🌊hua_gradio Checkpoint Loader

Hua_CheckpointLoaderSimple (🌊hua_gradio Checkpoint Loader) is the checkpoint loader you use when you want the person on the other end of your packaged webui to be able to pick the model themselves. Functionally it's a straight rename of ComfyUI's CheckpointLoaderSimple - same three outputs, same loading path - with one extra name input. The point of the rename is that the pack's Gradio frontend scans saved workflows for nodes by their exact class name, and this one gets a "choose checkpoint" dropdown in the UI. It's a wiring hook, not a new engine.

The mechanism is standard ComfyUI: ckpt_name is a dropdown populated from your models/checkpoints folder, and on execution it loads the checkpoint with load_checkpoint_guess_config, returning the diffusion MODEL, the CLIP text encoder, and the VAE. Same contract as the stock node, so you can swap this in anywhere CheckpointLoaderSimple already lives - which is exactly what you want when converting an existing workflow into a packaged one.

Inputs that matter:

• ckpt_name - pick from checkpoints you already have. This is what the Gradio dropdown controls.
• name - label; defaults to "Hua_CheckpointLoaderSimple".

Outputs: MODEL, CLIP, VAE. Wire them exactly as you would from the stock loader: MODEL into sampler/LoRA, CLIP into the text encoders, VAE into decode.

Install: ComfyUI Manager → "ComfyUI_to_webui", or

cd ComfyUI/custom_nodes
git clone https://github.com/kungful/ComfyUI_to_webui.git

then restart. No model files are downloaded by the pack - it uses whatever checkpoints you already have, and the dropdown only shows those. If the dropdown is empty in the webui, your checkpoints folder is empty or the "刷新模型" (refresh models) button hasn't been pressed.

A few honest notes. First, this node is a singleton in the frontend: exactly one checkpoint dropdown per workflow, no matter how many of these you place. Second, it's for combined checkpoint files. If you're running Flux-style split models, the sibling 🌊hua_gradio_UNET Loader is the one you want, since those checkpoints ship as separate UNet + CLIP files and there's no VAE output to speak of. And third, the model-caching means switching models in the UI costs a load but subsequent runs reuse the cached one - normal ComfyUI behavior. If all you need is a static model in a packaged workflow, the stock loader plus a Primitive works fine; this node earns its keep when the model is a choice the user gets to make.