Load Magic Clothing Pipeline With Path&Diffusers

The sibling that skips the conversion step

This is the "already have it" variant of Load Magic Clothing Pipeline. Where that node takes a plain .safetensors checkpoint and converts it to diffusers format on first use, this one just points at a diffusers-format model directory and loads it directly. Same pack, same Magic Clothing diffusers pipeline underneath, same three outputs - minus the conversion wait.

You'll want this node if you already keep models in diffusers layout. Drop the official ShineChen1024/MagicClothing download (or any diffusers-format SD1.5 model) into ComfyUI/models/diffusers/<name>/, and it'll show up in the model_path dropdown. It also reads from the pack's own conversion/ folder, so if you've already run its sibling once, the cached conversion appears here too and you can reuse it without re-converting.

If you've only got .safetensors checkpoints and no diffusers folders sitting around, this node is just an empty dropdown that makes you feel dumb. Go use Load Magic Clothing Pipeline instead - that's the one that does the converting.

Inputs and outputs

Three inputs, and only the first is worth your attention:

• model_path - the dropdown of diffusers-format model directories. This is the whole job of the node.
• dtype - default, float32, float16, bfloat16. float16 is the standard pick for a consumer GPU.
• device - default, cpu, cuda, cuda:0, cuda:1. Leave it on default unless you're steering a multi-GPU setup.

Outputs are the same trio as its sibling: PIPELINE, AUTOENCODER, and SCHEDULER. They feed the same downstream chain - Diffusers Model Makeup &MC, then Load Magic Clothing Adapter &Diffusers, then RUN Magic Clothing Diffusers Model to get an IMAGE. This node is the loading dock, not the generator.

Install

Identical to the rest of the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/longgui0318/comfyui-oms-diffusion

Or find "comfyui-magic-clothing" in ComfyUI Manager and restart. Like every node in the pack's diffusers branch, this one imports diffusers and transformers at module load, and the repo has no requirements.txt, so a fresh portable install will likely need:

pip install diffusers transformers safetensors

And you still need the Magic Clothing UNet weights in ComfyUI/models/unet for the adapter node further down the chain. A path-based loader doesn't spare you that.

Gotchas

The dropdown only lists what's actually on disk - no folder, no entry, so if model_path looks empty, check you put a proper diffusers directory (with its model_index.json intact) under ComfyUI/models/diffusers/. And this whole branch is SD1.5-only, same as Magic Clothing generally. If the sibling node's conversion ceremony is what's putting you off, this one is the lazy path - but only if you have the folders to point it at.