README.md

---
title: "Material Maps - Normal, Height, Roughness from One Image"
emoji: 🧱
colorFrom: indigo
colorTo: purple
sdk: gradio
sdk_version: 6.1.0
python_version: "3.12"
app_file: app.py
pinned: false
models:
  - InvokeAI/pbr-material-maps
  - jingheya/lotus-normal-g-v1-1
  - openai/clip-vit-base-patch32
license: apache-2.0
short_description: "PBR normal, height, roughness, metallic maps from a texture."
---

# Material Maps - Normal, Height and Roughness from One Image

Turn one picture of a surface, a texture tile or a photo, into the maps a PBR renderer needs: a tangent-space normal map (OpenGL, or DirectX on request), a 16-bit height map, roughness and metallic. The maps come back at the picture's size, up to 1024 pixels, and a seamless picture gives seamless maps: every network pads by wrapping around the edges.

Height is not guessed from brightness. Nets trained on texture sets (the Material Map Generator ESRGAN models) read painted bricks as raised and white mortar as sunk, which a brightness-based height map gets backwards. Lotus-G, a surface-normal diffusion model, adds the broad shape of stones and bricks, and CLIP names the material to set how rough it is and whether it is metal.

API endpoint, one GPU call, about a second on the GPU:

- `/material_maps(image, directx=False)` returns `normal.png`, `height.png` (16-bit), `roughness.png`, `metallic.png` and a JSON note (`material`, `classes`, `roughness_level`, `metal`, `seconds`).

## Free on 3D Valley

Used by the texture generator and the normal map tool on [3D Valley](https://3dvalley.com). Built by [Upsampler](https://upsampler.com), which also offers AI image generation, editing, upscaling, and enhancement tools.