Sign In

Artist Style LoRA: Feng Zikai, 丰子恺

36
198
1.3k
6
Verified:
SafeTensor
Type
LoRA
Stats
198
1,274
9
Reviews
Published
Nov 3, 2023
Base Model
SDXL 1.0
Training
Steps: 1,560
Epochs: 2
Usage Tips
Clip Skip: 1
Trigger Words
Feng Zikai
Hash
AutoV2
E4AD9C5A9D

I am having issues uploading image previews to CivitAI, apologies. The photos here are some of the ORIGINAL work of Feng Zikai. For now the generated work can be accessed on this Github repo, unfortunately without generation parameters.

LoRA trained in style of 20th Century Chinese painter Feng Zikai. I find CLIP and model strength around 0.8-0.9 as well as CFG scale around 6.0 works fairly well, with higher CFG creating more saturated/cartoony images.

Trained on 39 images of his work, however these were fairly low resolution and the computer got upset so I had to blow them up. This can lead to a little less clarity in the images that I am disappointed in, so I plan to come back to this with higher quality data in the future.