AMD has launched a brand new Stable Diffusion 3 Medium synthetic intelligence (AI) mannequin optimised for XDNA 2 neural processing items (NPUs). The chipmaker claimed that it’s the world’s first AI mannequin that processes outputs within the BF16 format. The mannequin can be supported by the newer Ryzen AI laptops with at the very least 24GB RAM, after customers obtain Tensorstack’s Amuse 3.1 beta software program. The Stable Diffusion 3 Medium is an on-device picture era mannequin that doesn’t require Internet connectivity.
AMD’s Image Generation Model Can Generate Print-Ready Images
In a press launch, the Santa Clara-based tech big detailed the brand new picture era mannequin. The AI mannequin is predicated on Stable Diffusion 3 Medium, which is optimised for the corporate’s XDNA NPUs and are outfitted within the Ryzen AI laptops launched in 2024 and newer.
The firm claims the mannequin can be utilized to generate stock-quality photos from textual content prompts. The mannequin generates 1024×1024 decision photos, that are then upscaled to 2048×2048 print-ready decision utilizing the NPU’s capabilities.
The new AI mannequin is a part of AMD and Tensorstack’s new Amuse 3.1 desktop app, which is free to obtain and set up. Since the picture era mannequin runs totally regionally, it even works when the gadget shouldn’t be related to the Internet. The data-processing happens on-device, powered by the XDNA 2 NPUs.
AMD mentioned it has labored on the reminiscence necessities of the AI mannequin, and it now requires 24GB RAM, as a substitute of 32GB RAM which was essential for the Stable Diffusion XL Turbo mannequin. Additionally, the brand new picture mannequin consumes solely 9GB of RAM whereas energetic. The firm achieved this through the use of the block floating level 16 or block fp16 (BF16) memory-efficient format.
The tech big highlighted that the Stable Diffusion 3 Medium AI mannequin strictly adheres to the immediate, construction, and order. AMD mentioned customers making an attempt out the mannequin ought to first describe the kind of picture, then the structural elements, and eventually particulars and different context. Negative prompts can be utilized to take away components from the picture, and placement of full stops can change the context understanding of the mannequin.
