Style LoRA trained on Commodore 64 "Petscii" art.

Training images were converted to use the RGB colors from this GIMP palette:

https://github.com/denilsonsa/gimp-palettes/blob/master/palettes/HW-Commodore-64.gpl

Color accuracy isn't perfect, the model will sometimes add extra shades.

Artist tags can improve composition but may affect color accuracy.

Recommended Settings

• Model: NoobAI-XL Epsilon-pred 1.0 (Other NoobAI-based models work but are less accurate)

• Resolution: 1280x800 (40x25 characters) or 1024x1024

• LoRA strength: 1.0

• Sampler: Euler A

• Scheduler: Align Your Steps

• Steps: ~30

• CFG Scale: 4-5

• CFG Rescale: 0.5-0.7

• Discrete Sampling Type: eps

• Zero SNR: Off

Example Prompt

<lora:petscii:1.0> petscii, masterpiece, best quality, solo, anthro, mouse, green goggles, mad scientist, obscured eyes, hands on head, explosion, open mouth, scared, black background

Negative prompt: worst quality, low quality