Prompt Studio Model Loader

Prompt Studio Model Loader is a drop-in replacement for ComfyUI's standard diffusion-model loader that quietly picks the right loader for your file. Most of the time that means the normal UNETLoader. But when the model's safetensors header says its weights are INT8, it switches to Load Diffusion Model INT8 (W8A8) with the fixed Krea 2 defaults - the loader you'd otherwise have to hunt for and wire up by hand. The node's own description says it exactly: "Automatically selects the standard ComfyUI diffusion-model loader or Load Diffusion Model INT8 (W8A8)."

Its real job is the same as the LoRA loader's: be the attachment point Prompt Studio uses to swap models from the chat UI. Set it up with a model_type folder name, and Studio's Model selector appears in the sidebar, offering only the diffusion models inside that folder. When you pick one, Studio injects the choice into the queued workflow; the selection is recorded with the generated image so you can restore it later.

How it works

Two inputs, one output:

• model_type - a top-level folder name under your ComfyUI diffusion-models directory (for example krea2). Only models inside that folder are offered in Studio; the match is case-insensitive and includes nested folders. Leave it empty and the node acts as a plain loader.
• unet_name - the actual diffusion model, as a dropdown populated from your diffusion_models folder. Outside Prompt Studio, whatever you select here is loaded; when model_type is set, the name is validated against that folder.

The output is a single MODEL, same as any loader, so it slots in where your checkpoint/UNET loader used to be.

The INT8 detection is the interesting part. The node reads the safetensors header to check the weight dtype. If it's INT8, it routes through Load Diffusion Model INT8 (W8A8) (OTUNetLoaderW8A8 in the node registry) using the fixed Krea 2 settings - that's the path that makes quantized Krea 2 checkpoints work with the same graph as full-precision ones. Otherwise it uses the standard loader. This matters because the open Krea 2 scene shipped quantized weights alongside the raw ones, and mixing loaders up is an easy way to get silent quality or crash failures.

Installing it

Same pack, same install as the rest of Prompt Studio:

cd ComfyUI/custom_nodes
git clone https://github.com/tiko13/ComfyUI_PromptStudio

Restart ComfyUI, or search "ComfyUI_PromptStudio" in ComfyUI Manager. No pip dependencies for the pack itself - but there's a real dependency lurking: the INT8 path requires the ComfyUI-INT8-Fast custom node (that's where OTUNetLoaderW8A8 lives). If a model needs it and it's not installed, the node raises a clear error instead of silently loading the wrong way.

Where people get burned

• "Requires Load Diffusion Model INT8 (W8A8) but ComfyUI-INT8-Fast is not installed." Install ComfyUI-INT8-Fast (or use a non-INT8 checkpoint). This is the one extra dependency in the whole pack, and it only triggers for INT8 models.
• Model won't appear in Studio. model_type doesn't match a real top-level folder, or the file isn't under it. Same scoping rule as the LoRA loader.
• Expecting it to download models. It doesn't. Like every ComfyUI loader, the files have to be in your models/diffusion_models folder already.