poor quality when used with picxReal
I'm using SD15 4Steps version with picxReal, however the result is quite noisy and blurry.
Hi,
@CHNtentes
Could you please tell us what sampler you are using right now?
Are the results getting better when using our 8-Step LoRA?
4 steps, cfg 1.0, ddim with sgm_uniform.
And with 8-step lora it is much better:
But there are still some artifacts especially in the face.
Hi,
@CHNtentes
These results should be expected.
We are currently working on CFG-Preserved Hyper-SD15/SDXL that facilitate negative prompts and larger guidance scales.
You may try removing these artifacts by the functionality of negative prompt after releasing.
This is picxReal without lora, 8 steps cfg 4.0. IMO it is cleaner than HyperSD.
Hi,
@CHNtentes
Could you try different lora weights? Our acceleration lora will inevitably change the base model.
Thanks for your attention.
This is picxReal without lora, 8 steps cfg 4.0. IMO it is cleaner than HyperSD.
@CHNtentes , we have uploaded the CFG-preserved hyper-SD15 LoRA and hyper-SDXL LoRA just now, higher cfg and negative prompts may be helpful in your usecase, looking forward to your use and feedback!