LoRA based on Marvel vs Capcom's character select portraits.

It will usually produce a complete face. The split portraits in my gallery here were cropped in post.

Generation is intended for 1024x1024, and then downscale by 4 for (sort of) a pixel art look. Unfortunately the pixel result is a bit messy, the training images were too big for 8x pixel scale training.

No trigger word, but recommended: portrait, looking at viewer, straight-on, facing viewer, blue background, simple background, (hatching \(texture\):0.8),

You may need to add "nude" to the negative prompt or define clothes in your prompt or you'll end up with a bare shouldered character.

Higher LoRA strength can induce a more comic book like look but can also make the character's skin turn blue for some reason.