Sansheng Gemini Video

Use when 用户要用 Gemini 原生多模态联合分析本地或在线视频的画面、声音与时序，支持 YouTube 直读，以及按需下载 B站、小红书、抖音或经一次性授权解析微信视频号；触发词：分析视频、录屏质检、视频链接分析、批量验收。只要字幕、只下载、或要剪辑出片（交给 sandy-video）时不用此 Skill。

5 starsImages, Video & Audio

Install

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

npx skills add sanshengai/sansheng-gemini-video --skill sansheng-gemini-video

Manual install

git clone --depth 1 https://github.com/sanshengai/sansheng-gemini-video.git /tmp/sansheng-gemini-video
mkdir -p ~/.claude/skills
cp -r /tmp/sansheng-gemini-video/. ~/.claude/skills/sansheng-gemini-video

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

From the SKILL.md

# sansheng-gemini-video · 视频之眼

把整段视频交给 Gemini 原生多模态处理，保留画面、声音和时间关系，输出结构化 JSON。不要退回“抽几帧让 Agent 猜”的弱路径。

## 核心路由

先判断输入，再执行最短可行路径：

| 输入 | 路由 | 是否下载 |
|---|---|---|
| 本地视频/音频绝对路径 | inline base64；超上限时 ffmpeg 自动压缩 | 已在本地 |
| 公共 YouTube URL | Gemini 原生 `fileData.fileUri` | 否 |
| B站/小红书/抖音/TikTok/其他网页视频 | `yt-dlp` 下载到临时目录，再 inline 分析 | 是 |
| 上述短视频平台且 `yt-dlp` 失败 | 仅在已配置 `AI_DOUYIN_API_KEY` 时调用解析代理，再下载 | 是 |
| 微信视频号 `weixin.qq.com/sph/...` | 一次性元宝 Cookie 授权后纯后台换直链、下载、分析；已就绪的本地服务仅作兜底 | 是 |

下载到临时目录的文件在分析后自动清理；传 `--download-dir` 或 `--keep-download` 才保留。微信本地服务管理的下载文件不自动删除。

处理非 YouTube URL、配置 Cookie/代理，或排查下载失败时，先读 [references/url-sources.md](references/url-sources.md)。

## 三条硬规则

1. **能直读就不下载** -- Gemini 官方只承诺公共 YouTube URL 可直接作为视频输入；其他网页 URL 不冒充可直读。
2. **整段原生理解** -- 下载只是取材，分析仍把完整媒体交给 Gemini，不以抽帧替代视听时序。
3. **凭证必须显式** -- 默认不读取浏览器 Cookie。微信授权由用户显式提供：Windows 可运行 `python scripts/wechat_auth.py` 隐藏粘贴并用 DPAPI 保存；macOS/Linux 使用 `GEMINI_VIDEO_YUANBAO_COOKIE_FILE` 指向本人创建、权限为 600 的本地凭证文件，或当前进程的 `GEMINI_VIDEO_YUANBAO_COOKIE`；失效只提示，不自动打开微信/浏览器。配置 `AI_DOUYIN_API_KEY` 视为允许把对应短视频链接发送给该解析服务。

## 快速命令

| 用户意图 | 命令 |
|---|---|
| 分析一个本地文件或 URL | `python scripts/analyze_video.py <源>` |
| 素材筛选 | `python scripts/analyze_video.py <源> --intent screening` |
| 逆向视觉/音频手法 | `python scripts/analyze_video.py <源> --intent reverse` |
| 自定义结构化分析 | `python scripts/analyze_video.py <源> --prompt "..." --schema schema.json` |
| 批量验收本地文件夹 | `python scripts/batch_eval.py <文件夹>` |

Read the full skill on GitHub

Details

Skill name: sansheng-gemini-video
Repository: sanshengai/sansheng-gemini-video
License: MIT
Repo updated: 19 September 2026

FAQ

What does the Sansheng Gemini Video skill do?

How do I install the Sansheng Gemini Video skill?

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

Is Sansheng Gemini Video popular?

It lives in sanshengai/sansheng-gemini-video, which has 5 GitHub stars and 0 forks.

Related skills

Superpartner

coffeeandburger72/superpartner

Build a living profile of your partner from chat history. Analyze personality, interests, important dates, and generate gift ideas and dating plans. Supports WhatsApp, WeChat, and LINE exports.

Images, Video & Audio

Roblox Blender Modelling

WildCake/roblox-blender-modelling

Model, UV unwrap, consolidate and prepare Blender assets for Roblox. Use for props, vehicles, mechanisms and static environments; reference reconstruction; stretched textures, inconsistent texel density or unintended UV overlap; excessive MeshParts, segmented exports and duplicate mesh uploads. Covers the official Roblox Blender add-on and import validation, not general gameplay or UI implementation.

Images, Video & Audio

Tiktok Ugc Seedance

xigua0626/tiktok-ugc-seedance

Convert a target ecommerce product and one TikTok competitor video into a TikTok Shop UGC strategy package: product positioning, category/search keywords for finding competitor videos, competitor video transcript and conversion-structure breakdown, 15-second English UGC voiceover script, visual shot plan, and Seedance 2.0 prompt. Use when the user provides a product link/image plus a TikTok reference link, asks to analyze a competitor TikTok video, write UGC TikTok scripts, or turn the script into Seedance prompts for cross-border ecommerce ads.

Images, Video & Audio

Melies

melies-co/melies-cli

AI filmmaking CLI with 148 built-in actors, 98 visual styles, and smart model selection. Generate images, videos, posters, and thumbnails without prompt engineering. Text-to-image, text-to-video, image-to-video pipeline, upscaling, background removal. Use --actor, --art-style, --lighting, --mood flags instead of writing prompts.

Images, Video & Audio

Video Debug

vladzima/video-debug

Use when the user references a video, screencast, screen recording, or any path ending in .mp4, .mov, .webm, .mkv, or .gif — especially when asking to debug a UI bug, visual glitch, or reproduction recording. Extracts key frames via ffmpeg scene detection so the agent can visually inspect what went wrong.

Images, Video & Audio

Launch Video Maker

SigmaMatrix-AI/launch-video-maker

Use when the user asks for an autonomous product launch video or promo from product materials, a website, screenshots, recordings, or source code, or to continue, edit, or review such a film project.

Images, Video & Audio

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