StreamDiffusion requires CUDA GPU, Python 3.10, PyTorch 2.1, and optional TensorRT for optimal performance

StreamDiffusion has a layered installation: (1) conda/venv with Python 3.10; (2) PyTorch 2.1 with matching CUDA version (11.8 or 12.1); (3) pip install streamdiffusion[tensorrt] for the full install with TensorRT support, then python -m streamdiffusion.tools.install-tensorrt to install TensorRT bindings; (4) optional Docker path that pre-bundles CUDA + TensorRT. The [tensorrt] extra is required for the highest performance mode. Windows users additionally need pip install --force-reinstall pywin32. Without a CUDA-capable NVIDIA GPU, realtime performance is not achievable—CPU inference is orders of magnitude slower. A Docker image (stream-diffusion:latest) provides a TensorRT-ready environment without manual dependency management.

Examples

conda create -n streamdiffusion python=3.10
conda activate streamdiffusion
pip3 install torch==2.1.0 torchvision==0.16.0 xformers --index-url https://download.pytorch.org/whl/cu121
pip install git+https://github.com/cumulo-autumn/StreamDiffusion.git@main#egg=streamdiffusion[tensorrt]
python -m streamdiffusion.tools.install-tensorrt

Assessment

A user has an AMD GPU. Which StreamDiffusion features are unavailable to them and why? Identify which installation step they would skip.