Hugging Face's logo

ByteDance
/
Hyper-SD

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

cfg-sdxl lora seems working poorly in animatediff

#30
by superag - opened
Discussion
superag

it works out with great results with image generation, but very flickerly video result in animatediff with the sdxl animatediff model.
tried different samplers and different cfg values, all the same...

image.png
Yanzuo
 ByteDance org

Hi, @superag
Is the normal LoRA working?
We will try figuring it out.
superag

Hi, @superag
Is the normal LoRA working?
We will try figuring it out.

normal 8 steps works just fine

image.png
Yanzuo
 ByteDance org

Hi, @superag
Can you share the example scripts / workflows of these two samples with us?
It's not expected.
superag

this workflow will show clearly the compare results. changing animatediff beta scheduler to linear SDXL will help a bit with the stableness, but you can still see the output being highly unstable compare to normal 8 step lora
https://drive.google.com/file/d/1Rzz7lZEQrQ3fOvQfBhUie1pjkeK-O2nL/view?usp=sharing

Sign up or log in to comment