This is an Extension for Forge Neo, which quantizes models via comfy-kitchen, or casts models to save space
Tip
- Use the Cast section to cast an arbitrary model (e.g. ControlNet) into
dtype - Use the Quantize section to quantize a diffusion model into
format
Note
Supported Formats: fp8_scaled / nvfp4 / mxfp8 / int8 / int8_convrot / convrot_w4a4
Important
Can only convert non-quantized models (i.e. fp16 / bf16)
Warning
This Extension does not include per-model config ; quality may not match other dedicated tools
Caution
This Extension is currently Experimental ; use at your own risk