TRELLIS Text-to-3D API

REST API for Microsoft TRELLIS text-to-3D generation.

Quick Start

docker compose up --build

API available at http://localhost:8000 with interactive docs at /docs.

Test with: python test_api.py

• application starting up may take some time, as well as downloading model weight in the first run - the weights will be cached for future use under ./cache by volume-mounting the container.

Alternatively, run the docker container manually with docker run -it --gpus all -v $(pwd)/cache:/root/.cache -p 8000:8000 --name trellis trellis and run python text_to_3d_api.py --host 0.0.0.0 --port 8000.

API Usage

# Generate 3D model
curl -X POST "http://localhost:8000/generate" \
  -H "Content-Type: application/json" \
  -d '{"prompt": "A red sports car", "formats": ["mesh", "gaussian"]}'

# Download files from response URLs
curl -O "http://localhost:8000/files/{job_id}/model.glb"

Features

The Dockerfile also provides a way to use an updated torch+cuda environment to original TRELLIS repo.

• REST API for text-to-3D generation
• Multiple formats: GLB, PLY, MP4 preview videos
• Docker deployment with GPU support
• File management with automatic cleanup
• Health monitoring and error handling

See API_README.md for detailed documentation.