Video Watch

看视频与动图。把视频/GIF/录屏翻译成可核算的图版序列并产出观看交付：ffmpeg 逐帧量测（切点/冻结/运动/像素突变/静音）→ 骨架优先采样（低运动单帧、高运动成组）→ 联络表与位置映射清单 → 画面文字 OCR + 语音 ASR（按需）→ 完整描述 + 部分细节 + 带时间节点的转写。Invoke when 用户要求看视频/看动图/分析录屏/GIF 拆帧/视频里发生了什么/某时刻画面是什么/把视频里的文字或语音转写出来。

29 starsImages, Video & Audio

Install

skills CLI (Claude Code, Cursor, Codex and more)

npx skills add CFITCorporation/video-watch-skill --skill video-watch

Manual install

git clone --depth 1 https://github.com/CFITCorporation/video-watch-skill.git /tmp/video-watch-skill
mkdir -p ~/.claude/skills
cp -r /tmp/video-watch-skill/. ~/.claude/skills/video-watch

Personal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.

From the SKILL.md

# video-watch — 看视频与动图

把连续时间轴翻译成**离散图版序列**，让只有单帧视觉的模型也能给出带时间戳的结论。

**核心机理**：ffmpeg 看全部帧，我看其中几十帧，二者靠**时间戳**缝合。
**交付目标**：给出**人能用的观看结果**（完整描述 + 细节 + 按需转写），不是抖动审计。

本文件写给**执行这个技能的智能体**：下文"我"即执行者。

## 安装

### 依赖分档

| 依赖 | 用途 | 档次 |
|---|---|---|
| **Python 3.8+** | 运行 | 必需 |
| **ffmpeg + ffprobe** | 抽帧、量测、烧索引 | 必需（见下"完整构建"） |
| **ImageMagick** | 拼贴图版、差分合成 | 必需 |
| **TrueType 字体** | `drawtext` 烧帧号/时间码 | 必需（各平台通常自带） |
| Pillow | `gif` 子命令 | 可选 |
| numpy + Pillow | `vwtools` 专用脚本 | 可选 |
| faster-whisper | `asr` 子命令 | 可选 |
| rapidocr-onnxruntime | `ocr` 子命令 | 可选 |

**主链路的 Python 侧零第三方依赖**（`probe`/`plan`/`sheet`/`grid`/`seq`/`read`/`report` 全部只用标准库）。可选依赖都是函数内延迟导入，不装不影响主链路。

### ffmpeg 必须是完整构建

需要这些滤镜：`scdet` `freezedetect` `silencedetect` `tblend` `signalstats` `tile` `drawtext`。

其中 **`drawtext` 需要 ffmpeg 编译时带 libfreetype**，而部分发行版与精简构建**不带**。缺它时 `sheet`/`grid`/`seq` 会烧不上帧号索引，且报错落在抽帧那一步，离真正原因很远 —— 所以工具在启动时就校验，缺了当场明确报错。

**版本要求：ffmpeg ≥ 5.1**（这条只管 `vwtools`）。它的逐帧解帧用 `-fps_mode`，而那个选项从 5.1 起才有；
老的 `-vsync` 已被 **ffmpeg 9.0 移除** —— 两边各有一段失效区间，只能选覆盖更长的那个。
`doctor` 会检查这个构建认不认 `-fps_mode`（多跑一次进程，只在 doctor 里做），
所以不用等用到 `vwtools` 才发现。主链路不受此限，滤镜齐即可。

### 安装命令

**下面三组按你的系统选一组，不是依次全跑**：

Read the full skill on GitHub

Details

Skill name: video-watch
Repository: CFITCorporation/video-watch-skill
License: MIT
Repo updated: 4 October 2026

FAQ

What does the Video Watch skill do?

How do I install the Video Watch skill?

skills CLI (Claude Code, Cursor, Codex and more): npx skills add CFITCorporation/video-watch-skill --skill video-watch. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.

Is Video Watch popular?

It lives in CFITCorporation/video-watch-skill, which has 29 GitHub stars and 0 forks.

Related skills

Manim Video

ApliroAI/manim-video-lab

Create production-quality Manim Community Edition explainer videos, math animations, algorithm visualizations, paper/PDF explainers, data stories, and architecture diagrams. Use when a user asks for 3Blue1Brown-style programmatic animation, Manim scene planning/coding/rendering, Docker-based Manim setup, or troubleshooting Manim renders.

Images, Video & Audio

Awesome AI Product Video Workflows

HiAPIAI/awesome-ai-video-workflows

Use when planning, creating, reviewing, or researching AI product videos from product images, ecommerce listings, or product briefs. Routes ecommerce sellers, ad teams, and AI creators through product-image preparation, hero visuals, storyboards, image-to-video generation, UGC ads, social variants, and media QC without inventing product claims.

Images, Video & Audio

Remotion Video Toolkit

shreefentsar/remotion-video-toolkit

Complete toolkit for programmatic video creation with Remotion + React. Covers animations, timing, rendering (CLI/Node.js/Lambda/Cloud Run), captions, 3D, charts, text effects, transitions, and media handling. Use when writing Remotion code, building video generation pipelines, or creating data-driven video templates.

Images, Video & Audio

Visual Multimedia

CheshireMew/visual-multimedia

将确认内容制成可编辑、预览、导出的技术图解、GIF、HTML 动画、Live Photo、产品功能宣传片、其它视频、音频或播客，并处理文案。Use when the user asks to 制作网页视觉、剪辑或对齐实拍/访谈/录屏、为视频查找或添加免费可商用且不要求公开署名的 BGM、从本地视频或链接提取金句并制作字幕拼图、建立无需 Live2D 的二次元口播角色，或编辑音频与播客；不用于独立静态设计、PPTX、研究核查、长文、拍摄规划、账号运营或发布。

Images, Video & Audio

Tech Article Image

ZD-AI-Lab/Tech-Article-Illustration-Master

为技术文章生成高质量主题配图。触发场景：(1)用户要求为文章配图/生成题图/制作封面图 (2)用户上传技术文章并要求生成配图 (3)用户提到"给文章配个图"、"生成主题图"、"做张封面"。核心能力：分析文章主题和情绪，从25种反AI油腻感的风格中匹配最佳选项，生成专业级图像提示词。支持MCP生图工具自动生成，无MCP时输出提示词供手动使用。

Images, Video & Audio

Video Sharingan

nikolasfaria/video-sharingan

Reverse-engineer the audiovisual production of a reference video (motion design, product launch, changelog, app promo, trailer) into measured numbers and a recreation brief. Use this when the user shares a video and says "make ours like this", "copy this style/camera/pacing", "what camera moves does this use", "how fast is the camera", "recreate this launch video", or before generating a video with Hyperframes, /brag, Remotion or After Effects from a reference. Measures cuts and shot durations, camera language (truck/pedestal/pan/push-in/pull-out/orbit, speed in px/s and %W/s, scale %/s, easin

Images, Video & Audio

Browse the most popular Claude skills, all images, video & audio skills, or write your own with the skills generator.