AMD has unveiled the first image-generating AI based on Stable Diffusion 3.0 Medium, specifically designed for XDNA 2 NPU (Neural Processing Unit) cores, which runs locally on Ryzen AI laptops.
According to reports, this model is capable of generating high-quality images offline, without requiring internet access or cloud services. It has been developed in collaboration with Stability AI.
AMD Introduces Offline Stable Diffusion 3.0 Medium AI Model for XDNA 2
This text-to-image AI model is based on Stable Diffusion 3.0 Medium and has been optimized specifically for the BF16 format, tailored for execution on devices equipped with XDNA 2 NPUs. The model initially generates images at a resolution of 1024×1024, which are then upscaled to 2048×2048 using an internal pipeline within the NPU, resulting in a 4-megapixel final output. AMD emphasizes that the image quality is suitable for professional use and printing.
To run this model, users will need a computer powered by an AMD Ryzen AI 300 or Ryzen AI MAX+ processor, an XDNA 2 NPU capable of delivering at least 50 TOPS (Tera Operations Per Second), and a minimum of 24 GB of RAM. The model itself consumes around 9 GB of RAM during image generation, indicating a heavy processing load. The key advantage of this model is its ability to run locally on the device, ensuring fast generation speeds and operation without reliance on an internet connection or cloud-based services.
This AI model is primarily aimed at content creators and designers who require customizable images. It supports advanced prompting features for precise control over image composition.
To use this model, users must install the latest AMD Adrenalin Edition drivers and the Amuse 3.1 Beta software developed by Tensorstack. Once installed, users should open the Amuse application, switch to EZ Mode, set the slider to HQ, and enable the “XDNA 2 Stable Diffusion Offload” option.
Usage of this model falls under the Stability AI Community License, making it free for individuals and small businesses with annual revenue under $1 million. However, licensing terms may change in the future. It’s also worth noting that the Amuse software is still in beta, so users may encounter stability and performance issues.
