Kemmy Model Set Library

cn_04_kemmy_ComfyUI-Model-Set-Library

A ComfyUI custom node for saving and switching reusable model sets inside one model family.

What a preset stores

• Diffusion model (UNETLoader compatible)
• Weight dtype
• Text encoder and encoder type
• Text encoder device
• VAE
• Any number of LoRAs
• Per-LoRA enable state, model strength, and CLIP strength

The node outputs MODEL, CLIP, and VAE, so the rest of a workflow can stay unchanged while the complete loader configuration changes from one preset to another.

Intended use

Keep separate workflows for model families that require different latent, sampling, or conditioning structures. Use this node to switch presets within a compatible family—for example several Krea2 production sets or several MOSS-Image LoRA combinations.

Installation

ComfyUI Manager / Registry

After the first Registry release, search for Kemmy Model Set Library in ComfyUI Manager, install it, and restart ComfyUI.

Manual installation

Clone the repository into ComfyUI/custom_nodes, restart ComfyUI, and add:

Add Node → kemmy → Loaders → cn_04_kemmy_Model Set Library

No third-party Python dependencies are required.

Repository URL:

https://github.com/kemmy2302/cn_04_kemmy_ComfyUI-Model-Set-Library

Basic usage

1. Choose a diffusion model, text encoder type/file, and VAE.
2. Add zero or more LoRAs.
3. Set each LoRA's model and CLIP strengths. Use 0 for CLIP strength when a LoRA should affect only the diffusion model.
4. Click Save preset.
5. Connect model, clip, and vae to the existing workflow.
6. Select another preset with its radio button whenever a different production set is needed.

Storage

The shared preset library is stored at:

ComfyUI/user/kemmy_model_set_library/presets.json

The selected preset is also embedded as a snapshot in the workflow JSON. Model files themselves are never copied, moved, or bundled.

Compatibility notes

• The diffusion model must be loadable by ComfyUI's standard UNETLoader.
• The text encoder must be loadable by the standard CLIPLoader.
• The VAE must be loadable by the standard VAELoader.
• Presets do not replace model-specific latent or sampling nodes.
• Missing files produce the same clear loader error as standard ComfyUI nodes.

License

MIT

For maintainers: publishing a release

1. Update version in pyproject.toml. Registry versions cannot be overwritten.
2. Add the GitHub Actions repository secret REGISTRY_ACCESS_TOKEN.
3. Open Actions → Publish to Comfy Registry → Run workflow.
4. Confirm the new version appears in the Comfy Registry.

Do not commit the Registry token to this repository.