Silly experimental model. Trained off the pixel art portraits from Capcom's Children of the Atom. Dataset is small so inference should be up in the air.
This lora is very base model sensitive and using finetunes etc will likely cause it to lose most of its effect. As you can see from the samples, it's also pretty inconsistent so you are likely to have to try multiple times for any given image. Overly trained character loras may overwrite the "comic superhero" stylization with their own (usually anime) style. That's what happens if you have a very small sample size.
Directions: Generate at 1536 x 1536, downscale to 96x96 (original training data size), upscale it back up if you want.
Suggested Weight: 1.0 (You can try to go a bit higher like 1.2 but the pixel art gets rougher)
Suggested Resolution: 1536x1536
Trigger Word: CCOTA, pixelart, simple background, pink background, close-up, portrait