A style LoRA that combines several of my previous experiments into a single model.
The main trigger word is u3da, which activates the overall 3D-style look. For the pony version, use 3D, for Illustrious, it is not necessary, but it may improve the result.
Additional tokens allow you to control specific parts of the style:
u3da_figurine – anime figurine style
u3da_doll – anime doll style
u3da_render – 3D anime render look (Blender-like renders)
u3da_curvy – more exaggerated body proportions
u3da_gradbg – simple gradient background
Accordingly, you can add the parts you don't want to see to the negative prompt.
Older versions used different trigger tokens. See the changelog for details.
Description
Trained on Illustrious with softer parameters. More versatile and stable.
The dataset has been updated: common trigger tokens have been replaced with unique ones, which has reduced the influence of the base model. Training parameters have also been optimised. This has led to a stronger manifestation of style and slightly improved generation stability.