A small Agent Skill for making Unity URP textures with GPT Image, built around the constraints of standalone Meta Quest.
I made this for architectural VR scenes: concrete, plaster, decals, light patterns, glowing surfaces, fog, smoke, and other environment details. The goal is simple—describe the texture you need, get back a usable PNG, catch obvious tiling problems, and know which Unity import settings to start with.
This is not a Unity package or editor extension. Nothing is installed into your Unity project. The skill gives your coding agent a repeatable texture-generation and QA workflow.
This was the prompt:
Generate a seamless 1024×1024 light-gray fair-faced concrete exterior wall texture for standalone Quest and URP. Keep it restrained and modern, without obvious grime.
| Generated texture | 3×3 tiling check |
|---|---|
![]() |
![]() |
The result is a 1024×1024 RGB PNG. For Unity, the suggested starting point is sRGB on, Wrap Mode set to Repeat, mipmaps on, and ASTC 6×6 for Android. The 3×3 preview did not show an obvious seam.
This example was generated and checked with Codex. It has not been profiled inside a shipping Quest build, so treat the import settings as a practical starting point rather than a promise.
- Seamless architectural surfaces such as concrete, plaster, stone, painted walls, and subtle floor materials
- Grayscale Light Cookies for spot and directional lights
- Emission masks for windows, signs, strips, and other glowing materials
- Fog, smoke, dust, particle sprites, and reusable noise textures
- Decals and small environment details
For each asset, the agent reports the image path, dimensions, intended shader role, and the Unity settings that matter. Tileable surfaces also get a 3×3 repeat check.
The focus is environments and lighting. It does not model characters, build meshes, unwrap UVs, or generate point-light cubemap Cookies.
python3 ~/.codex/skills/.system/skill-installer/scripts/install-skill-from-github.py \
--repo okokelly/unity-texturegen \
--path skills/unity-texturegenStart a new Codex task after installation, then try:
Use $unity-texturegen to make a seamless concrete wall texture for Quest URP.
Install with the cross-agent Skills CLI
This option requires Node.js and npm:
npx skills add https://github.com/okokelly/unity-texturegen \
--skill unity-texturegen \
--agent codex \
--global \
--yesManual installation
git clone https://github.com/okokelly/unity-texturegen.git
# Codex
cp -R unity-texturegen/skills/unity-texturegen ~/.codex/skills/unity-texturegen
# Claude Code
cp -R unity-texturegen/skills/unity-texturegen ~/.claude/skills/unity-texturegen
# Another Agent Skills client—confirm its discovery path first
cp -R unity-texturegen/skills/unity-texturegen ~/.agents/skills/unity-texturegenRestart or reload the agent session after copying the skill.
Generate a seamless, warm-gray plaster wall for a quiet brutalist interior. Keep the surface variation subtle and check it for seams.
Create a 512×512 grayscale Light Cookie with soft window-blind shadows for a Unity spotlight.
Make a black-background additive ground-fog sprite for a standalone Quest scene. Keep the edges soft and avoid a bright center.
You can name an output folder in the request. If you do not, the agent saves the file somewhere stable in the current workspace and tells you the full path.
For a tileable surface, the skill repeats the generated image in a 3×3 grid and looks for visible borders, sudden tonal changes, or repeated landmarks. If it finds a problem, it makes one repair attempt and checks again.
The loop is intentionally short. Repeated image edits tend to change the material's scale and character, so after one repair the skill reports the result honestly as verified, failed, or unverified.
The preview script uses only the Python 3 standard library and produces a self-contained HTML file:
python3 skills/unity-texturegen/scripts/make_tile_preview.py texture.png \
--output /tmp/texture-3x3.html- A coding agent with a built-in raster image generator, such as GPT Image
- Python 3 for the tiling preview
- A Unity URP project
I tested the full workflow with Codex and its built-in image generator. The folder follows the Agent Skills format, so other compatible clients can read it, but they must also be able to generate images and open the local HTML preview.
- Generative images are not mathematically seamless. The repeat preview is a visual check, not a proof.
- Separately generated normal, roughness, and ambient-occlusion maps may not align pixel-for-pixel. The skill therefore does not promise a complete PBR texture set by default.
- Quest recommendations are conservative defaults. Final texture size and compression should still be tested on the target headset.
- The skill does not import assets into Unity or change project settings for you.
skills/unity-texturegen/
├── SKILL.md
├── agents/
│ └── openai.yaml
├── references/
│ ├── prompt-recipes.md
│ └── quest-texture-specs.md
└── scripts/
└── make_tile_preview.py
SKILL.mdcontains the workflow and discovery metadata.prompt-recipes.mdcontains prompts for different texture types.quest-texture-specs.mdcontains the Unity and Quest defaults.make_tile_preview.pybuilds the repeat preview without third-party packages.
The images used above live in examples/ and are not installed with the skill.
Issues and pull requests are welcome, especially for new environment texture recipes, better seam checks, and Quest settings backed by real measurements. I would like to keep the core workflow small, readable, and dependency-light.

