RIFE model (rife49.pth) auto-download fails with 404 → workflow crashes after generation

[gcadoret]

Hi 👋

While using the ksampler branch (via the RunPod hub), the workflow fails during the ComfyUI-Frame-Interpolation (RIFE) step because the auto-download URLs for rife49.pth return HTTP 404.

The main Wan generation completes successfully, but the job crashes when VFI attempts to fetch the model.

⸻

Problem Details

During execution, the node attempts to download rife49.pth from multiple GitHub release URLs:
• https://github.com/styler00dollar/VSGAN-tensorrt-docker/releases/download/models/rife49.pth
• https://github.com/Fannovel16/ComfyUI-Frame-Interpolation/releases/download/models/rife49.pth
• https://github.com/dajes/frame-interpolation-pytorch/releases/download/v1.0.0/rife49.pth

All return HTTP 404, causing the pipeline to fail.

Relevant log:

Exception: Tried all GitHub base urls to download rife49.pth but no success urllib.error.HTTPError: HTTP Error 404: Not Found

Likely Cause

It appears that the hardcoded download URLs in vfi_utils.py are outdated and no longer host rife49.pth.

This results in:
• Successful Wan sampling
• Failure during frame interpolation
• Full job termination after compute time has already been consumed

⸻

Suggested Fix

Instead of relying on outdated GitHub release URLs, one of the following would make the workflow stable:
1. Update the download URL to a currently valid and maintained location
2. Host rife49.pth inside this repository (e.g., release asset or model directory)
3. Allow specifying a custom download base URL
4. Gracefully skip interpolation if the model cannot be retrieved

If the file already exists in the image or expected path, perhaps the fallback logic should check the local path first before attempting remote downloads.

⸻

Thanks for the project 🙏
Reporting this in case it helps improve stability for serverless deployments.