← Ash FoundryHugging Face Lane
GitHub Repo

Diagnostics Artifact

Hugging Face Video Tester

A browser-facing tester and documentation surface for Hugging Face video generation routes, especially the newer pattern where direct hf-inference model endpoints fail but a Hugging Face Space with exposed Gradio endpoints still returns a real MP4.

Video focus Space API aware Client-side tester

What this page is for

This tester exists because Hugging Face video access turned out to be structurally different from the image lane. The obvious text-to-video models on the hub were visible, but the direct hf-inference provider path returned unsupported-model errors. A different route ended up working: a Hugging Face Space with exposed Gradio API endpoints.

The current core question is: which Hugging Face video routes are truly reachable from this environment right now?

Seeded candidate logic

  • Lightricks/ltx-video-distilled , confirmed working via Space Gradio API
  • ali-vilab/modelscope-damo-text-to-video-synthesis , visible on hub but unsupported by current hf-inference path
  • ali-vilab/text-to-video-ms-1.7b , visible on hub but unsupported by current hf-inference path

Run video route tests

How it works

  • Direct model routes: tested against https://router.huggingface.co/hf-inference/models/{model} and expected to fail or succeed as raw provider checks.
  • Space routes: tested against Hugging Face Space Gradio endpoints such as /gradio_api/call/text_to_video.
  • The browser submits the job, then listens to the returned event stream until completion or error.
  • If a route returns a real MP4, the page embeds it inline as a preview.

Interpretation guide

  • 200 + MP4 preview: the route is genuinely live.
  • 400 unsupported by provider: visible on the hub, but not available through the current hf-inference path.
  • Space event error: route exists, but the generation failed at runtime.
  • Space event complete: best current pattern for Hugging Face video recovery.