Microsoft

Introduction

Microsoft's AI models are integrated into the Runware platform through our unified API, providing access to high-fidelity 3D asset generation technology that transforms 2D images into fully textured 3D models with complex topology and physically based rendering materials.

Through the settings object, you can access TRELLIS.2's unique features such as texture resolution control, mesh decimation targeting, voxel grid resolution, and per-stage sampler configuration, while maintaining the consistency of Runware's standard API structure. This page documents the technical specifications, parameter requirements, and model-specific settings for all Microsoft models available through our platform.

3D models

TRELLIS.2

TRELLIS.2 is a 4-billion-parameter 3D generative model that transforms 2D images into fully textured 3D assets with complex topology, sharp features, and physically based rendering materials. It uses a novel field-free sparse voxel representation and scalable structured latent space to enable efficient, high-resolution 3D generation from single image input.

Model AIR ID: microsoft:trellis-2@4b.

Supported workflows: Image-to-3D.

Technical specifications:

Input image: Required via inputs.image.
Texture size: 1024–4096 pixels (multiples of 1024, default: 2048). Controls the resolution of the output texture map.
Decimation target: 100,000–1,000,000 polygons (default: 500,000). Controls mesh simplification for lighter models.
Resolution: 512, 1024, or 1536 (default: 1024). Sets the voxel grid resolution for 3D generation.
Remesh: Boolean (default: true). Whether to remesh the output for cleaner topology.
Sampler configuration: Three independent samplers for different generation stages:
- Sparse structure: Controls the initial sparse voxel grid that establishes basic 3D geometry.
- Shape SLAT: Controls the Structured Latent stage that refines 3D geometry.
- Texture SLAT: Controls the Structured Latent stage that generates surface appearance.
Each sampler supports: guidanceStrength (1–10), guidanceRescale (0–1), steps (1–50), rescaleT (1–6).

Basic image-to-3D

{
  "taskType": "3dInference",
  "taskUUID": "f47ac10b-58cc-4372-a567-0e02b2c3d479",
  "model": "microsoft:trellis-2@4b",
  "inputs": {
    "image": "c64351d5-4c59-42f7-95e1-eace013eddab"
  }
}

High-resolution output

{
  "taskType": "3dInference",
  "taskUUID": "6ba7b810-9dad-11d1-80b4-00c04fd430c8",
  "model": "microsoft:trellis-2@4b",
  "inputs": {
    "image": "c64351d5-4c59-42f7-95e1-eace013eddab"
  },
  "settings": {
    "textureSize": 4096,
    "resolution": 1536,
    "decimationTarget": 1000000,
    "remesh": true
  }
}

Custom sampler configuration

{
  "taskType": "3dInference",
  "taskUUID": "550e8400-e29b-41d4-a716-446655440000",
  "model": "microsoft:trellis-2@4b",
  "inputs": {
    "image": "c64351d5-4c59-42f7-95e1-eace013eddab"
  },
  "seed": 42,
  "settings": {
    "textureSize": 2048,
    "resolution": 1024,
    "sparseStructure": {
      "guidanceStrength": 7.5,
      "steps": 25
    },
    "shapeSlat": {
      "guidanceStrength": 5,
      "guidanceRescale": 0.5,
      "steps": 30
    },
    "texSlat": {
      "guidanceStrength": 6,
      "steps": 20,
      "rescaleT": 2
    }
  }
}

Lightweight mesh

{
  "taskType": "3dInference",
  "taskUUID": "a770f077-f413-47de-9dac-be0b26a35da7",
  "model": "microsoft:trellis-2@4b",
  "inputs": {
    "image": "c64351d5-4c59-42f7-95e1-eace013eddab"
  },
  "settings": {
    "textureSize": 1024,
    "resolution": 512,
    "decimationTarget": 100000,
    "remesh": true
  }
}