# Neyralfoundry Runner

Three paths to turn a Studio bundle JSON into a real cinematic render.

## Path 1 — ComfyUI (easiest if you already have it)

1. Open ComfyUI.
2. Load → select `comfyui_workflow.json`.
3. Replace the `CheckpointLoaderSimple.ckpt_name` with your local video diffusion model (HunyuanVideo, CogVideoX, etc.).
4. Open your Studio-exported bundle JSON in any text editor. Copy:
   - `prompt` → CLIPTextEncode positive prompt node
   - `seed` → KSampler seed
   - `cfg` → KSampler cfg
   - `steps` → KSampler steps
5. Queue prompt. Render lands in `output/` folder.
6. Burn watermark via ffmpeg:
   ```
   ffmpeg -i input.mp4 -vf "drawtext=text='made with neyralfoundry.com':fontcolor=white@0.55:fontsize=22:x=w-tw-30:y=h-th-30:box=1:boxcolor=black@0.18:boxborderw=8" -c:v libx264 -crf 18 reel.mp4
   ```

## Path 2 — Python runner (if you have diffusers / HuggingFace)

```bash
pip install diffusers torch accelerate
python neyralfoundry_runner.py bundle.json --out reel.mp4
```

The runner is a thin glue. The `render_with_diffusers()` function inside is a placeholder — adapt it to your local pipeline. Watermark is applied automatically.

## Path 3 — Native runtime (waitlist)

We are building a one-command CLI runtime that bundles a chosen model and renders without configuration:

```bash
neyralfoundry render bundle.json --out reel.mp4
```

Ships in beta. Join the waitlist through the contact form at neyralfoundry.com/apply.html.

## Watermark obligation

The visible "made with neyralfoundry.com" identifier must remain on outputs you publish or distribute, until you have a separate written agreement with us removing this requirement. See https://neyralfoundry.com/legal.html for the full posture.

## License

MIT for these scripts. Your model weights and any output rights are governed separately by their respective licenses.