# TRELLIS - Core.Today AI API

> Microsoft's TRELLIS image-to-3D (836K+ runs — the most-used 3D model on Replicate). Turns one or more images into a textured GLB 3D asset, with turntable render videos and optional Gaussian PLY.

- **Provider**: Microsoft
- **Model ID**: firtoz/trellis
- **Category**: Image Generation
- **Credits**: 77 per generation
- **Speed**: Medium
- **Quality**: High

## Features
- The most-used 3D model on Replicate
- Single or multi-image input
- Textured GLB export (generate_model)
- Color/normal turntable render videos
- Gaussian point cloud PLY export

## Use Cases
- Product photos to 3D assets
- Game asset prototyping
- AR/VR content creation
- 3D printing previews
- E-commerce 3D viewers

## API Endpoint

Base URL: https://api.core.today/v1
Create Prediction: POST /predictions
Get Status: GET /predictions/{job_id}

## Authentication
Header: X-API-Key: YOUR_API_KEY

## Input Parameters

### Required
- **images**: array - One or more images of the object (multiple angles improve quality)

### Optional
- **seed**: integer (default: 0) - Random seed (randomize_seed overrides)
- **randomize_seed**: boolean (default: true) - Randomize seed
- **generate_color**: boolean (default: true) - Generate a color turntable render video
- **generate_normal**: boolean (default: false) - Generate normal video render
- **generate_model**: boolean (default: false) - Generate the textured GLB 3D model file
- **save_gaussian_ply**: boolean (default: false) - Save Gaussian point cloud as PLY file
- **return_no_background**: boolean (default: false) - Return the preprocessed images without background
- **ss_guidance_strength**: number (default: 7.5) - Stage 1: Sparse Structure Generation - Guidance Strength
- **ss_sampling_steps**: integer (default: 12) - Stage 1: Sparse Structure Generation - Sampling Steps
- **slat_guidance_strength**: number (default: 3) - Stage 2: Structured Latent Generation - Guidance Strength
- **slat_sampling_steps**: integer (default: 12) - Stage 2: Structured Latent Generation - Sampling Steps
- **mesh_simplify**: number (default: 0.95) - Mesh simplification ratio for the GLB
- **texture_size**: integer (default: 1024) - GLB texture resolution (when generate_model is true)

## Examples

### Product Photo to GLB
Create a textured 3D model from a single product photo
```json
{
  "model": "firtoz/trellis",
  "input": {
    "images": [
      "https://example.com/product.png"
    ],
    "generate_model": true,
    "texture_size": 1024
  }
}
```

## Response Format
```json
{
  "job_id": "abc123",
  "status": "pending | processing | completed | failed",
  "result": "URL or data (when completed)"
}
```

## Usage Flow
1. POST /predictions with model and input -> receive job_id
2. GET /predictions/{job_id} -> poll until status is completed or failed
3. Result contains output URL(s)

## Tips
- Set generate_model: true to get the GLB file — the default only renders preview videos
- Clean background photos work best; remove backgrounds first for tricky shots
- Provide 2-4 angles of the same object for noticeably better geometry
- Result is a JSON with model_file (GLB), render videos, and optional PLY

## Documentation
https://replicate.com/firtoz/trellis