DoRA is trained on images of the back of cards from the game Hearthstone.
Usage tips: I recommend using the DoRA weight between 0.7 and 0.85. Use card back and hearthstone logo to trigger DoRA. Describe the back of the card in detail and vividly.
Problems: Almost always DoRA in the background can draw the consequences of my editing the dataset. Try writing dark grey background in the prompt, sometimes it helps.
Example prompt: a card back that reflects the shield of the warrior of darkness, in the center is the purple glowing hearthstone logo, solid light grey background
Recommended model: DoRA is trained on the base SDXL 1.0 model, so I think it's okay to use this DoRA on any model that is based on SDXL 1.0 (except anime models). I used the Juggernaut XL checkpoint to generate the images.
Clip skip: The clip skip of the DoRA is 1.