fal.ai ships PATINA

Physically based rendering is the ruleset game engines and visual effects tools use to decide how a surface reflects light. fal.ai said PATINA turns a prompt, a flat texture, or a photo into those surface maps for three-dimensional scenes. (blog.fal.ai) fal.ai published the launch on its blog about two days ago and exposed three endpoints on its model hub: `/patina` for image-to-PBR maps, `/patina/material` for text-to-material generation, and `/patina/material/extract` for pulling a material out of a photo. The company said the stack supports seamless tiling materials up to 8K through upscaling. (blog.fal.ai; fal.ai; fal.ai; fal.ai) A PBR map set is the bundle of images that tells rendering software what a surface is made of. fal.ai’s examples list five common outputs — base color, normal, roughness, metallic, and height — and its `/patina` endpoint says a 1024 by 1024 image with all five maps costs $0.06. (fal.ai; github.com) The release targets a bottleneck between image models and computer-generated imagery pipelines. fal.ai said PATINA is meant to bridge “image generation and CGI,” where artists often need tileable textures and lighting-aware material maps rather than a single finished picture. (blog.fal.ai) The text-to-material endpoint starts with a written prompt such as “weathered copper patina” and generates a seamless material at up to 2048 pixels before optional 2x or 4x upscaling pushes the maps to 8K. The extraction endpoint starts from a photo, flattens and normalizes the chosen surface, then rebuilds it as a tileable material. (fal.ai; fal.ai) fal.ai priced the text-to-material endpoint from $0.08 for a full material set, according to its launch post. The photo-extraction endpoint lists $0.10 base pricing plus $0.02 per megapixel and $0.01 per megapixel for each map type, with extra charges for 2x or 4x upscaling. (blog.fal.ai; fal.ai) The company also disclosed some of the model design. fal.ai said PATINA builds on its klein image model and adds an adapter with a DINOv2 backbone so the system can use semantic segmentation — a pixel-by-pixel scene labeler — to predict materials. (blog.fal.ai) That puts PATINA in a market where artists already use tools such as Adobe Substance 3D Sampler and other material-capture software to turn photos into reusable surfaces. fal.ai’s pitch is that the same API can now generate, extract, and upscale those assets inside a model-serving platform. (blog.fal.ai; github.com) For developers, the immediate change is practical: PATINA is live as callable fal.ai endpoints now, with example code and pricing on the model pages. For fal.ai, it is a push beyond image generation into the asset-making tools that feed game, design, and visual effects workflows. (fal.ai; fal.ai; fal.ai)