The model training is based on 2TGB of MMD video, the data has been cleaned by the open source aesthetic scoring and data cleaning toolset (AAV) produced by L_A_X.
https://huggingface.co/spaces/Laxhar/anime-thetic
Animagine XL V3.1 was used as training base, which was tested to be better than the 1.5 version of MMDlora.
This model can generate an MMD model with a fixed style, and effect is even more amazing.
