← Ash FoundryHugging Face Lane
GitHub Repo

Diagnostics Artifact

Hugging Face API Tester

A browser-side testing surface for probing Hugging Face image routes with a manually pasted token, now biased toward models we have real evidence are supported by the current hf-inference provider path. For video-route testing, use the companion Hugging Face Video Tester.

Manual token entry Image-model focus Client-side fetch Companion video tester available

What this page is for

This tester is designed to stop us guessing. Hugging Face is a model hub, but not every model exposed on the site is actually usable through the same inference provider path for a free token. This page lets us test candidate image models directly from the browser and see which ones really return an image.

The core question is simple: which Hugging Face image routes are truly live for the current token and provider path?

Important lane update: Hugging Face video recovery turned out to require a different pattern. The direct image-oriented hf-inference approach documented here is not the same thing as the Space-based Gradio workflow that currently makes video possible.

Seeded candidate logic

The list now starts with models we have real support evidence for on the current route, then keeps a smaller set of adjacent candidates that still deserve probing.

  • black-forest-labs/FLUX.1-schnell , confirmed working
  • stabilityai/stable-diffusion-3-medium-diffusers , confirmed working
  • black-forest-labs/FLUX.1-dev , same family, but currently deprecated on this path
  • black-forest-labs/FLUX.1-Krea-dev , related FLUX family candidate
  • black-forest-labs/FLUX.1-Canny-dev , related FLUX family candidate
  • stabilityai/stable-diffusion-3.5-medium , adjacent Stability route worth testing

Run model tests

How it works

  • Your token stays in the browser tab and is not stored in the repository.
  • Each selected model is called through https://router.huggingface.co/hf-inference/models/{model}.
  • The test sends a basic JSON payload with inputs set to the image prompt.
  • If a model returns an image blob, the page shows a thumbnail preview inline.
  • If a model fails, the page shows the raw error text so we can distinguish unsupported, deprecated, gated, or quota-limited routes.

Interpretation guide

  • 200 + image preview: the route is genuinely live for the current token.
  • 400: often means the model is not supported by the current hf-inference provider path.
  • 401/403: bad token, gated access, or permission issue.
  • 410: deprecated route or removed provider support.
  • Timeout / fetch failed: unstable upstream or browser/CORS issue.