Skip to content
Get started
GENERATION API CALLS
Image Model Generation

Patina AI

This page is auto-generated from model configurations. Last updated: 2026-09-01.

This reference lists all available Patina AI image generation models and their parameters. Use these parameter names when calling the Generation API.


Predict seamless PBR maps (base color, normal, roughness, metalness, height) from an existing texture or photo. The source image is analyzed, not rewritten.

Model ID: model_patina

Capabilities: img2img

LLM Markdown: https://app.scenario.com/api/models/model_patina/markdown

ParameterTypeRequiredDefaultMinMaxAllowed ValuesDescription
imagefileYes----A flat texture or photo of a surface. PATINA predicts PBR maps from it without changing the original.
mapsstring_arrayNobasecolor,normal,roughness,metalness,height--basecolor, normal, roughness, metalness, heightWhich PBR maps to generate. Defaults to all five.
seednumberNo-02147483647-A number that makes results repeatable. Reusing the same seed and settings produces the same maps; leave it empty for a different result each time.

Generate a seamlessly tiling PBR material from a text prompt, with optional image-to-image variations and inpainting. Outputs a texture plus base color, normal, roughness, metalness, and height maps, up to 8K with upscaling.

Model ID: model_patina-material

Capabilities: txt2img, img2img

LLM Markdown: https://app.scenario.com/api/models/model_patina-material/markdown

ParameterTypeRequiredDefaultMinMaxAllowed ValuesDescription
promptstringYes----Text description of the material to generate, for example weathered copper patina or mossy stone wall.
imagefileNo----An optional starting texture. Without a mask, PATINA creates variations. With a mask, it inpaints the marked region.
imageStrengthnumberNo0.60.011-How much to transform the input image. Higher values change it more. Only used when an image is provided.
maskfileNo----A black-and-white mask for inpainting. White regions are regenerated, black regions are preserved. Requires an input image.
widthnumberNo10245122048-Output width in pixels. Upscaling applies to PBR maps only, not the base texture.
heightnumberNo10245122048-Output height in pixels. Upscaling applies to PBR maps only, not the base texture.
mapsstring_arrayNobasecolor,normal,roughness,metalness,height--basecolor, normal, roughness, metalness, heightWhich PBR maps to predict. Deselect all to skip PBR estimation and only generate the texture.
upscaleFactornumberNo0--0, 2, 4Seamless upscale for predicted PBR maps. 0 keeps map resolution, 2 doubles it, 4 quadruples it (up to 8K from a 2048 texture). The base texture is not upscaled.
tilingModestringNoboth--both, horizontal, verticalWhich directions the material repeats seamlessly. Both repeats in every direction; Horizontal repeats left-to-right; Vertical repeats top-to-bottom.
numOutputsnumberNo114-Number of texture images to generate from the same prompt.
numInferenceStepsnumberNo818-Number of denoising steps for texture generation. Higher values take longer.
enablePromptExpansionbooleanNotrue---Expand the prompt with an LLM for richer material detail.
tileSizenumberNo12832256-Seamless-tiling window in latent space. 64 β‰ˆ 512px, 128 β‰ˆ 1024px, 256 β‰ˆ 2048px. Match this to the output resolution for cleaner repeats.
tileStridenumberNo6416128-How far the tiling window steps between passes. Smaller than tile size adds overlap and usually smoother seams; larger is faster but can show a grid.
seednumberNo-02147483647-A number that makes results repeatable. Reusing the same seed and settings produces the same material; leave it empty for a different result each time.

Extract a clean, seamlessly tiling PBR material from a photo of a real surface. Isolates the chosen material, flattens it, and predicts base color, normal, roughness, metalness, and height maps.

Model ID: model_patina-material-extract

Capabilities: img2img

LLM Markdown: https://app.scenario.com/api/models/model_patina-material-extract/markdown

ParameterTypeRequiredDefaultMinMaxAllowed ValuesDescription
promptstringYes----Describe which material to extract from the photo, for example the brick wall or the wooden floor.
imagefileYes----A photo that contains the material you want. Perspective, shadows, and surrounding scene are flattened into a seamless texture.
imageStrengthnumberNo0.60.011-How much to transform the extracted material while making it tileable. Higher values change it more.
widthnumberNo10245122048-Output width in pixels. Upscaling applies to PBR maps only, not the base texture.
heightnumberNo10245122048-Output height in pixels. Upscaling applies to PBR maps only, not the base texture.
mapsstring_arrayNobasecolor,normal,roughness,metalness,height--basecolor, normal, roughness, metalness, heightWhich PBR maps to predict. Deselect all to skip PBR estimation and only extract the texture.
upscaleFactornumberNo0--0, 2, 4Seamless upscale for predicted PBR maps. 0 keeps map resolution, 2 doubles it, 4 quadruples it (up to 8K from a 2048 texture). The base texture is not upscaled.
tilingModestringNoboth--both, horizontal, verticalWhich directions the extracted material repeats seamlessly. Both repeats in every direction; Horizontal repeats left-to-right; Vertical repeats top-to-bottom.
numOutputsnumberNo114-Number of texture variations to extract from the same photo.
numInferenceStepsnumberNo818-Number of denoising steps for texture generation. Higher values take longer.
enablePromptExpansionbooleanNotrue---Expand the prompt for richer material detail.
tileSizenumberNo12832256-Seamless-tiling window in latent space. 64 β‰ˆ 512px, 128 β‰ˆ 1024px, 256 β‰ˆ 2048px. Match this to the output resolution for cleaner repeats.
tileStridenumberNo6416128-How far the tiling window steps between passes. Smaller than tile size adds overlap and usually smoother seams; larger is faster but can show a grid.
seednumberNo-02147483647-A number that makes results repeatable. Reusing the same seed and settings produces the same extract; leave it empty for a different result each time.