AMD and Stability AI Bring Local AI Image Generation to NPU-Powered Laptops
AMD and Stability AI have teamed up to adapt Stable Diffusion 3.0 Medium for the Stability Amuse AI art creator. This collaboration enables improved image generation and text handling directly on Neural Processing Units (NPUs), making AI-driven creativity more accessible on laptops.
Advancements Introduced at Computex 2024
At Computex 2024, AMD showcased the first block FP16 stable diffusion model, SDXL Turbo, developed with Stability AI. This model achieved FP16 accuracy while running at INT8 performance levels. Building on that, the two companies have now released the block FP16 Stable Diffusion 3.0 Medium model, which delivers higher image quality optimized for AMD Ryzen AI XDNA 2 NPUs.
What This Means for Users
Users can access this new model immediately through Amuse 3.1 by Tensorstack, running in HQ mode on compatible Ryzen AI-powered laptops. The updated block FP16 SD 3 Medium model requires less memory than previous versions, functioning efficiently on laptops with 24 GB of RAM while using only around 9 GB. This means high-quality AI image generation is now possible locally on laptops with under 32 GB of memory, without sacrificing precision due to excessive quantization.
Enhanced Image Quality with Two-Stage Pipeline
The system features a two-stage pipeline powered by the AMD XDNA 2 NPU that upscales output images from 2 megapixels (1024 x 1024) to 4 megapixels (2048 x 2048). Previously, Amuse required GPU processing to handle Stable Diffusion Medium, which limited its accessibility. Now, users have the flexibility to choose between GPU or NPU processing depending on their setup.
How to Get Started
- Download and install the latest AMD Adrenalin Driver.
- Install the Amuse 3.1 Beta version.
- Open Amuse, switch to EZ Mode, move the slider to HQ, and toggle on the XDNA 2 Stable Diffusion Offload option.
These steps will enable you to experience local AI image generation powered by the NPU on supported Ryzen AI laptops.
For those interested in expanding their AI skills and exploring more about AI tools and practical applications, consider visiting Complete AI Training for a range of courses and resources.
Your membership also unlocks:
