AMD’s New BF16 SD 3 Medium Model Lets You Run Stable Diffusion Locally on Ryzen AI Laptops—How to Use It

AMD and Stability AI have partnered once again—this time to launch a compact, power-efficient AI image model called "BF16 SD 3 Medium", designed specifically for AMD’s XDNA 2 NPUs. Unlike other AI tools that depend heavily on cloud servers or bulky desktop GPUs, this latest advancement runs directly on Ryzen AI laptops, using a new bfloat16 (BF16) precision format to generate high-quality images with minimal hardware demands.

If you own a Ryzen AI-powered laptop with the right specs, you can try it right now through Amuse 3.1 Beta.

What Is BF16 SD 3 Medium?

BF16 SD 3 Medium is a Stable Diffusion 3.0 image generation model optimized to run efficiently on AMD’s AI-capable chips. It’s the first NPU-native model of its kind to use BF16 precision, which preserves most of the accuracy of full-precision FP16 but dramatically lowers memory usage.

The result? A smooth, local text-to-image generation experience, with no cloud access, GPU, or subscription required.

What Hardware Do You Need?

To run SD 3 Medium in high quality mode on your laptop, you'll need:

• An AMD Ryzen AI 300 series or Ryzen AI MAX+ laptop
• A XDNA 2 NPU capable of 50 TOPs or higher
• At least 24GB of system RAM

The model uses just around 9GB of memory, meaning it can comfortably run on many new laptops without maxing out system resources.

How Does It Work?

The generation process is split into two stages:

1. First, the model creates a base image at 1024×1024 resolution (2MP).
2. Then, it uses the NPU to upscale it to 2048×2048 (4MP), giving you print-ready images on the go.

This two-stage pipeline allows you to create detailed visuals on battery power—even without Wi-Fi.

How to Try It on Your AMD AI Laptop

Here’s a simple setup walkthrough:

1. Install the latest AMD Adrenalin graphics driver
2. Download Amuse 3.1 Beta
3. Open Amuse, go to EZ Mode
4. Push the slider to HQ (high quality)
5. Turn on “Stable Diffusion Offload” to use your XDNA 2 NPU

From there, you can start generating images directly on your device—no cloud login or internet connection needed.

Why It Matters

Running Stable Diffusion locally has long been possible, but it often required powerful GPUs and large memory allocations. The shift to BF16 enables lower memory usage without sacrificing image detail. This makes high-quality AI generation more accessible to creators working on the move, especially in low-connectivity environments.

Designers can quickly prototype marketing visuals, social media assets, or concept art—even at a coffee shop or on a flight. Since everything runs locally, iterations are fast, private, and cost-free.

Via

Source