Trained on better images, though because it’s a 1.5 base model it sometimes produces text, but rarely watermarks.
This was NOT trained at huggingface, and was our first FAST DREAMBOOTH collab notebook attempt!
Only reason this isn’t a “VERSION” is that any future retrains will be updated on THIS model not the Dreams one. That one may still get updates but may only get updated on AI outputs like it’s base.
Euler for the TEENY Specifics
SDE is good, and KARRAS mixes are good.
All up to how you prompt and guide the model.
Certain times it IS SPECIFICALLY adding text around edges, but it’s never adding the “STOCK IMAGES” watermark, i’ll see what I can do about a further retrain.