fal.ai releases PATINA

fal.ai has released PATINA, a model that turns text or images into physically based rendering material maps for three-dimensional graphics workflows. (blog.fal.ai) Physically based rendering is the standard way game engines and visual effects tools describe surfaces such as brick, wood, or metal, using separate maps for color, roughness, metalness, height, and surface direction. PATINA’s material endpoint says it generates seamless tiling maps up to 8K resolution, while its image endpoint says it can infer those maps from a reference image without changing the original picture. (fal.ai, fal.ai) fal.ai published the launch on its blog on April 10, 2026, and the company’s model pages list commercial use and application programming interface access through endpoints including `fal-ai/patina`, `fal-ai/patina/material`, and `fal-ai/patina/material/extract`. The extract endpoint says a 1024 by 1024 material with five maps and no upscaling costs $0.17, while a 2048 by 2048 material with four-times upscaling to 8K costs $0.70. (blog.fal.ai, fal.ai, fal.ai) The pitch is aimed at a long-running production problem: image generators can make a surface that looks convincing in one frame, but computer graphics pipelines need reusable material controls that still behave correctly when lighting or camera angles change. fal.ai wrote that PATINA is closer to spatially varying bidirectional reflectance distribution function estimation, which means recovering the hidden surface properties a renderer needs rather than just guessing depth or edges. (blog.fal.ai) fal.ai said it trained PATINA on public-domain material libraries including AmbientCG and Poly Haven, then rendered those materials under many lighting setups with a custom Cook-Torrance renderer. The company said that process produced paired examples linking ordinary-looking images to the underlying maps used by rendering software. (blog.fal.ai) The pricing model also reflects that production framing. fal.ai’s image-to-maps endpoint charges a $0.01 base fee plus $0.01 per megapixel for each output map, and the company’s pricing documentation says image models on the platform are commonly billed per image or per megapixel rather than by subscription tier. (fal.ai, fal.ai) PATINA enters a crowded market for image generation, but a narrower one for tools that package outputs as renderer-ready materials. fal.ai’s own launch post compares PATINA with geometry-focused systems such as MiDaS, Depth Anything v2, Marigold Depth, Marigold Normal, and CHORD, arguing that those tools recover shape cues while PATINA tries to separate shape from gloss, reflectivity, and micro-surface detail. (blog.fal.ai) The immediate test is whether artists and developers treat PATINA as more than a demo. fal.ai is selling it as an application programming interface product with explicit 1K-to-8K cost examples, which puts the model directly into the budgeting math of game, design, and visual effects pipelines. (fal.ai, fal.ai, fal.ai)