edenfunf/reelmimic

ReelMimic – AI‑powered video‑style remix

What it does

• You give ReelMimic a reference video (file, phone capture, or YouTube link) and a short description of the new video you want.
• The system analyses the reference, extracting rhythm, shot lengths, transitions, colors, framing and camera moves.
• It then generates a plan – a storyboard with suggested characters, assets and style frames – which you can review and edit via an integrated chat.
• Once approved, up to six AI agents work in parallel to create each shot. After a shot is rendered, a different agent reviews it, so every piece is double‑checked.
• All fixes are shown with before/after screenshots and must be approved before the video is considered finished.
• The whole pipeline runs locally, using your own Claude Code or Codex API keys, so the cost is billed to your account.

Key technical points

• 2‑D only – seven built‑in rendering engines (vector/motion‑graphics, hand‑painted watercolor, crayon picture‑book, pixel art, paper cut‑out stop‑motion, whiteboard doodle, anime cel). New styles can be added with a simple Markdown file.
• Workflow – reference analysis → plan generation → multi‑agent production → review → final video. Logs, agent thoughts and screenshots are visible throughout.
• Dependencies – Node 22.18+, Python 3.10+, FFmpeg, Chrome, and an Anthropic Claude Code or OpenAI Codex account.
• Platform support – primarily tested on Windows; macOS/Linux should work.
• Optional posting – finished videos can be sent to TikTok, Instagram, YouTube etc. via the external Upload‑Post service.

Typical usage

1. Install with the provided install.sh/install.bat script.
2. Open http://localhost:4318, drop a reference video, write a one‑line brief, and pick Claude Code or Codex.
3. Review the automatically generated storyboard, tweak via chat, then hit Approve and start.
4. Watch the production line tab as agents render and review each shot.
5. When done, add any final notes and optionally post the video.

Why it matters ReelMimic demonstrates a practical, end‑to‑end AI‑agent orchestration for creative video generation. It combines video analysis, prompt‑driven storyboard creation, multi‑agent parallel rendering, and human‑in‑the‑loop review, all on a user’s own hardware. This makes high‑level style‑transfer for short animated videos accessible without needing large cloud‑hosted pipelines.

License MIT for the core code; third‑party assets keep their original licenses.