Hugging Face's logo

ByteDance
/
Hyper-SD

Text-to-Image
Diffusers
lora
stable-diffusion
flux
Model card Files
xet
 Community
82

SD15 + 8steps-lora works good, but SD15 + 8steps-CFG-lora generate terrible results

#32
by Million - opened
Discussion
Million

use the workflow : Hyper-SD15-Nsteps-lora-workflow.json

SD15 + 8steps-lora :

image.png

SD15 + 8steps-CFG-lora

image.png
Yanzuo
 ByteDance org
•
edited May 7, 2024

Hi, @Million
You would need to set the cfg scale (5~8) for CFG LoRA.
Million

@Yanzuo Thanks, it works fine now, still not good at complicated prompts but I guess it's a limitation of the base model.
renyuxi changed discussion status to closed

Sign up or log in to comment