I have been unhappy the way tattoos tend to generate on flux models. This hopes to offer a solution.

This Lora is trained on 68 images generated on RealVisXL 5.0

Usage, so far I have tested this mainly on Schnell models, its output is varied depending on the model used.

So far I have had the best results on schnellMODE (FLUX.S) - FP8 using a fairly low weight 0.55 - 0.75.

Some models seem to need higher weights and active prompting, so you are going to have to experiment!

Anyway I hope this is of some use, and as always have fun!