)
|
||||
|
||||
need = limit if (limit is not None and limit > 0) else None
|
||||
|
||||
@@ -37,7 +37,6 @@ class DoubaoClient:
|
||||
self.base_url: str = settings.doubao_base_url.rstrip("/")
|
||||
self.timeout: int = settings.doubao_timeout
|
||||
self.max_retries: int = settings.doubao_max_retries
|
||||
self.vision_model: str = settings.doubao_vision_model
|
||||
|
||||
@property
|
||||
def is_available(self) -> bool:
|
||||
@@ -104,99 +103,6 @@ class DoubaoClient:
|
||||
logger.error("豆包API调用最终失败: %s", last_error)
|
||||
return None
|
||||
|
||||
def vision_completion(
|
||||
self,
|
||||
messages: list[dict],
|
||||
images: list[str] | None = None,
|
||||
max_tokens: int = 2048,
|
||||
temperature: float = 0.3,
|
||||
timeout: int | None = None,
|
||||
) -> Optional[str]:
|
||||
"""调用豆包视觉理解 API（OpenAI 兼容多模态格式）.
|
||||
|
||||
将 images 附加到最后一条 user message 的 content 中，
|
||||
使用 vision_model（默认 doubao-1-5-vision-pro-250915）。
|
||||
|
||||
Args:
|
||||
messages: 对话消息列表。最后一条 user message 会被注入图片内容。
|
||||
images: 图片列表，支持 base64 data URI 或 HTTP(S) URL。
|
||||
max_tokens: 最大生成 token 数，默认 2048。
|
||||
temperature: 采样温度，默认 0.3（视觉任务偏低更稳定）。
|
||||
timeout: 单次请求超时秒数，不传则使用默认 self.timeout。
|
||||
|
||||
Returns:
|
||||
模型返回的文本内容，失败返回 None。
|
||||
"""
|
||||
if not self.is_available:
|
||||
return None
|
||||
|
||||
# 构造多模态 content：先追加文本，再追加图片
|
||||
vision_messages = []
|
||||
for msg in messages:
|
||||
vision_messages.append(dict(msg))
|
||||
|
||||
# 将图片注入最后一条 user message
|
||||
if images and vision_messages:
|
||||
# 找到最后一条 user message
|
||||
for i in range(len(vision_messages) - 1, -1, -1):
|
||||
if vision_messages[i].get("role") == "user":
|
||||
text_content = vision_messages[i].get("content", "")
|
||||
multi_content: list[dict[str, Any]] = []
|
||||
if text_content:
|
||||
multi_content.append({"type": "text", "text": text_content})
|
||||
for img in images:
|
||||
if img.startswith("data:") or img.startswith("http://") or img.startswith("https://"):
|
||||
multi_content.append({"type": "image_url", "image_url": {"url": img}})
|
||||
else:
|
||||
# 当作 base64 编码
|
||||
multi_content.append(
|
||||
{"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{img}"}}
|
||||
)
|
||||
vision_messages[i]["content"] = multi_content
|
||||
break
|
||||
|
||||
url = f"{self.base_url}/chat/completions"
|
||||
headers = {
|
||||
"Authorization": f"Bearer {self.api_key}",
|
||||
"Content-Type": "application/json",
|
||||
}
|
||||
payload: dict[str, Any] = {
|
||||
"model": self.vision_model,
|
||||
"messages": vision_messages,
|
||||
"temperature": temperature,
|
||||
"max_tokens": max_tokens,
|
||||
}
|
||||
|
||||
req_timeout = timeout or self.timeout
|
||||
last_error: Optional[Exception] = None
|
||||
for attempt in range(self.max_retries + 1):
|
||||
try:
|
||||
response = httpx.post(
|
||||
url,
|
||||
headers=headers,
|
||||
json=payload,
|
||||
timeout=req_timeout,
|
||||
)
|
||||
response.raise_for_status()
|
||||
data = response.json()
|
||||
content = data["choices"][0]["message"]["content"]
|
||||
return content.strip()
|
||||
except Exception as e:
|
||||
last_error = e
|
||||
if attempt < self.max_retries:
|
||||
wait = 0.5 * (2**attempt)
|
||||
logger.warning(
|
||||
"豆包视觉API调用失败，%.1fs后重试 (第%d/%d次): %s",
|
||||
wait,
|
||||
attempt + 1,
|
||||
self.max_retries + 1,
|
||||
e,
|
||||
)
|
||||
time.sleep(wait)
|
||||
|
||||
logger.error("豆包视觉API调用最终失败: %s", last_error)
|
||||
return None
|
||||
|
||||
|
||||
# ── 单例 ─────────────────────────────────────────────────────────────────────
|
||||
|
||||
|
||||
@@ -1,292 +0,0 @@
|
||||
"""#1970 P2 片段级 AI 标签模块测试。
|
||||
|
||||
测试范围：
|
||||
- build_vision_prompt: 返回有效 prompt
|
||||
- parse_vision_response: 正常/异常/空值
|
||||
- tag_atom_clip: 成功/MediaKit不可用/视觉API失败/超时降级
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import json
|
||||
from dataclasses import dataclass, field
|
||||
from datetime import UTC, datetime
|
||||
|
||||
import pytest
|
||||
|
||||
from packages.domain.atom_clip_tagger import (
|
||||
build_vision_prompt,
|
||||
parse_vision_response,
|
||||
tag_atom_clip,
|
||||
)
|
||||
|
||||
# ── Fake 对象 ──────────────────────────────────────────────────────────────
|
||||
|
||||
|
||||
@dataclass
|
||||
class FakeClip:
|
||||
id: str = "clip-001"
|
||||
asset_id: str = "asset-001"
|
||||
start_time: float = 0.0
|
||||
end_time: float = 5.0
|
||||
duration: float = 5.0
|
||||
clip_index: int = 0
|
||||
tags: list[str] = field(default_factory=lambda: ["tag1", "tag2"])
|
||||
ai_tags: dict | None = None
|
||||
|
||||
|
||||
class FakeDoubaoClient:
|
||||
"""模拟豆包客户端."""
|
||||
|
||||
def __init__(self, available: bool = True, response: str | None = None, raise_error: bool = False):
|
||||
self._available = available
|
||||
self._response = response
|
||||
self._raise_error = raise_error
|
||||
self.vision_calls: list[dict] = []
|
||||
|
||||
@property
|
||||
def is_available(self) -> bool:
|
||||
return self._available
|
||||
|
||||
def vision_completion(self, messages, images=None, timeout=None, **kwargs):
|
||||
self.vision_calls.append({"messages": messages, "images": images, "timeout": timeout})
|
||||
if self._raise_error:
|
||||
raise RuntimeError("API error")
|
||||
return self._response
|
||||
|
||||
|
||||
class FakeMediaKitClient:
|
||||
"""模拟 MediaKit 客户端."""
|
||||
|
||||
def __init__(self, available: bool = True, frames: list[dict] | None = None):
|
||||
self._available = available
|
||||
self._frames = frames
|
||||
|
||||
@property
|
||||
def is_available(self) -> bool:
|
||||
return self._available
|
||||
|
||||
def extract_frames(self, video_url, strategy=None, max_frames=None, **kwargs):
|
||||
return self._frames
|
||||
|
||||
|
||||
# ── build_vision_prompt ────────────────────────────────────────────────────
|
||||
|
||||
|
||||
class TestBuildVisionPrompt:
|
||||
def test_returns_non_empty_string(self):
|
||||
prompt = build_vision_prompt()
|
||||
assert isinstance(prompt, str)
|
||||
assert len(prompt) > 100
|
||||
|
||||
def test_contains_required_keys(self):
|
||||
prompt = build_vision_prompt()
|
||||
assert "scene" in prompt
|
||||
assert "objects" in prompt
|
||||
assert "action" in prompt
|
||||
assert "shot" in prompt
|
||||
assert "has_text" in prompt
|
||||
|
||||
def test_requests_json_format(self):
|
||||
prompt = build_vision_prompt()
|
||||
assert "JSON" in prompt or "json" in prompt
|
||||
|
||||
|
||||
# ── parse_vision_response ──────────────────────────────────────────────────
|
||||
|
||||
|
||||
class TestParseVisionResponse:
|
||||
def test_valid_json(self):
|
||||
response = json.dumps(
|
||||
{
|
||||
"scene": ["工厂", "车间"],
|
||||
"objects": ["产品", "机器"],
|
||||
"action": ["演示"],
|
||||
"shot": "特写",
|
||||
"has_text": True,
|
||||
}
|
||||
)
|
||||
result = parse_vision_response(response)
|
||||
assert result["scene"] == ["工厂", "车间"]
|
||||
assert result["objects"] == ["产品", "机器"]
|
||||
assert result["action"] == ["演示"]
|
||||
assert result["shot"] == "特写"
|
||||
assert result["has_text"] is True
|
||||
|
||||
def test_json_with_markdown_code_block(self):
|
||||
response = '```json\n{"scene": ["办公室"], "objects": ["电脑"], "action": ["说话"], "shot": "中景", "has_text": false}\n```'
|
||||
result = parse_vision_response(response)
|
||||
assert result["scene"] == ["办公室"]
|
||||
assert result["has_text"] is False
|
||||
|
||||
def test_json_embedded_in_text(self):
|
||||
response = '这是一些说明文字\n{"scene": ["户外"], "objects": ["汽车"], "action": ["展示"], "shot": "远景", "has_text": false}\n结束'
|
||||
result = parse_vision_response(response)
|
||||
assert result["scene"] == ["户外"]
|
||||
|
||||
def test_empty_response(self):
|
||||
assert parse_vision_response("") == {}
|
||||
assert parse_vision_response(None) == {}
|
||||
assert parse_vision_response(" ") == {}
|
||||
|
||||
def test_invalid_json(self):
|
||||
assert parse_vision_response("这不是JSON") == {}
|
||||
|
||||
def test_partial_fields(self):
|
||||
response = json.dumps({"scene": ["工厂"]})
|
||||
result = parse_vision_response(response)
|
||||
assert result["scene"] == ["工厂"]
|
||||
assert result["objects"] == []
|
||||
assert result["shot"] == ""
|
||||
assert result["has_text"] is False
|
||||
|
||||
def test_invalid_shot_value(self):
|
||||
response = json.dumps({"scene": [], "objects": [], "action": [], "shot": "全景", "has_text": False})
|
||||
result = parse_vision_response(response)
|
||||
# "全景" 不在有效值 ("特写", "中景", "远景") 中
|
||||
assert result["shot"] == ""
|
||||
|
||||
def test_string_values_converted_to_list(self):
|
||||
response = json.dumps(
|
||||
{"scene": "工厂", "objects": "产品", "action": "演示", "shot": "特写", "has_text": "true"}
|
||||
)
|
||||
result = parse_vision_response(response)
|
||||
assert result["scene"] == ["工厂"]
|
||||
assert result["objects"] == ["产品"]
|
||||
assert result["has_text"] is True
|
||||
|
||||
def test_non_dict_json(self):
|
||||
assert parse_vision_response("[1, 2, 3]") == {}
|
||||
assert parse_vision_response('"hello"') == {}
|
||||
|
||||
|
||||
# ── tag_atom_clip ──────────────────────────────────────────────────────────
|
||||
|
||||
|
||||
class TestTagAtomClip:
|
||||
def test_success_with_mediakit(self):
|
||||
"""MediaKit 可用 + 视觉 API 成功 → 返回完整 AI 标签."""
|
||||
clip = FakeClip()
|
||||
fake_doubao = FakeDoubaoClient(
|
||||
response=json.dumps(
|
||||
{
|
||||
"scene": ["工厂"],
|
||||
"objects": ["产品"],
|
||||
"action": ["演示"],
|
||||
"shot": "特写",
|
||||
"has_text": False,
|
||||
}
|
||||
)
|
||||
)
|
||||
fake_mediakit = FakeMediaKitClient(
|
||||
frames=[
|
||||
{"image_url": "https://example.com/frame1.jpg", "timestamp": 0.0},
|
||||
{"image_url": "https://example.com/frame2.jpg", "timestamp": 2.5},
|
||||
{"image_url": "https://example.com/frame3.jpg", "timestamp": 5.0},
|
||||
]
|
||||
)
|
||||
|
||||
result = tag_atom_clip(
|
||||
clip=clip,
|
||||
video_url="https://example.com/video.mp4",
|
||||
doubao_client=fake_doubao,
|
||||
mediakit_client=fake_mediakit,
|
||||
)
|
||||
|
||||
assert result["scene"] == ["工厂"]
|
||||
assert result["objects"] == ["产品"]
|
||||
assert result["shot"] == "特写"
|
||||
assert result["inherited_tags"] == ["tag1", "tag2"]
|
||||
assert len(fake_doubao.vision_calls) == 1
|
||||
|
||||
def test_doubao_unavailable_returns_inherited(self):
|
||||
"""DoubaoClient 不可用 → 返回 inherited_tags."""
|
||||
clip = FakeClip()
|
||||
fake_doubao = FakeDoubaoClient(available=False)
|
||||
|
||||
result = tag_atom_clip(
|
||||
clip=clip,
|
||||
video_url="https://example.com/video.mp4",
|
||||
doubao_client=fake_doubao,
|
||||
)
|
||||
|
||||
assert result == {"inherited_tags": ["tag1", "tag2"]}
|
||||
assert len(fake_doubao.vision_calls) == 0
|
||||
|
||||
def test_mediakit_unavailable_no_ffmpeg(self):
|
||||
"""MediaKit 不可用 + 无 ffmpeg → 降级 inherited_tags."""
|
||||
clip = FakeClip()
|
||||
fake_doubao = FakeDoubaoClient()
|
||||
fake_mediakit = FakeMediaKitClient(available=False)
|
||||
|
||||
result = tag_atom_clip(
|
||||
clip=clip,
|
||||
video_url="https://example.com/video.mp4",
|
||||
doubao_client=fake_doubao,
|
||||
mediakit_client=fake_mediakit,
|
||||
)
|
||||
|
||||
# 没有 ffmpeg 的情况下，帧提取失败
|
||||
assert result == {"inherited_tags": ["tag1", "tag2"]}
|
||||
|
||||
def test_vision_api_error_returns_inherited(self):
|
||||
"""视觉 API 抛异常 → 降级 inherited_tags."""
|
||||
clip = FakeClip()
|
||||
fake_doubao = FakeDoubaoClient(raise_error=True)
|
||||
fake_mediakit = FakeMediaKitClient(frames=[{"image_url": "https://example.com/frame.jpg", "timestamp": 0.0}])
|
||||
|
||||
result = tag_atom_clip(
|
||||
clip=clip,
|
||||
video_url="https://example.com/video.mp4",
|
||||
doubao_client=fake_doubao,
|
||||
mediakit_client=fake_mediakit,
|
||||
)
|
||||
|
||||
assert result == {"inherited_tags": ["tag1", "tag2"]}
|
||||
|
||||
def test_vision_api_empty_response(self):
|
||||
"""视觉 API 返回空 → 降级 inherited_tags."""
|
||||
clip = FakeClip()
|
||||
fake_doubao = FakeDoubaoClient(response=None)
|
||||
fake_mediakit = FakeMediaKitClient(frames=[{"image_url": "https://example.com/frame.jpg", "timestamp": 0.0}])
|
||||
|
||||
result = tag_atom_clip(
|
||||
clip=clip,
|
||||
video_url="https://example.com/video.mp4",
|
||||
doubao_client=fake_doubao,
|
||||
mediakit_client=fake_mediakit,
|
||||
)
|
||||
|
||||
assert result == {"inherited_tags": ["tag1", "tag2"]}
|
||||
|
||||
def test_vision_api_invalid_json_response(self):
|
||||
"""视觉 API 返回无效 JSON → 降级 inherited_tags."""
|
||||
clip = FakeClip()
|
||||
fake_doubao = FakeDoubaoClient(response="这不是JSON格式")
|
||||
fake_mediakit = FakeMediaKitClient(frames=[{"image_url": "https://example.com/frame.jpg", "timestamp": 0.0}])
|
||||
|
||||
result = tag_atom_clip(
|
||||
clip=clip,
|
||||
video_url="https://example.com/video.mp4",
|
||||
doubao_client=fake_doubao,
|
||||
mediakit_client=fake_mediakit,
|
||||
)
|
||||
|
||||
assert result == {"inherited_tags": ["tag1", "tag2"]}
|
||||
|
||||
def test_clip_with_empty_tags(self):
|
||||
"""空素材标签 → inherited_tags 为空列表."""
|
||||
clip = FakeClip(tags=[])
|
||||
fake_doubao = FakeDoubaoClient(available=False)
|
||||
|
||||
result = tag_atom_clip(
|
||||
clip=clip,
|
||||
video_url="https://example.com/video.mp4",
|
||||
doubao_client=fake_doubao,
|
||||
)
|
||||
|
||||
assert result == {"inherited_tags": []}
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
pytest.main([__file__, "-q"])
|
||||
@@ -1,306 +0,0 @@
|
||||
"""#1970 P2 叙事匹配 AI 标签加权测试。
|
||||
|
||||
测试范围：
|
||||
- AI 标签命中时权重 2.0
|
||||
- 无 AI 标签时降级到素材标签权重 1.0
|
||||
- 混合场景（部分素材有 AI 标签，部分只有素材标签）
|
||||
- compute_tag_match_score 归一化得分
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import datetime as dt
|
||||
import random
|
||||
from dataclasses import dataclass, field
|
||||
|
||||
import pytest
|
||||
|
||||
from packages.domain.narrative_match import (
|
||||
AI_TAG_WEIGHT,
|
||||
ASSET_TAG_WEIGHT,
|
||||
_compute_ai_score,
|
||||
_extract_ai_tag_names,
|
||||
compute_tag_match_score,
|
||||
match_assets_by_script_tags,
|
||||
pick_narrative_assets,
|
||||
)
|
||||
|
||||
|
||||
@dataclass
|
||||
class FakeAsset:
|
||||
id: str
|
||||
tag_ids: list[str] = field(default_factory=list)
|
||||
tags: list[str] = field(default_factory=list)
|
||||
status: str = "ready"
|
||||
file_type: str = "video"
|
||||
duration: float = 10.0
|
||||
quality_score: float | None = None
|
||||
created_at: object = None
|
||||
metadata: dict = field(default_factory=dict)
|
||||
|
||||
|
||||
def _make_old_dt():
|
||||
return dt.datetime(2020, 1, 1, tzinfo=dt.UTC)
|
||||
|
||||
|
||||
# ── _extract_ai_tag_names ─────────────────────────────────────────────────
|
||||
|
||||
|
||||
class TestExtractAiTagNames:
|
||||
def test_extracts_all_keys(self):
|
||||
ai_tags = {
|
||||
"scene": ["工厂", "车间"],
|
||||
"objects": ["产品"],
|
||||
"action": ["演示"],
|
||||
"shot": "特写", # shot 不参与标签匹配
|
||||
"has_text": False,
|
||||
}
|
||||
names = _extract_ai_tag_names(ai_tags)
|
||||
assert names == {"工厂", "车间", "产品", "演示"}
|
||||
|
||||
def test_empty_dict(self):
|
||||
assert _extract_ai_tag_names({}) == set()
|
||||
|
||||
def test_none_values(self):
|
||||
ai_tags = {"scene": None, "objects": None, "action": None}
|
||||
assert _extract_ai_tag_names(ai_tags) == set()
|
||||
|
||||
def test_case_insensitive(self):
|
||||
ai_tags = {"scene": ["Factory"], "objects": [], "action": []}
|
||||
names = _extract_ai_tag_names(ai_tags)
|
||||
assert "factory" in names
|
||||
|
||||
|
||||
# ── _compute_ai_score ─────────────────────────────────────────────────────
|
||||
|
||||
|
||||
class TestComputeAiScore:
|
||||
def test_single_clip_hit(self):
|
||||
wanted = {"工厂", "演示"}
|
||||
clips = [{"scene": ["工厂"], "objects": [], "action": ["演示"]}]
|
||||
score = _compute_ai_score("a1", wanted, {"a1": clips})
|
||||
# 命中 2 个 × 2.0 = 4.0
|
||||
assert score == 2 * AI_TAG_WEIGHT
|
||||
|
||||
def test_multiple_clips_takes_best(self):
|
||||
wanted = {"工厂", "演示"}
|
||||
clips = [
|
||||
{"scene": ["工厂"], "objects": [], "action": []}, # 1 hit = 2.0
|
||||
{"scene": ["工厂"], "objects": [], "action": ["演示"]}, # 2 hits = 4.0
|
||||
]
|
||||
score = _compute_ai_score("a1", wanted, {"a1": clips})
|
||||
assert score == 2 * AI_TAG_WEIGHT # best = 2 hits
|
||||
|
||||
def test_no_match(self):
|
||||
wanted = {"美食"}
|
||||
clips = [{"scene": ["工厂"], "objects": [], "action": ["演示"]}]
|
||||
score = _compute_ai_score("a1", wanted, {"a1": clips})
|
||||
assert score == 0.0
|
||||
|
||||
def test_no_clips_for_asset(self):
|
||||
wanted = {"工厂"}
|
||||
assert _compute_ai_score("a1", wanted, {}) == 0.0
|
||||
assert _compute_ai_score("a1", wanted, None) == 0.0
|
||||
|
||||
def test_empty_wanted(self):
|
||||
clips = [{"scene": ["工厂"], "objects": [], "action": []}]
|
||||
assert _compute_ai_score("a1", set(), {"a1": clips}) == 0.0
|
||||
|
||||
|
||||
# ── match_assets_by_script_tags with AI tags ──────────────────────────────
|
||||
|
||||
|
||||
class TestMatchWithAiTags:
|
||||
def test_ai_tag_hit_puts_in_matched(self):
|
||||
"""有 AI 标签命中 → 进入命中池."""
|
||||
assets = [FakeAsset("a1", created_at=_make_old_dt())]
|
||||
clip_ai_tags = {"a1": [{"scene": ["工厂"], "objects": [], "action": []}]}
|
||||
|
||||
matched, unmatched = match_assets_by_script_tags(
|
||||
assets,
|
||||
script_tags=["工厂"],
|
||||
clip_ai_tags_by_asset=clip_ai_tags,
|
||||
)
|
||||
|
||||
assert [a.id for a in matched] == ["a1"]
|
||||
assert unmatched == []
|
||||
|
||||
def test_ai_tag_no_match_puts_in_unmatched(self):
|
||||
"""AI 标签未命中 → 进入未命中池."""
|
||||
assets = [FakeAsset("a1", created_at=_make_old_dt())]
|
||||
clip_ai_tags = {"a1": [{"scene": ["办公室"], "objects": [], "action": []}]}
|
||||
|
||||
matched, unmatched = match_assets_by_script_tags(
|
||||
assets,
|
||||
script_tags=["工厂"],
|
||||
clip_ai_tags_by_asset=clip_ai_tags,
|
||||
)
|
||||
|
||||
assert matched == []
|
||||
assert [a.id for a in unmatched] == ["a1"]
|
||||
|
||||
def test_asset_tag_still_works_without_ai_tags(self):
|
||||
"""无 AI 标签时，素材标签仍按权重 1.0 匹配."""
|
||||
assets = [FakeAsset("a1", tags=["工厂"], created_at=_make_old_dt())]
|
||||
|
||||
matched, unmatched = match_assets_by_script_tags(
|
||||
assets,
|
||||
script_tags=["工厂"],
|
||||
)
|
||||
|
||||
assert [a.id for a in matched] == ["a1"]
|
||||
|
||||
def test_mixed_ai_and_asset_tags(self):
|
||||
"""混合场景：一个素材有 AI 标签，另一个只有素材标签."""
|
||||
assets = [
|
||||
FakeAsset("a1", created_at=_make_old_dt()), # AI 标签命中
|
||||
FakeAsset("a2", tags=["工厂"], created_at=_make_old_dt()), # 素材标签命中
|
||||
FakeAsset("a3", tags=["美食"], created_at=_make_old_dt()), # 无命中
|
||||
]
|
||||
clip_ai_tags = {"a1": [{"scene": ["工厂"], "objects": [], "action": []}]}
|
||||
|
||||
matched, unmatched = match_assets_by_script_tags(
|
||||
assets,
|
||||
script_tags=["工厂"],
|
||||
clip_ai_tags_by_asset=clip_ai_tags,
|
||||
)
|
||||
|
||||
assert {a.id for a in matched} == {"a1", "a2"}
|
||||
assert [a.id for a in unmatched] == ["a3"]
|
||||
|
||||
def test_ai_tag_and_asset_tag_both_hit(self):
|
||||
"""同一素材 AI 标签和素材标签都命中 → 仍在命中池."""
|
||||
assets = [FakeAsset("a1", tags=["工厂"], created_at=_make_old_dt())]
|
||||
clip_ai_tags = {"a1": [{"scene": ["工厂"], "objects": [], "action": []}]}
|
||||
|
||||
matched, unmatched = match_assets_by_script_tags(
|
||||
assets,
|
||||
script_tags=["工厂"],
|
||||
tag_names_by_id={"a1": ["工厂"]},
|
||||
clip_ai_tags_by_asset=clip_ai_tags,
|
||||
)
|
||||
|
||||
assert [a.id for a in matched] == ["a1"]
|
||||
|
||||
|
||||
# ── compute_tag_match_score ───────────────────────────────────────────────
|
||||
|
||||
|
||||
class TestComputeTagMatchScore:
|
||||
def test_ai_only_score(self):
|
||||
"""仅 AI 标签命中."""
|
||||
clip_ai_tags = {"a1": [{"scene": ["工厂"], "objects": [], "action": ["演示"]}]}
|
||||
score = compute_tag_match_score(
|
||||
"a1",
|
||||
script_tags=["工厂", "演示"],
|
||||
clip_ai_tags_by_asset=clip_ai_tags,
|
||||
)
|
||||
# AI: 2 hits × 2.0 = 4.0; asset: 0; max = 2 × 3.0 = 6.0
|
||||
assert abs(score - 4.0 / 6.0) < 0.01
|
||||
|
||||
def test_asset_only_score(self):
|
||||
"""仅素材标签命中."""
|
||||
score = compute_tag_match_score(
|
||||
"a1",
|
||||
script_tags=["工厂", "演示"],
|
||||
tag_names_by_id={"a1": ["工厂"]},
|
||||
)
|
||||
# AI: 0; asset: 1 hit × 1.0 = 1.0; max = 2 × 3.0 = 6.0
|
||||
assert abs(score - 1.0 / 6.0) < 0.01
|
||||
|
||||
def test_both_ai_and_asset_score(self):
|
||||
"""AI 标签 + 素材标签同时命中."""
|
||||
clip_ai_tags = {"a1": [{"scene": ["工厂"], "objects": [], "action": []}]}
|
||||
score = compute_tag_match_score(
|
||||
"a1",
|
||||
script_tags=["工厂", "演示"],
|
||||
tag_names_by_id={"a1": ["工厂"]},
|
||||
clip_ai_tags_by_asset=clip_ai_tags,
|
||||
)
|
||||
# AI: 1 hit × 2.0 = 2.0; asset: 1 hit × 1.0 = 1.0; max = 2 × 3.0 = 6.0
|
||||
assert abs(score - 3.0 / 6.0) < 0.01
|
||||
|
||||
def test_no_match_score_zero(self):
|
||||
"""无命中 → 得分 0."""
|
||||
score = compute_tag_match_score(
|
||||
"a1",
|
||||
script_tags=["工厂"],
|
||||
tag_names_by_id={"a1": ["美食"]},
|
||||
)
|
||||
assert score == 0.0
|
||||
|
||||
def test_full_match_score_one(self):
|
||||
"""全命中 → 得分接近 1.0."""
|
||||
clip_ai_tags = {"a1": [{"scene": ["工厂"], "objects": ["产品"], "action": ["演示"]}]}
|
||||
score = compute_tag_match_score(
|
||||
"a1",
|
||||
script_tags=["工厂", "产品", "演示"],
|
||||
clip_ai_tags_by_asset=clip_ai_tags,
|
||||
)
|
||||
# AI: 3 hits × 2.0 = 6.0; max = 3 × 3.0 = 9.0 → 6/9 = 0.667
|
||||
# 注意：仅 AI 标签命中不可能达到 1.0（因为 max 包含素材权重）
|
||||
assert score > 0.5
|
||||
|
||||
def test_empty_script_tags(self):
|
||||
"""空文案标签 → 得分 0."""
|
||||
assert compute_tag_match_score("a1", script_tags=[]) == 0.0
|
||||
|
||||
|
||||
# ── pick_narrative_assets with AI tags ────────────────────────────────────
|
||||
|
||||
|
||||
class TestPickNarrativeWithAiTags:
|
||||
def _assets(self):
|
||||
old = _make_old_dt()
|
||||
return [
|
||||
FakeAsset("ai_match", created_at=old), # AI 标签命中
|
||||
FakeAsset("asset_match", tags=["工厂"], created_at=old), # 素材标签命中
|
||||
FakeAsset("no_match", tags=["美食"], created_at=old), # 无命中
|
||||
]
|
||||
|
||||
def test_ai_match_prioritized(self):
|
||||
"""AI 标签命中的素材进入命中池."""
|
||||
clip_ai_tags = {"ai_match": [{"scene": ["工厂"], "objects": [], "action": []}]}
|
||||
|
||||
picked = pick_narrative_assets(
|
||||
self._assets(),
|
||||
script_tags=["工厂"],
|
||||
clip_ai_tags_by_asset=clip_ai_tags,
|
||||
limit=2,
|
||||
rng=random.Random(0),
|
||||
)
|
||||
|
||||
ids = {a.id for a in picked}
|
||||
assert "ai_match" in ids
|
||||
assert "asset_match" in ids
|
||||
|
||||
def test_fallback_when_no_ai_match(self):
|
||||
"""AI 标签和素材标签都未命中 → 降级."""
|
||||
clip_ai_tags = {"ai_match": [{"scene": ["办公室"], "objects": [], "action": []}]}
|
||||
|
||||
picked = pick_narrative_assets(
|
||||
self._assets(),
|
||||
script_tags=["不存在"],
|
||||
clip_ai_tags_by_asset=clip_ai_tags,
|
||||
limit=2,
|
||||
rng=random.Random(0),
|
||||
)
|
||||
|
||||
assert len(picked) == 2 # 从全量中选取
|
||||
|
||||
def test_backward_compat_without_ai_tags(self):
|
||||
"""不传 clip_ai_tags_by_asset 时行为与之前完全一致."""
|
||||
picked = pick_narrative_assets(
|
||||
self._assets(),
|
||||
script_tags=["工厂"],
|
||||
limit=2,
|
||||
rng=random.Random(0),
|
||||
)
|
||||
|
||||
# 仅素材标签匹配
|
||||
ids = {a.id for a in picked}
|
||||
assert "asset_match" in ids
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
pytest.main([__file__, "-q"])
|
||||
Reference in New Issue
Block a user