This is a little lora I trained overnight with a dataset of 130 images, it requires Flux-1dev, but it kinda works with Flux-Schnell as well.