Stability AI and NVIDIA Boost Performance and Simplify Enterprise Deployment with Stable Diffusion 3.5 NIM

Stability AI has partnered with NVIDIA to launch the Stable Diffusion 3.5 NIM microservice. This allows enterprises to deploy the latest image generation model faster and more easily.

SD 3.5 NIM maximizes image generation performance on enterprise hardware. In tests on NVIDIA H100 GPU, the TensorRT-optimized generation method (3,700ms) recorded roughly 1.8x faster performance compared to the existing PyTorch method (6,800ms).

Deployment convenience has also improved significantly. SD 3.5 NIM integrates the SD 3.5 Large model along with Depth and Canny ControlNet into a single container. Users can efficiently manage complex workflows through this optimized package without needing to deploy each model separately.

This service supports NVIDIA's Ada and Blackwell GPUs, bringing large-scale batch processing and multi-model workflows to a practical level.

The model can be used for commercial and non-commercial purposes under the Stability AI Community License. However, companies with annual revenue of $1 million or more must use Enterprise Licensing, which includes customized support.