PATINA
fal.ai
- Web
- API
What PATINA is
PATINA is a paid web and API model for generating material textures from text or images. Text prompts produce complete, seamless PBR materials, with upscaling up to 8K. The available maps are basecolor, normal, roughness, metalness, and height. An image URL can generate texture variations, while an image and mask can regenerate selected regions. Tiling supports both directions, horizontal, or vertical settings, with configurable tile size and stride. Requests can return one to four texture images and use one to eight denoising steps. Prompt expansion is enabled by default and adds approximately $0.0025. The model listing marks commercial use. Exports include PNG, JPEG, and WebP. API examples are provided for JavaScript, Python, and cURL, using a FAL_KEY environment variable. Pricing is billed per request: $0.01 base, plus listed per-megapixel and map charges; upscaling adds further charges.
Who it suits
It suits artists and developers needing text-to-texture, image variation, inpainting, seamless tiling, and PBR maps. API users can work from JavaScript, Python, or cURL.
Good to know, and what to watch
- Generates five listed PBR maps
- Supports seamless horizontal, vertical, or two-way tiling
- Text and image inputs are supported
- Resolution reaches 8192 pixels
- PNG, JPEG, and WebP exports
- Paid per request with additional map charges
- Prompt expansion adds approximately $0.0025
- Requests allow one to four texture images
- Only one to eight denoising steps are supported
Plans
Every plan and pricePATINA material inference
The verdict
PATINA brings text generation, image variation, masking, tiling, and PBR map output into one model endpoint. Its per-request pricing and configurable generation limits should be considered when planning usage.
What the maker publishes
- Commercial use
- The model listing is marked for commercial use. source
- Dataset
- fal says PATINA's training dataset was built from CC0 material libraries including AmbientCG, Poly Haven, and other public-domain sources. source
- Developer access
- The API page provides JavaScript, Python, and cURL examples and uses a FAL_KEY environment variable for authentication. source
- Generation limits
- The endpoint supports 1–4 texture images per request and 1–8 denoising steps. source
- Inpainting
- An input image and mask can be used to regenerate white mask regions while preserving black regions. source
- Maker background
- fal says it was founded in 2021 and is headquartered in San Francisco. source
- Output maps
- The five available PBR maps are basecolor, normal, roughness, metalness, and height. source
- Prompt controls
- Prompt expansion is enabled by default and adds approximately $0.0025. source
- Related photo workflow
- The material extraction endpoint isolates a material from a photo, flattens it, makes it seamless, and predicts PBR maps. source
- Safety
- The endpoint has an enabled-by-default safety checker option for filtering outputs. source
- Support
- fal lists [email protected] as its contact address. source
- Text to material
- Generates complete, seamlessly tiling PBR materials from text prompts, with upscaling up to 8K. source
- Texture variations
- With an input image URL, the endpoint can create texture variations through image-to-image generation. source
- Tiling controls
- Tiling can be set to both directions, horizontal, or vertical; tile size and stride are configurable. source
Checked against the maker's pages on 7 Oct 2026.