LoRA trained on Gio Scotti. I've found it works best with RealisticVision.

All images are untouched generations- no img2img, controlnet, photoshop, or upscaling.

For reproducibility, use a VAE and Clip Skip 2.

Trained on Base 1.5

I've heard reports that choosing a specific checkpoint may increase quality, so sometime in the future I might make a version finetuned on RealisticVision or maybe some other model.