Diffusion Bridge

The diffusion bridge connects AI image generators (DALL-E, Stable Diffusion, Midjourney) to the PAX pixel art pipeline. Generate a reference image, and PIXL converts it into a clean, palette-constrained tile.

Warning
DALL-E generation requires an OPENAI_API_KEY environment variable. Stable Diffusion and Midjourney images can be imported without an API key using pixl convert.

Pipeline

Text promptdescribe your sprite
DALL-Egenerates 1024×1024
Grid detectfind pixel blocks
Samplecenter of each block
Paletteextract colors
Quantizemap to palette
CleanupAA + outlines
PAX tilegame-ready

Pixel grid detection

AI generators render "pixel art" at high resolution — each art pixel is a 20-40px block. PIXL scans horizontal and vertical runs of identical pixels to find the actual grid size, then samples the center of each block instead of blurring with traditional downscaling.

Background removal

Even when asked for transparency, generators often add glowing halos around sprites. PIXL flood-fills from the image corners with color tolerance to detect and strip these opaque-but-unwanted background pixels.

Auto-palette

Colors are extracted directly from the generated image using median-cut quantization. The actual darkest pixel (for outlines) and lightest pixel (for highlights) are always preserved — median-cut alone would average them away.

Default: 32 colors. Adjustable with max_colors.

Supported image sources

Any PNG or JPEG works as input:

  • DALL-E / GPT Image — built-in via OPENAI_API_KEY
  • Stable Diffusion — generate externally, import with pixl convert
  • Midjourney — download the image, import with pixl convert
  • Any pixel art image — screenshots, references, existing assets

CLI usage

# Generate via DALL-E
pixl generate-sprite tileset.pax \
  --prompt "wizard with staff" \
  --name wizard \
  --out wizard.png

# Import existing image
pixl convert reference.png --width 32 --colors 32