roderickgrc/agentic-edition-studio
Developer tools이 페이지는 AI로 자동 번역되었습니다. 기술적인 세부 사항은 원문을 확인하세요. 원문 보기
Teach an AI agent to edit video: perception (sureshot), Remotion editing, wiki-based judgment. By RodGrc.
개요
A starter for teaching an AI coding agent to edit video. Picture and sound become timed evidence. The edit is React. The judgment lives in files the agent can open one page at a time. This repository is the empty studio. It does not contain finished videos, personal notes, or a sound library. Editing is not only cutting. Someone looks, listens, and decides what is worth telling. An agent has four gaps for that job: Whatever fails in verify goes back to decide. A render with no error can still have bad cuts, drifting audio or empty stretches. - Node.js 18 or newer - Windows, with Python 3.10 to 3.13 for sureshot and the sound scripts sureshot installs the rest (ffmpeg, speech, OCR, Gemini key) with a guided command that asks before each step: doctor confirms everything works. Details: .agents/skills/sureshot/INSTALL.md. OCR uses Tesseract by default; PaddleOCR is optional (setup --ocr paddle).
README
원문 README의 일부입니다. 원문 보기.
Agentic Edition Studio
by RodGrc · built while making Haiku
A starter for teaching an AI coding agent to edit video. Picture and sound become timed evidence. The edit is React. The judgment lives in files the agent can open one page at a time.
This repository is the empty studio. It does not contain finished videos, personal notes, or a sound library.
Why it exists
Editing is not only cutting. Someone looks, listens, and decides what is worth telling. An agent has four gaps for that job:
| Gap | What goes wrong | What this repo does about it |
|---|---|---|
| Video | The model sees sampled stills, not motion. A 60 s clip at 30 fps is 1,800 frames; sending 1 per second skips 29 of every 30. | sureshot picks frames by events (cuts, motion, text) and keeps a timeline. |
| Audio | The model gets, at best, a transcript. | sureshot adds speech with timestamps, silences and signal events. |
| Editing environment | Editors are made for hands and eyes. | The edit is a Remotion (React) project: every cut is a line you can review. |
| Context | Your brand and rules do not fit in one prompt. | AGENTS.md, PROJECT.md and a wiki/ with an index, read page by page. |
Pieces
| Piece | Path | Role |
|---|---|---|
| Editor | this folder | Remotion project, one folder per video |
| Perception | .agents/skills/sureshot/ |
Speech, on-screen text, cuts, motion, silence in one timeline |
| Remotion rules | .agents/skills/remotion-best-practices/ |
How to write a composition |
| Judgment | wiki/ |
Index plus the pages that apply to the job |
| Sounds | assets/audio/ |
Shelves and a catalog shape. No audio files yet |
| Manual | AGENTS.md |
How the agent should work in any project |
Workflow
ingest -> understand -> decide -> produce -> verify
ffmpeg sureshot agent Remotion sureshot + you
(+ Gemini) + wiki + FFmpeg
Whatever fails in verify goes back to decide. A render with no error can still have bad cuts, drifting audio or empty stretches.
Requirements
- Node.js 18 or newer
- Windows, with Python 3.10 to 3.13 for sureshot and the sound scripts
sureshot installs the rest (ffmpeg, speech, OCR, Gemini key) with a guided command that asks before each step:
python .agents/skills/sureshot/scripts/sureshot.py setup --plan
python .agents/skills/sureshot/scripts/sureshot.py setup
python .agents/skills/sureshot/scripts/sureshot.py doctor
doctor confirms everything works. Details: .agents/skills/sureshot/INSTALL.md. OCR uses Tesseract by default; PaddleOCR is optional (setup --ocr paddle). A free Gemini key from Google AI Studio is only needed for the agent and digest commands.
The sound scripts need numpy:
python -m pip install -r scripts/requirements.txt
Media you add later (mp3, wav, mp4, png) is marked for Git LFS in .gitattributes. Install Git LFS before committing those files.
Quick start
추천 도구
다른 키워드를 입력하거나 필터를 제거해 보세요.
설치
npx skillfish add roderickgrc/agentic-edition-studio