AMD has released a new stable Defusion 3 Medium Artificial Intelligence (AI) model adapted to XDNA 2 Neural Processing Units (NPU). The chipmaker claimed that it is the world’s first AI model that processes the output in the BF16 format. The model will be supported by the new Ryzen AI laptop with at least 24GB of RAM, when users download the amuse 3.1 beta software of Tensorsstack. Stable spread 3 medium is an on-device image generation model that does not require internet connectivity.
AMD’s image generation model can produce print-taiyar images
In a press release, Santa Clara-based tech veteran expanded the new image generation model. The AI model is based on the stable spread 3 medium, adapted to the company’s XDNA NPU and equipped in the Ryzen AI laptop issued in 2024 and new.
The company claims that the model can be used to generate stock-quality images from text signals. The model produces 1024 × 1024 resolution images, which are then ready for 2048 × 2048 print-redi resolution using NPU’s capabilities.
The new AI model is part of the AMD and Tensorstack’s new AMuse 3.1 desktop app, which is free to download and install. Since the image generation model runs perfectly locally, it works even when the device is not connected to the Internet. The data-processing on-device is powered by XDNA 2 NPU.
AMD stated that it has worked on the memory requirements of the AI model, and now it requires 24 GB RAM instead of 32 GB RAM that was required for the stable spread XL Turbo model. Additionally, the new image model consumes only 9GB RAM when the model is active. The company achieved this by using the Block Floating Point 16 or Block FP16 (BF16) memory-skilled format.
Tech veteran stated that stable spread 3 medium AI model strictly follows the early, structure and order. AMD stated that users trying the model should describe the type of image, then structural components, and finally details and other references. Negative signals can be used to remove elements from the image, and the placement of full break can change the reference understanding of the model.
