coldframelisted
Install: claude install-skill Razee4315/coldframe
# Cloud renders with coldframe
coldframe renders Remotion compositions on free cloud machines. Two backends:
| Backend | Use for | Speed |
|---|---|---|
| **GitHub Actions** (default) | Final renders | Frames split across up to 20 machines in parallel |
| **Google Colab** via the `colab-mcp` MCP server | Experiments, GPU tests, very long single jobs | One 2-vCPU machine (sometimes a T4 GPU) |
Keep local work light: preview with `npx remotion still <Comp> --scale=0.25 --frame=N`, never full local renders unless asked.
## GitHub Actions (default)
Requirements: `gh` signed in, the Remotion project pushed to GitHub, and `.github/workflows/coldframe.yml` in it (add it with `npx github:Razee4315/coldframe init`).
1. Commit and push the changes to render. The cloud renders what is on GitHub, not local edits.
2. Run:
```bash
npx github:Razee4315/coldframe render <CompositionId> --chunks 8
```
This starts the workflow, waits (`gh run watch`), and downloads the MP4 to `out/cloud/`.
For long renders, use `--no-wait` and check with `gh run list --workflow coldframe.yml`, or run it in the background.
3. Check the result: extract a few frames with ffmpeg and look at them, and confirm the job summary says the frame count matched.
Useful inputs: `--props '<json>'`, `--name file.mp4`, `--ref <branch>`.
WebGL / three.js content renders with `--gl=swangle` on CPU runners (the workflow default).
If the `RCLONE_CONF` repo secret is set, the MP4 is also copied to Google Drive