视频处理视频抽帧与时间轴标注工作台 AI 提示词 (Prompts)

使用说明（读者导向）

下面每条都是可直接复制给 AI 的 Prompt，目标是让 AI 产出一个可运行/可部署的「在线视频抽帧 + 时间轴标注工作台」。注意：抽帧=从现有视频提取现有帧，不涉及生成图片/渲染图。

Web 工作台：上传视频 + 时间轴缩略图 + 按时间点抽帧 + 导出标注

适合做一个“审片/沟通”页面：看时间轴、点时间点抽帧、对片段做备注，然后导出JSON给同事或客户。

English Prompt: You are a senior full-stack engineer. Build a production-ready web tool named "Video Frame Extractor & Timeline Annotator". Goal - Users upload a video file (mp4/mov/webm). The app generates a timeline with thumbnails (e.g., 1 thumbnail every 2 seconds, configurable). - Users can click any timestamp to extract the exact frame as PNG/JPEG (this is extraction from the source video, NOT image generation). - Users can create annotations for time ranges (start/end), tags, and notes. - Users can export annotations as JSON and CSV, and download extracted frames as a ZIP. Tech constraints - Use Next.js (App Router) + Type + Tailwind. - Backend: Node.js API routes. - Use ffmpeg to extract frames. Prefer ffmpeg.wasm for local-browser processing; if too heavy, provide a server-side ffmpeg fallback with clear security limits. - Store data in SQLite (better-sqlite3) with a simple schema. Deliverables 1) Full project code and file tree. 2) Exact install/run commands. 3) Deployment guide (Docker + one-click deploy steps). 4) Security notes: upload limits, allowed mime types, temp file cleanup. 5) At least 8 QA checks (upload edge cases, corrupt video, large file, export correctness, timestamp accuracy). UI/UX - A timeline strip with thumbnails + a playhead. - A right panel listing annotations with search and filter. - A "Validate" button that runs basic integrity checks and shows results. Include tests - Unit tests for timestamp parsing and export. - E2E test plan for upload and extraction.

中文释义: 让 AI 生成一个完整的在线工具：上传视频后自动生成缩略图时间轴；点击时间点即可抽取该帧；对片段做区间标注；支持JSON/CSV导出与帧打包下载，并附带可运行命令、部署说明与QA清单。

轻量版：纯前端（浏览器本地处理）抽帧与标注，不上传服务器

适合对隐私敏感的审片场景：视频不出本地，所有处理在浏览器完成。

English Prompt: Build a client-only web app for "Local Video Frame Extraction & Annotation". Requirements - No server upload. Use ffmpeg.wasm and Web Workers. - Allow: select local video, scrub timeline, extract frame at current time, save as PNG, and create annotations for ranges. - Export: annotations.json and annotations.csv. - Provide a lightweight thumbnail strategy (e.g., extract a thumbnail every N seconds with progress). Tech - Vite + React + Type . - Use IndexedDB to persist the annotation data and extracted frames data. Deliverables - Full repository code + file tree. - Run/build commands. - Performance notes and limits. - At least 6 QA checks (memory, long videos, browser support, worker failures, export validation).

中文释义: 生成一个“纯前端”版本，重点是隐私与离线可用：视频只在本地处理，仍需可运行项目、文件结构、运行命令与QA检查。

服务器版：FastAPI + ffmpeg 抽帧API，前端做时间轴与标注管理

适合团队协作：视频可集中管理，后端负责抽帧与导出，前端负责标注流程。

English Prompt: Design and implement a web tool with a Python backend. Backend - FastAPI + Uvicorn. - Endpoints: - POST /upload (stores video, returns videoId) - GET /videos/{id}/thumbnails?intervalSec=2 - GET /videos/{id}/frame?timeMs=12345&format=png - CRUD /videos/{id}/annotations - GET /videos/{id}/export (json/csv) - Use ffmpeg (system binary in Docker) for extraction; strictly validate inputs. - Store data in SQLite. Frontend - React + Type . - Timeline with thumbnails and annotation editor. Deliverables - Dockerfile + docker-compose. - File tree + full source. - Run commands. - At least 8 QA checks (time parsing, invalid timestamps, concurrency, cleanup, security limits). Important - This tool must only extract frames from existing videos; do not include any image generation instructions.

中文释义: 生成一个后端可部署的抽帧服务（FastAPI+ffmpeg）+ 前端时间轴标注界面，强调接口设计、输入校验、安全限制与部署可用。

标注增强：片段列表、快捷键、批量标签、审核流状态

适合内容审核/素材归档：对片段打状态（待审/通过/驳回）并批量导出。

English Prompt: Extend the "Video Timeline Annotator" into an audit-ready workflow tool. Features - Annotation has fields: startMs, endMs, label, severity, status (todo/approved/rejected), reviewer, updatedAt. - Keyboard shortcuts for play/pause, jump +/-1s, mark in/out, add label. - Bulk edit: apply tag/status to multiple annotations. - Export includes workflow fields. Deliverables - Full code changes (assume an existing React+API app; show all modified files). - Migration s for SQLite. - At least 10 QA checks including shortcut conflicts, export correctness, and status transitions.

中文释义: 让 AI 生成“审核流”增强能力：快捷键、批量标签、状态机字段、数据库迁移与更完整的QA校验，仍以可运行项目为交付。

精确时间：基于关键帧/PTS 的时间戳校准与误差提示

适合需要精确抽帧的场景：对长视频、可变帧率（VFR）做时间校准与误差可视化。

English Prompt: Implement "Timestamp Accuracy Mode" for a video frame extraction tool. Requirements - Detect if input is VFR and show a warning. - Use ffprobe to read stream time , frame rate, and keyframe info. - When users request a frame at timeMs, return: - requestedTimeMs - actualExtractedTimeMs - deltaMs - Provide a UI badge if deltaMs exceeds a threshold. Deliverables - Full implementation + commands. - At least 6 tests/QA items focusing on accuracy, VFR files, and edge timestamps.

中文释义: 让工具更“专业”：抽帧时返回实际抽取时间与误差，并提示VFR风险；输出完整实现与测试要点。

导出交付：一键生成“审片报告”HTML/PDF（仅导出，不生成图片）

适合对外沟通：把标注、时间点与截图文件名汇总成可阅读的报告。

English Prompt: Add an "Audit Report" export to the tool. Rules - The report must reference extracted frames (existing files) by filename/path and timestamp. - Do NOT generate new images. Only use frames extracted from the video. Output - report.html and report.pdf (PDF rendered from HTML using Playwright or a server-side renderer). - Include summary stats: number of annotations, statuses, labels frequency. Deliverables - Code + file tree changes. - Commands to produce the report. - At least 5 QA checks (missing frames, unicode notes, long text wrap, PDF render failures).

中文释义: 增加“交付件”导出能力：把标注和已抽取帧做成HTML/PDF报告（仅引用已抽帧文件，不做任何图像生成），并给出可复现命令与QA清单。

权限与分享：登录、项目空间、只读分享链接（团队协作）

适合多成员协同：项目级权限与只读链接，便于外部审阅。

English Prompt: Turn the tool into a multi-user app. Requirements - Auth: email+password + session cookies. - Workspaces: each workspace has projects (videos) and annotations. - Roles: owner/editor/viewer. - Share s: create a signed, time-limited, read-only to view a video timeline and annotations. Deliverables - Full implementation with data schema. - Security checklist (OWASP basics, upload abuse prevention). - At least 8 QA checks including permission boundaries.

中文释义: 让 AI 生成一套可落地的权限与分享机制（登录、空间、角色、只读分享链接），并强调安全与权限边界的QA。

可观测性：抽帧任务队列、进度条、失败重试与日志追踪

适合更大文件或多人并发：抽帧/生成缩略图是耗时任务，需要队列和可观测性。

English Prompt: Add background jobs for thumbnail generation and bulk frame extraction. Tech - Node.js + BullMQ (Redis) or a lightweight queue. - Progress reporting via SSE or WebSocket. - Structured logs with requestId/jobId. Deliverables - Full code + docker-compose including Redis (if used). - Run commands. - At least 8 QA checks: retries, cancellation, partial failures, concurrent jobs, cleanup.