Unet Model Selector

If you run Flux, SD3, or any of the split-architecture models, you know the checkpoint isn't one file anymore - the UNet/diffusion backbone lives separately from the text encoders and VAE. Unet Model Selector is the pack's node for choosing that backbone: a dropdown over your models/unet folder that packages the pick as an IPT-Model value and lets the pack's metadata system remember it.

It's the same selector family as Checkpoint Selector - same design, same outputs, same right-click tricks - just pointed at a different folder. Which one you use comes down to one question: are you choosing a whole checkpoint (model + CLIP + VAE in one file) or a diffusion backbone on its own?

How it works

At execution, the node takes the filename from its unet_model dropdown, resolves it against what's actually in ComfyUI/models/unet, and wraps it as an IPT-Model value. It doesn't load the weights - that happens later, once, when the value reaches Load New Model. The loader then handles the accompanying CLIP and VAE however your workflow supplies them, since a bare UNet file contains none.

The outputs mirror the other selectors:

• model - the IPT-Model value for the loader pipeline.
• unet_name - a passthrough of the selected filename, handy for labels, XY-plot axes, or feeding metadata downstream.

The optional sha256 input feeds the right-click View Model Info... action, which looks the file up on Civitai by hash - a useful sanity check that the file in your unet folder is the exact release you think it is, and the reason you should leave that field alone unless you know what you're doing with it.

When to use it vs. Checkpoint Selector

The rule of thumb is boring and practical: use Checkpoint Selector when your model ships as a single .safetensors checkpoint (SD1.5, SDXL, most fine-tunes). Use Unet Model Selector when you're running the split-style architectures where the diffusion model is a separate file - Flux dev/schnell UNets, SD3-style setups, or any time you've downloaded just the backbone and are pairing it with separate CLIP and VAE choices.

The pack's Load New Model node understands both, and supports a vae_fallback for diffusion-model selections - so once you've got the value, the plumbing handles the rest.

The one thing that trips people up

Folder placement. This node reads ComfyUI/models/unet (there's also a diffusion_models folder the pack's Diffusion Model Selector uses - don't mix them up). If you've dropped a UNet file in checkpoints because that's what the download page said, it won't appear here. Move it into unet and restart ComfyUI so the file list refreshes.

Installing

Part of kinorax/comfyui-info-prompt-toolkit. ComfyUI Manager, search "Info-Prompt-Toolkit":

cd ComfyUI/custom_nodes
git clone https://github.com/kinorax/comfyui-info-prompt-toolkit.git
cd comfyui-info-prompt-toolkit
pip install -r requirements.txt

Restart, and it appears under Info-Prompt-Toolkit/ImageInfo. No extra model downloads - the node only references UNet files you already own. If you're running SDXL and don't use split checkpoints, you can ignore this one entirely and grab Checkpoint Selector instead.