Stability AI’s Stable Diffusion 3 is the latest iteration of its popular AI image-generation tool. This powerful tool delivers photorealistic results, boasting what the company calls the ‘most advanced text-to-image open model yet.’ One of the key advancements in Stable Diffusion 3 is its ability to address challenges like recreating repeating patterns and accurately rendering human hands – issues that have plagued previous AI image generators. This impressive model can run on a range of hardware, from an Apple M2 Ultra to a powerful GeForce RTX 4090-powered rig.
The GeForce RTX 4090, with its 24GB of GDDR6X memory and unlocked Ada Lovelace architecture, delivers unparalleled speed for AI image generation. However, even with TensorRT acceleration and optimization, generating images in real-time requires a powerful GPU. NVIDIA showcased this capability at SIGGRAPH 2024, demonstrating SDXL Turbo generating an image of ‘a hot rod, racing in the desert at sunset’ in real-time. The model even added details like a canyon to the background as the user typed in the prompt.
This real-time performance is possible thanks to TensorRT optimizations, which deliver a 60% speedup in Stable Diffusion 3.0. NVIDIA has made an optimized Stable Diffusion 3 NIM microservice available for preview on ai.nvidia.com, allowing users to experience this real-time generation firsthand. The performance boost extends to video generation as well, with a 40% speedup. While generating video still requires a few seconds, the ability to create complex AI-generated images in real-time on a local machine is a significant milestone.
Although the GeForce RTX 4090 is not a mainstream GPU, it’s readily available and represents a significant step forward for AI image generation. The real-time capabilities of Stable Diffusion 3 on a powerful GPU like the GeForce RTX 4090 usher in a new era of accessibility and speed for AI-powered creative workflows.
