It's a dream booth training based on stable diffusion 1.5. Dataset is consist of different cover art from dnd books
As far as i can see, too many tokens in negative prompt can make results worse, so dont go too much. Something like 1-3 tokens is ok
Previously was uploaded to huggingface -> https://huggingface.co/sd-dreambooth-library/dndcoverart-v1
You can support me and request Lora/Model creation at https://boosty.to/smgbuddies/donate