This NVIDIA extension brings TensorRT acceleration to the Stable Diffusion Web UI (AUTOMATIC1111), delivering the best Stable Diffusion performance on NVIDIA RTX GPUs. It converts diffusion models into optimized TensorRT engines for your exact GPU, so image generation runs significantly faster than the standard PyTorch path. The extension supports Stable Diffusion 1.5, 2.1, SDXL, SDXL Turbo and LCM models, and has gathered around 2,000 GitHub stars; NVIDIA last updated it in 2024.
After installing the extension from the Web UI’s Extensions tab, you generate a TensorRT engine for your checkpoint and resolution range once; subsequent generations run through the compiled engine with optimizations such as fused kernels and reduced-precision execution. NVIDIA recommends a GPU with 12 GB or more of VRAM for SDXL-class engines because of model size and compute requirements. For RTX owners generating many images locally, the speed-up versus stock PyTorch is the difference between interactive iteration and batch waiting.
Free. The extension is open source under the MIT license; it requires an NVIDIA RTX GPU and the free AUTOMATIC1111 Web UI.
Stable Diffusion Web UI users with NVIDIA RTX cards who generate at volume and want the fastest possible local inference without changing their workflow.
Because TensorRT engines are compiled for a specific GPU architecture and resolution profile, the extension trades a one-time build step and extra disk space for consistently faster sampling, and engines can be reused across sessions once built. It is the official route NVIDIA documents for accelerating AUTOMATIC1111 on RTX hardware, and it is commonly combined with the Web UI’s standard controls for ControlNet, LoRAs and upscalers, which continue to work through the normal interface.