MMCPMarkets
WV

worldwonderer/video-recap-skills

Developer tools

This page was translated automatically. Technical details may differ from the original. View the original

600 stars Quality 76 Trend 43MIT
Repository Install

Clip any video into a narration recap with claude code skill| Use the claude code skill to edit any video into a Chinese narration video, supporting Jianying export

Overview

Locally, as long as you use ffmpeg and the Xiaomi MiMo Token Plan API Key, no GPU or model downloads needed—macOS, Linux, and Windows can all run it. - ASR, VLM, TTS all run Xiaomi MiMo; locally, apart from ffmpeg, there's no other dependence. - When the title/plot is clear or the brief prompts indicate the material is too thin, save character relationships and plot background into the background_research.json so VLM can more easily identify who. - Narration is told in segments in sequence, with full dubbing in whole segments, leaving blank space between sections to play the exciting original audio at full volume—roughly 70/30. - --edit-mode cut First edit the long video into a video, then write commentary on the finished video, with the timeline naturally aligned. - Upload multiple videos at once, select source_id segments and cut them into a single clip; Each video's analysis accumulates into a file system material library, which can be reused next time with GREP without recalculation. - Optional export of schema-driven multi-track clip drafts, with original, narration, BGM, and subtitles each occupying one track; FFMPEG remains the final criterion for final production. (No pip install: pure standard library + ffmpeg on PATH, Python 3.10+): Subtitles are burned into the screen by default and require ffmpeg—these packages are basically all pre-installed. If your ffmpeg hasn't been programmed with libass, it will immediately report an error and prompt before running (you can also add --no-burn-subtitles to output MP4 + .srt external subtitles without the blackout). Use python3 skills/video-recap/scripts/recap.py --doctor for self-check. (One key simultaneously drives ASR / VLM / TTS; Register first on platform.xiaomimimo.com to obtain it): Pay-as-you-go sk-* keys default to https://api.xiaomimimo.com/v1。其它都有默认值；想分别配 key/URL or to modify models, tones, volume, subtitles, etc., as detailed in the configuration manual. The engine is pure Python + ffmpeg + a MiMo key, independent of the specific agent, so it can run in other agent CLIs as well. Prepare common prerequisites: on the PATH, set up , (same as above). - (Verified) — Directly reads the .claude-plugin/marketplace.json of the repository:

Overview (machine translated)

Video Recap Skills

Turn one or several videos into Chinese commentary videos: Seven skills (six production + one on-demand reference) are loaded into your programming agent. Locally, you only need ffmpeg; remotely, just a Xiaomi MiMo key. The finished video can be exported into a clip draft and edited with one click.

Project homepage  ·  Installation  ·  Look at its output  ·  English

The above 59-second landscape commentary “This Second Goes Overboard” was selected from four short dramas, edited, written, dubbed, mixed, packaged, and revised multiple rounds of video review before final delivery; All its creative products (story planning, voice division, narration, timeline, QC reports, video review and revision records, Remotion packaging source code) are all present examples/guohuo-60s/, all excerpts below come from these files.

What is this?

Seven skills (six production + one on-demand reference) are loaded into Claude Code, Codex CLI, OpenCode, or OpenClaw. You use natural language to provide the video path and desired footage, while the Agent understands the visuals and dialogue. Decide on story and audiovisual plans, editing, writing, dubbing, mixing, and subtitles. Supports .mp4 / .mov / .mkv / .webm.

  • **One key, only ffmpeg needs locally. ** ASR, VLM, TTS all run Xiaomi MiMo; locally only the Python standard library and ffmpeg are used, no need for GPU, no need for pip install, and no model download. Dubbing can be replaced with Fish Audio, only replacing this dubbing segment.
  • **Make creative decisions first, then assign sounds. ** When planning for the first time, the Agent should consider a feasible editing direction; If a specified structure or feedback is provided, revise according to the current decision. Write audience commitments, POV, dramatic issues, and changes for each beat in recap_story_plan.json, then assign a visual task and sound ownership to each beat: narration only provides full dubbing when a clear task is given; strong dialogue, action sounds, or silence can fully dominate a beat.
  • **Edit first, then compose, with natural timeline alignment. ** Editing mode first cuts long videos into clips, then writes commentary on them; Multiple videos can be uploaded at once, and selected by source_id to create a single main storyline; Video analysis can be saved to the file system material library as needed, and later reused when matching sources and settings.
  • **You can continue editing beyond the final video. ** Multi-track timeline timeline.json allows one-click export of clip drafts, including original footage, narration, BGM, subtitles, and image overlays; Comes with an accurate subtitle file as the preferred source for original subtitles.
  • **Each step leaves verifiable records. ** Voiceover lint, assembly QC, and final QC generate machine-readable reports; Video modifications can additionally record changes made in this round, frozen items, and verification results.

Installation

ClawHub

ClawHub entries and dependencies are managed by the Explicit Release List.

This page was translated automatically. Technical details may differ from the original.

Recommended Tools

Try a different keyword or remove a filter.

Install

npx skillfish add worldwonderer/video-recap-skills
New to MCP? Start with the mcpskill.org guide →