Prompt with character name (without space) + umamusume girl:
As an example, the positive prompt for the following grid is masterpiece, artwork of [...] umamusume girl in rain, best quality, detailed, beautiful eye
I don't remember the negative prompt
Training details:
- Trained with kanewallmann Dreambooth repository using character name + umamusume girl as caption
- Data set: around 200 images for each character + thousands of customized reg images
- Trained to 26500 steps (probably at the default learning rate 1e-6) but the 22500 step version may be better
- I don't remember the remaining details
Problems:
- The generated images tend to have high contrast
- It often forgets the horse tails
More example generations:
