About the model

It was trained partially on synthetic set with images in style just similar to the units' portraits from the game series Disciples II. Paradoxically it's less similar to the style from the game than my model trained on SD 1.5, it's more painting-like, and with less visible 3D model under painting edits. I personally like it more, even if it's less similar to the game style, it still has its vibe.

Generation tips

LoRA trigger / modifier:

This is a digital painting of a fantasy character from Disciples2

Start your prompt with:

Fantasy portrait of {subject: human, elf, goblin, etc.}

• Do not specify background other than simple color or gradient. If you do it it will become less Disciples-like.

• Works better for portraits only, but it can generate whole character if specified in the prompt

• It was trained for singular character only, with not work well with multiple characters at once or landscapes

• Don't work for NSFW generation if your base Krea 2 model isn't "unlocked"

• Prompting subjects that aren't fit to the category of "fantasy character" probably result in generation in different style

Known problems:

Sometimes it will produce glitched / smudged edges because of problems with training set (original portraits had frames). In the most cases another generation should be fine. I'll fix it in another version.