rea-video skill
rea-video is an agent skill that breaks a video down shot by shot, calling out real versus AI footage and building a remake plan.
- 7,760 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- Frames sent to a video-capable model reach an external provider (Gemini or Doubao) using a supplied access key.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/rea-videoDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Frames sent to a video-capable model reach an external provider (Gemini or Doubao) using a supplied access key.
This is third-party code. Review the repository files before installing.
What it does
rea-video installs a skill and CLI that, given a video (talking head, tabletop, AI video, 3D, motion graphics, mixed edit), determine for every shot how it was made: live footage with camera position, lighting and color grade, or AI generation with a reverse-engineered prompt and how many generations the shot took. It also breaks down captions and styled text with timing (exporting to SRT or ASS), stickers, and cut pacing against the beat and audio: human or AI voice, music with BPM, and where sound effects land. The output is a plan covering what to film, what to generate and what to build with code, plus a full `report.html` with per-shot frames and evidence for each call.
Who it is for. Short-video and AI-video creators who need to break down someone else's edit shot by shot and reproduce it with their own tools.
Good fit when
- Figuring out which shots in someone else's video are real footage and which are AI-generated.
- A remake needs an exact plan: what to film, what to generate, what to build with code.
- Captions and styled text from a video need to be extracted with timing into SRT or ASS.
- A competitor's edit and sound need a second-by-second breakdown: cuts, music, BPM, effects.
Not a fit when
- The video has not been checked against its original platform copy: the author's own re-compressed samples lose platform tags after compression for the repository.
- No video-capable model such as Gemini or Doubao is available and high-precision prompts for AI shots matter: without one the agent estimates them roughly.
- Only subtitle translation is needed, without a visual and edit breakdown.
Example request
Install rea-video and break down this video shot by shot: /path/to/video.mp4.Limitations
Precise AI-shot prompts and confident live-versus-generated splitting need a video-capable model (Gemini or Doubao) with its own key; without one the agent checks uncertain shots itself if it can read images. `ffmpeg` and Python 3.10 or newer are required. The detailed documentation is still Chinese-only; the agent translates it on the fly and answers in whichever language the question was asked. The author's own samples under `docs/cases/videos/` are re-compressed to keep the repository small, which strips TikTok and Douyin platform tags; reproducing the same result needs the original platform download.
How to disable. Remove the installed CLI and the skill links left in `~/.claude`, `~/.codex` or `~/.agents` by the install script.
FAQ
Can I install it without Git?
Through the quick start the agent installs the CLI itself with `Install https://github.com/JuneYaooo/rea-video`; manual install uses `brew install ffmpeg` and `bash scripts/install_skill.sh` after cloning.
What do I get as output?
A one-line summary such as the AI-video share and seconds of real footage, a remake plan split by shot type, and a full `report.html` with evidence for every shot.
Is a video-capable model required?
No, the skill still works without one, but AI-shot prompt accuracy drops; an agent that can read images checks uncertain shots itself.
Why do my results differ from the README examples?
Samples under `docs/cases/videos/` are re-compressed for the repository and lost their platform tags; matching the published report needs the original TikTok or Douyin file.
What language does the report come out in?
The report comes out in the language of the user's request: the agent passes the `--lang` flag itself, even though the project's detailed documentation is Chinese-only.
Discussion
No questions or reviews yet.
Related
SkillsVideo skills for Claude Code and Codex: break a clip into a shot-by-shot analysis and render video with a storyboard overlay
Two skills for generative video and images: video writes Seedance, Kling and Veo prompts by dramaturgy rules, image picks the model and its syntax
Reelpack breaks a local video into transcript, cuts, on-screen text and music beats so an agent reads it without watching.
A skill that turns a coding agent into a studio for voiceover explainer videos: word-level sync, a motion library and Remotion assembly
A skill that turns an idea, a link or a screen recording into finished video: HeyGen avatar, b-roll, captions and quality checks
motion-video-production: a skill for motion videos
motion-video-production
A skill that takes an agent through seven stages for a 15 to 90 second video, a script checking each stage before the next
README in short
The author demonstrates the breakdown on a phone-shot ad and its AI remake: the report correctly identified the original as one set shot from two camera positions, with the remake built from eight shots across three generations.