This is a reprint of the original CIVITAI project, please support the author - maDcaDDie
About this version
The LoRA was trained on Flux-Dev.
It also works with Flux-Schnell, but I behaves slightly differently.
The showcase images are made with Flux-Dev.
For Flux Dev I recommend the following setting - Lora strength 0.7 - 0.8 greatly depending on the image, highres fix with denoising 0.3
I didn't test it much with Flux-Schnell, but I noticed you get better results with 8-10 steps rather then the usual 4