This is an attempt to produce 8 Step LoRA that produces better quality images (Compared to the various loras by Viggle and Pruna AI). When I use these three LoRAs in the Recipe together in ComfyUI, I get good quality images. However, the 8 Step LoRA I produced is not as good, but it works!
Recipe
Loras From Viggle: https://huggingface.co/Viggle/Qwen-Image-2.1-viggle-turbo
And Pruna AI https://huggingface.co/PrunaAI/Pruna-Qwen-Image-2.1/tree/main
v0.21 LoRA: Qwen-Image-2.1-viggle-turbo-v0.2.1-6step-lora-r128.safetensors -> 0.75
v0.1 LoRA: Qwen-Image-2.1-viggle-turbo-4step-lora-r64.safetensors -> 0.15
Pruna AI 8 Step LoRa: p_qwen_image_2.1_8step_v0.1.safetensors -> 0.2
Merged, Then exported.
Caveats
Can produce anatomical errors.
Can produce grid noise.
Description
FAQ
Comments (2)
I've found a lot of time Q2 needs CFG higher than 1.0, but that kind of makes the turbo loras go to shit, doesn't it?
That's true.