5/24/2026 - Updated v3.0 file to fix a PyTorch size mismatch error.
Grounded generates realistic images of realistic women. Not another merge of merges, Grounded is made from trained checkpoints, particularly ones that specialize in amateur photography, and, therefore, amateur women. It excels in believable faces, realistic breasts, and diverse body types.
All samples were generated without embeddings, loras, or upscaling.
Recommendations:
Clip skip: 1-2
CFG: 6-8 (4-6 for v3.0)
20-40 steps
512-768
DPM++ 2M Karras, DPM++ SDE Karras, UniPC
Avoid negative embeddings
Description
Used some add difference merges to basically strip out everything professional and glossy - everything that made it like Elevated. What I got, I think, is a distilled version of my original vision for the checkpoint. I’ve had so much fun with it that I actually decided to upload again. SD1.5 may be limited and old, but dammit, so is my hardware!