dcode

Text -> Polargraph Gcode via Stable Diffusion

Single end-to-end diffusion model that converts text prompts directly to polargraph-compatible gcode.

Architecture

text prompt -> CLIP text encoder -> UNet diffusion -> latent [4,64,64] -> GcodeDecoder -> gcode tokens

All components post-trained end-to-end on 175,952 image-gcode pairs.

Machine Specs

• Work area: 841mm x 1189mm (A0)
• Bounds: X [-420.5, 420.5], Y [-594.5, 594.5]
• Pen servo: 40 deg (down), 90 deg (up)

Links

• GitHub
• Model
• Dataset

License

MIT