What is the difference between SRPO-Refine-Quantized-v1.0 and SRPO-for-ComfyUI
It would be helpful if I know the difference before downloading
@bhanukhan , SRPO-for-ComfyUI is offical version, only convert it's transformer keys from diffusers to ComfyUI and did quantized; SRPO-Refine-Quantized is my refine version, generate image more clarity and less steps compare with offical SRPO.
In offical repo, the "diffusion_pytorch_model.safetensors" file is fp32 and diffusers transformer format, the keys is different with Flux in ComfyUI, then can't load and quantize by "diffusion model loader" node directly.
@bhanukhan , SRPO-for-ComfyUI is offical version, only convert it's transformer keys from diffusers to ComfyUI and did quantized; SRPO-Refine-Quantized is my refine version, generate image more clarity and less steps compare with offical SRPO.
what 'refine' you did to model ? i see it has very different composition and elements from same seed of original srpo , it try to add detail but backward it give that old flux not real elemnts 'if the object image should be static and it well known' , we can say your refine good for art and illustration more 'still realistic render' ? or it try to add detail ..randomly !?