Load Diffusion Model (Quantized, Simple)

The Simple variant of Load Diffusion Model (Quantized) is the one you'll actually use if you're not in the weeds: it's the full loader with the quant_format and kernel_backend dropdowns stripped away, hardwired to auto detection and the PyTorch backend. Pick the file from diffusion_models, it figures out the quant layout itself, and you get a MODEL out the other side.

This is the right node when you grabbed a quantized UNET from silveroxides' HuggingFace - int8, fp8, whatever - and just want it feeding your KSampler without thinking about formats. The full loader's format dropdown only earns its keep when you're forcing a specific layout or diagnosing a file that misbehaves under auto-detection.

The inputs that matter

• unet_name - the quantized diffusion model from your diffusion_models folder.
• disable_dynamic and low_memory - the pack's shared loading toggles (low_memory tooltip: fast low-impact loading, or ComfyUI's default when off).

Output is a single MODEL into your KSampler. CLIP and VAE are separate loaders for separate files.

Installing it

ComfyUI Manager → search "ComfyUI-QuantOps", or:

cd ComfyUI/custom_nodes
git clone https://github.com/silveroxides/ComfyUI-QuantOps

Restart and let requirements.txt pull in unifiedefficientloader>=0.5.2.

The catch

Auto-detection only recognizes the pack's own formats - int8, tensorwise int8, the fp8 variants, mxfp8, nvfp4. A file quantized by anything outside the convert_to_quant toolchain won't load just because this node is called Simple; you'll likely get a format error that the full loader at least lets you poke at. And the pack is deprecated: int8 ConvRot is native in ComfyUI (v0.27.0+) now, the author stopped maintaining, and on a current ComfyUI the stock Load Diffusion Model handles modern formats. Keep this node for the older quantized UNETs already on disk - it does exactly one job, and it does it without ceremony.