W4A8_mixed models quantized by me. (Makes my mixed int4 models obsolete) True int4 model size with int8 activations, near int8 quality, same speed as int8. For RAM/VRAM-constrained systems. Update to latest comfyui for asym_w4a8 support
25/06/2026 UPDATE - Replaced the models with versions that work with the new native implementation of INT8 in comfyui, for use with the native Load Diffusion Model node, Make sure you update you ComfyUI (Still works with the custom node but using native loads the model a bit faster for me)
Simple workflow Embeded into images
Krea 2 Int8 ConvRot quant.
For use in ComfyUI with https://github.com/BobJohnson24/ComfyUI-INT8-Fast or any other INT8 Custom node
Krea 2 is licensed under the Krea 2 Community License Agreement. For more information, visit https://krea.ai/krea-2-licensing.
Description
FAQ
Comments (5)
This is the Krea 2 Raw checkpoint, its not recommended for inference use. What are the advantages of using this to generate content?
You can use cfg/negative prompts if you use the turbo Lora with raw
quality i assumed
If you generated an image with a Turbo model, you shouldn't use Turbo with a Hi-Res Fix; you should use a non-Turbo (Base or Raw) model. That's because Turbo has very few steps (typically 8), and Hi-Res Fixes can use very low denoise values (eg. 0.3-0.4). The HRF KSampler doesn't have enough 'power' (steps * denoise) to get rid of the upscaling artifacts. A Base/Raw model does, with ~20 steps, does. The low denoise guarantees that the final image still looks very much like the one the Turbo model generated.
@tsolful theres some official turbo lora for raw?