Inline Studio Registry

The list of published extensions & models

Publishing an extension

1. Tag a release in your extension repo (e.g. v1.0.0). Installs pin to the commit behind the tag.
2. Add registry/<your-id>.json here, one file per extension so PRs never conflict:

{
  "id": "your-id",
  "name": "Your Extension",
  "description": "One line on what it does.",
  "repo": "https://github.com/you/your-extension",
  "author": "You",
  "tags": ["image"]
}

No version goes here. A listing names your repository; Inline Studio resolves the newest release tag at install and when checking for updates. Publish a new version by tagging it - you never open another PR here.

3. Open a PR. id must match both the filename and your manifest's id.

What CI checks

Your repo is cloned at the tag and put through the same checks the app runs at install time:

• the manifest validates, and its id matches this entry;
• the security scan finds nothing CRITICAL: declaring torch/diffusers/numpy as a dependency, exec over an encoded payload, a setup.py, or bundled CUDA/torch binaries all fail the build;
• HIGH/MEDIUM findings (subprocess, sockets, unrecognized network hosts) are reported as warnings. They don't block publication, but users must approve them at install.

index.json is generated from the validated entries; don't edit it.

What this is not

Listing here is not an endorsement or a safety guarantee. Extensions run in the same process as Inline Studio and can do anything it can. Review the code you install.

Model registry

The list of models Inline Studio shows under Settings → Models.

One file, models.json, and everything in it is verified. An unverified model needs a channel of its own so the default list stays trustworthy; add models.dev.json back when there is one to put in it, and point INLINE_MODEL_REGISTRY at it to see them.

Adding a model

Add an entry in models.json:

{
  "id": "flux-2-klein-4b",
  "label": "FLUX.2 Klein 4B",
  "filename": "flux-2-klein-4b.safetensors",
  "category": "diffusion_models",
  "group": "flux-2-klein-4b",
  "precision": "",
  "source": {
    "kind": "hf_file",
    "repo": "Comfy-Org/flux2-klein-4B",
    "path": "split_files/diffusion_models/flux-2-klein-4b.safetensors"
  },
  "verified": true,
  "size_bytes": null,
  "updated": "2026-08-17"
}

Run python scripts/validate_models.py before opening a PR.

What the app does with this

It resolves filenames. When a graph, a node or a training run names a file that is not on disk, the app looks the name up here and offers the download.

It never uses this to decide what a checkpoint is. A model is identified from its own tensor shapes, because diffusion_models/ is shared across architectures and two encoders can have identical shapes. A file listed under the wrong category, or a filename that does not match what the repo actually serves, will download and then fail at generation. Get those right.

In case you want a model listed here, open a PR.