Stability AI and Nvidia Partner to Enhance Performance of Text-to-Image AI
Stability AI and Nvidia have joined forces to improve the performance of Stability AI’s text-to-image generative AI product, Stable Diffusion XL (SDXL). This collaboration aligns with Stability AI’s goal of unlocking human potential through advanced AI technologies.
Stable Diffusion XL is a text-to-image model based on diffusion techniques, capable of generating detailed images from text prompts and modifying existing images with text guidance. It is a lightweight model that can run on consumer GPUs with at least 8 GB VRAM. Recently, a new version called Stable Diffusion XL was launched.
To enhance the speed and efficiency of SDXL, Stability AI has integrated Nvidia’s TensorRT, a high-performance optimization framework. TensorRT is an SDK for deep learning inference that optimizes neural network models and delivers low latency and high throughput on Nvidia GPUs. It is compatible with major frameworks like PyTorch, TensorFlow, and ONNX, as well as application-specific SDKs offered by Nvidia.
The integration of TensorRT with the converted ONNX model has resulted in impressive performance improvements. There has been a doubling of performance on Nvidia H100 chips, which feature a dedicated Transformer Engine that accelerates large language models. This advancement enables the generation of high-definition images in just 1.47 seconds.
Stability AI has hinted at further optimizations, such as the introduction of 8-bit precision, which is expected to increase both speed and accessibility of the AI model. With this collaboration, Stability AI and Nvidia are taking steps towards advancing the capabilities of text-to-image generative AI and unlocking new possibilities in AI-driven applications.
– Stability AI