pixel-diffusion-xl

For best outputs and usage:

• Sampler: Euler a

• Steps: 50, cfg 4 ~ 12, CLIP skip 2

• This model has its own VAE: XL_VAE_C1.safetensors.

• For the neg prompt: simple negs or just nothing will do. Example: "bad quality, bad anatomy, worst quality, low quality, low resolution, extra fingers, blur, blurry, ugly, wrong proportions, watermark, image artifacts, lowres, ugly, jpeg artifacts, deformed, noisy image, deformation, skin moles"

• For the pos prompt:

  • Keep the prompt simple, don't use extremely long prompts, focus on what you need

  • Use "Pixel Art" at the start or in the beginning part of the prompt, this way you can be sure to achive a pixel art type of image

  • Use words such as 16 bit, 32 and 64 bit to help the checkpoint understand the style if you are having a hard time getting pixel art

  • It's preferrable if you don't use words that imply a style that is very different from pixel art, especially in the beginning part of the prompt, prompts such as: realistic, Photography and hyper detailed may damage the pixel art aesthetic, use those only in the last parts of a prompt (if you want to use them at all)

  • Pay attention to the LoRAs you are using, some may completely remove the pixel art style from the image

  • The model understands keyword like "detailed", "best quality" or "masterpiece" and they actually may increase the output's quality.