PBR Texture Generator v2

Generate seamless PBR (Physically Based Rendering) texture maps from text descriptions with AI-powered depth estimation.

AI Models Used

Features

• AI Height Maps: Uses Intel DPT-Large for accurate depth estimation from textures
• Derived PBR Maps: Normal, AO, roughness, and emissive computed from AI-estimated height
• Hybrid Normals: Blend height-based (smooth) with color-based (detailed) normals
• Emissive Detection: Auto-detect glowing/neon areas for emissive maps
• Seamless Tiling: Edge blending with cosine interpolation
• OpenGL/DirectX: Choose normal map format for your engine
• 16-bit Height: Optional high-precision height map export

Output

Generates 7 texture maps from a single prompt:

Usage

import replicate

output = replicate.run(
    "vantilator2000/pbr-playground",
    input={
        "prompt": "seamless cyberpunk circuit board, neon blue and purple, glowing lines",
        "negative_prompt": "blurry, text, watermark",
        "resolution": 1024,
        "tiling_strength": 0.5,
        "num_steps": 25,
        "normal_strength": 1.0,
        "normal_detail": 0.25,
        "normal_format": "opengl",
        "height_contrast": 1.0,
        "ao_strength": 1.0,
        "seed": 42
    }
)

# output[0] = color
# output[1] = height
# output[2] = normal
# output[3] = roughness
# output[4] = ao
# output[5] = emissive
# output[6] = grid

Parameters

Generation

Tiling

Height Map

Normal Map

Ambient Occlusion

Roughness

Two Modes

1. Generate from Prompt

Leave input_image empty and provide a text prompt. The AI generates the texture and all PBR maps.

2. Generate from Image

Upload your own texture image - the model estimates height using AI and generates all PBR maps from it.

Normal Map Formats

Example Prompts

• seamless dark wood texture, oak, detailed grain
• seamless marble texture, white with grey veins
• seamless grass texture, top-down view, lawn
• seamless concrete texture, weathered, cracks
• seamless metal texture, brushed steel
• seamless cyberpunk circuit board, neon blue and purple, glowing lines
• seamless stone wall texture, medieval castle
• seamless leather texture, brown, worn

Tips

1. Include “seamless” in your prompt for better tiling
2. Use “top-down view” for floor/ground textures
3. Increase num_steps (30-40) for higher quality
4. Increase normal_strength (1.5-2.0) for more pronounced bumps
5. Adjust normal_detail (0.3-0.5) for more surface texture in normals
6. Set ao_radius higher for softer, more spread shadows
7. Use negative_prompt to avoid unwanted elements like “blurry, text, watermark”

Technical Details

Depth Estimation

Uses Intel DPT-Large (Dense Prediction Transformer) for monocular depth estimation. The depth is converted to a height map with optional scene-depth suppression to preserve texture detail.

Normal Generation

Normals are computed using Sobel operators on the height map. Optionally blends in high-frequency detail from the color image for more organic results.

Emissive Detection

Analyzes HSV color space to detect bright, saturated areas (neon colors glow more than white).

Seamless Tiling

Uses edge blending with cosine interpolation - blends opposite borders together for seamless wrapping without center artifacts.

Use Cases

• Game development (Unity, Unreal Engine, Godot)
• 3D rendering (Blender, Maya, 3ds Max)
• Architectural visualization
• Product design
• Digital art
• VFX and film production

License

This project uses models with Apache 2.0 licenses, suitable for commercial use.