RealVisXL Lightning Fill LoRA
Base model: SG161222/RealVisXL_V5.0_Lightning
ControlNet: OzzyGT/controlnet-union-promax-sdxl-1.0
VAE: madebyollin/sdxl-vae-fp16-fix
This LoRA was trained with random infill/outpaint masks:
- white mask pixels = fill region
- black mask pixels = preserved region
- ControlNet conditioning image = source image with masked region blacked out
Training files: 174
Validation files: 44
Resolution: 1024
LoRA rank: 8
Target modules: to_q,to_k,to_v,to_out.0
Mixed precision: fp16
