This is a forked project of the "FurryArt"
* Merge Lab series *
About:
This is a Latent Consistency Model (LCM) merge. It is capable of producing very sharp images with 3 or 4 steps and cfg 1-3.
However, you can use a usual configuration, like: cfg 7; 20 steps.
It accepts "Sampling method" LCM and lora lcm as well.
LCM technical Suggestions:
If you use LCM technique follow this suggestions....
CFG: between 1and 3
Steps: 3-10 steps with best results in 6 - 10 steps
Sampling method: Euler, Euler a, UniPC, DPM++ SDE, DDIM, LCM will work well
Note: If you get weird or distorted images, try lowering cfg and or steps (this is a LCM merge model, and a LCM model work differently).
Description
This is a Latent Consistency Model (LCM) merge. It is capable of producing very sharp images with 3 or 4 steps and cfg 1-3.
However, you can use a usual configuration, like: cfg 7; 20 steps.
It accepts "Sampling method" LCM and lora lcm as well.
FAQ
Comments (5)
Fantastic model, looking forward to trying low steps with animation. ^^
Thanks for your comment!!
"with animation."
Could you teach me how to make an animation?
I would love to see some animated characters😀
@A_B I've only played with animation once or twice and it was using Animatediff, however the results were always awful. ^^
I get great and fast results with this model even using simple prompts, thank you! :)
Thank you