On April 16, 2025, AMD, in collaboration with Stability AI and TensorStack, launched Amuse 3.0, a powerful AI image and video generation tool optimized for AMD Ryzen AI processors and Radeon GPUs. This beta release introduces AMD-optimized Stable Diffusion models and video diffusion capabilities, delivering up to 4.3x faster inference for local AI workflows, making it a game-changer for creators, gamers, and AI enthusiasts.
Source : AMD
Unleashing Generative AI with AMD-Optimized Models
Amuse 3.0 leverages AMD’s partnership with Stability AI to deliver ONNX-optimized Stable Diffusion models, including Stable Diffusion 3.5 Large, 3.5 Large Turbo, SDXL 1.0, and SDXL Turbo, available on Hugging Face with an “_amdgpu” suffix. These models achieve up to 3.8x faster inference compared to base PyTorch models, with the Radeon RX 9070 XT enabling 4.3x faster image generation for Stable Diffusion 1.5 and 3.1x for SDXL 1.0. Ryzen AI 9 HX 370 users experience 5-50% speed improvements, while the Ryzen AI MAX+ 395 (Strix Halo) delivers up to 3.3x faster performance for SD 3.5 models. These optimizations allow larger models, like SD 3.5 Large Turbo, to run efficiently on compatible AMD hardware, requiring the Adrenalin 24.30.31.05 driver or the upcoming 24.5.1 mainline release.
Source : AMD
Video Diffusion and Creative Tools
Amuse 3.0 introduces video diffusion, enabling local video restyling and text-to-image clip generation. Users can transform videos with draft-quality video-to-video models, all processed offline on AMD hardware. The software’s user-friendly interface, featuring EZ Mode and Advanced Mode, simplifies AI image generation with text-to-image, paint-to-image, and custom AI filter tools. AMD XDNA Super Resolution enhances output, upscaling images to 2048×2048 pixels, ideal for high-quality creative projects.
Source : AMD
Hardware Requirements and Accessibility
Optimized for AMD Ryzen AI 300 Series (24GB+ RAM), Ryzen 8040 Series (32GB RAM), and Radeon RX 7000/9000 Series GPUs, Amuse 3.0 requires a one-click installer from TensorStack’s website and specific AMD drivers. While designed for AMD’s XDNA architecture, it supports partial model offloading in “minimum” mode for lower VRAM setups, though performance depends on disk and RAM speeds. As a beta, Amuse may have bugs, and users should verify hardware compatibility.
Conclusion
Amuse 3.0 empowers AMD Ryzen AI and Radeon users to explore generative AI locally with unmatched speed and creativity. From optimized Stable Diffusion models to video diffusion, this release showcases AMD’s commitment to accessible AI. Download Amuse 3.0 today to transform your creative workflow, but act fast ensure your system meets the requirements for the best experience.