From the source
fal introduced PATINA, a model that generates high-resolution PBR maps from rendered images, bridging the gap between AI-generated images and traditional CGI workflows.
- PATINA uses a modified FLUX.2 klein backbone with a DINOv2 adapter and was trained on CC0 material libraries with a custom Cook-Torrance BRDF renderer.
- The model outputs basecolor, normal, roughness, metalness, and height maps, and is available via endpoints starting at $0.08 for a full material set.
