MiaoBi Checkpoint Loader (Diffusers)

MiaoBi is a Chinese-language SD1.5-architecture checkpoint, and this pack's own README offers two ways to get it running. Route one: hand-pick the CLIP safetensors and the UNet safetensors, rename them, and drop each into the matching ComfyUI folder - that's what the MiaoBi CLIP Loader plus a normal checkpoint loader gets you. Route two, the one this node handles: clone or download the entire Hugging Face repo as-is into ComfyUI/models/diffusers and load the whole thing in one shot. Less manual renaming, more disk used - you pull the tokenizer files and repo metadata along with the weights, even though ComfyUI only needs a subset of it.

Community footprint on MiaoBi specifically is thin - don't go in expecting a lot of workflow examples or forum troubleshooting for this exact model.

What it does

Diffusers-format loaders read a folder laid out the Hugging Face diffusers way - subfolders like unet/, text_encoder/, vae/, tokenizer/ - rather than a single merged safetensors file, and assemble ComfyUI's model objects from that structure directly.

• model_path - a dropdown of folders you've placed under ComfyUI/models/diffusers. Point it at wherever you cloned the MiaoBi repo.

Outputs, all in one go:

• MODEL - the diffusion UNet
• CLIP - the text encoder
• VAE - the matching VAE

That last one is worth flagging: most other loaders in this pack (and in ComfyUI generally) hand you a MODEL and CLIP and expect you to load a VAE separately. This node bundles all three, so you can skip a VAE Loader node entirely for this particular pipeline.

Installing it

Part of the whole ComfyUI_ExtraModels pack:

• ComfyUI Manager - search "Extra Models for ComfyUI", install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/city96/ComfyUI_ExtraModels, then pip install -r requirements.txt inside your venv. Restart.

Then clone MiaoBi's Hugging Face repo (ShineChen1024/MiaoBi) straight into ComfyUI/models/diffusers, keeping the repo's own folder structure intact. Restart ComfyUI and the folder should show up in the dropdown.

Common issues

This path needs the diffusers Python package, which ships in the pack's requirements.txt - if it's missing from whichever venv ComfyUI is actually running on, you'll get an import error at startup or node-load time rather than a clean "model not found," which can be confusing if you assumed the pack installed cleanly.

If model_path shows up empty, the loader didn't find a valid diffusers-structured folder under models/diffusers - check that you cloned the repo with its subfolders intact rather than dumping loose files flat into the directory; a flat dump won't be recognized.

If disk space is tight, this route is heavier than it needs to be - the CLIP-and-UNet-only path (MiaoBi CLIP Loader plus a checkpoint loader) pulls just the weights you'll actually use.