V3:

Updated version with a new dataset for illustrious. It's a bit unstable, but can get good results.

Trained on realism but works also with 2D. depending on the checkpoint 0.4 strength is enough, but some checkpoints can go up to 1.

Mix up the trigger words and experiment with the strengths

Known issues: sometimes the tongue blends into the lower lip. you can usually fix that with "lower lip" in the prompt.

V1:

works with low strengths, 0.6 is enough. depending on the model you are using it helps if you add thick lips, big lips, large lips, detailed lips.