This is my first Flux LoRA and intended as a test before I would try to train other. I arranged rank models from best to worse. Trained on fp8 version of dev model and 512x512 resolution images.

64 rank is a bit overfitted, but suited if requires more accuracy in style. But can have some bad anatomy.

32 rank is the most optimal when it comes to style and characteristics.

8-16 ranks are most flexible. I recommend using those if you don't want a bleeding of style (or at least if it has this).

First images are all from the same settings.