This version is undercooked compared with SDXL and Illustrious
Model description
This lora was trained on 302 pixel art images of The King of Fighters sprites of characters on victory portrait screen (Neogeo/MVS).
Results tend to has 6x6pixels since all dataset was upscaled to that factor, so you might be able to do pixel-perfect with the generated image by normalize it to a grid 6 times smaller.
(I fell the Flux version is undercooked since it had only 8 epoches because it take a lot longer to train. What I mean is, it doesn't have that ressamblance to SNK pixel art, but it performs well)