Lora trained on FluxGym with 25 images 1500 steps total trained on the jib model, works fairly well on that model have yet to try it on a different base model.

https://civarchive.com/models/686814/jib-mix-flux

Wonyoung is the trigger word.

DEIS Sampler and Beta scheduler, 25 sampling steps and 1.5 CFG scale, 0 distilled.

Normally start by doing the main prompt at a strength of 1 - 1.5 to get the basic shape of her face and scenario I want her in and then inpaint the face and body to fix everything else.

I usually inpaint at 1.2 to 1.35 but experiment to see what works best for you.

I use Forge UI for all my generations, unsure of how it would work in other UI's but let me know.

Another note, it works really well in inpainting on IRL photos as well.