I added the latest material to the dataset I used to train SD 1.5 and created an SDXL version!
It was created with AnimagineXL-V3-Base, so you should be able to use the model with AnimagineXL-V3.
Lastly, we recommend using Dynamic Thresholding (CFG Scale Fix)!
I changed the sampling stage from 30 to 50, and the quality has improved a lot. So I think it would be good to do it around 50