This model is best at producing landscapes and art pieces rather than game art or stand alone characters. Trained on 100+ 1k images where each 'pixel' is 4x4 pixels.
Trigger word:
pixel art
I included several versions of the Lora I was testing.
V1 - Illustrative shaded. This version tends to be the most 'pretty'. The images have more details and shading. It's also pretty flexible.
V1.1 - With game art. I added in game art maps to the model to make it more versatile, though it made the outputs slightly less aesthetic.
V1.2 - Most stylized flat. This version I let 'cook' longer. It is a better representation of the training data. It is more flat, colorful, and true to the pixel art style.
Most generations won't have outlines unless "outlined" is included in the prompt.
Additional tags for perspective:
top down
isometric
side perspective
Description
This version includes game art in the training data, making it more versatile. However you may want to use the 1.0 version if you want it to look more illustrative.