The problem with online image generation using artificial intelligence is always the same: its free tiers don't last long, or are very limited. The best way to escape those restrictions is to run local models based on Stable Diffusion, and although there are several interfaces on the web (Auto1111 and ComfyUI come to mind), now AMD joins in with version 2.0 of its Amuse system. What does it have to offer?
The hype of artificial intelligence is focusing a lot on chatbots and other assistance tools, but image generation hasn't gone anywhere. On the contrary, models keep gaining precision, detail, and complexity, which in turn places more pressure on hardware.
However, none of this can reach the user's hands without good software, and that's where the Amuse platform from AMD and TensorStack comes in. Activating Stable Diffusion models on AMD hardware has always required extra steps or adopting optimized forks, but Amuse is the closest thing to a "fire and forget" solution. The best part? No tokens, no accounts, no licenses, no subscriptions. If you have firepower in your machine, you can run this.
Generating Images with Amuse 2.0
Another positive aspect of Amuse is that it doesn't force us to use AMD hardware. The page clearly indicates that the program was optimized for Ryzen AI 300, Ryzen 8040, and Radeon RX 7000 graphics cards, but I have managed to generate images on a GeForce GTX 1660 Super without issues. Now, offline mode doesn't mean we can be completely free from the internet. Models have to come from somewhere, and when you choose one of the performance presets, Amuse will try to download everything it needs. The smallest model (SD Turbo, available with the Fast preset) takes about four gigabytes, so be patient.
First recommendation? Activate advanced mode. The initial interface is intuitive and not too aggressive, with support for negative prompts (must be enabled in Settings), but the control of seeds, custom resolutions, the number of steps and the schedulers are all hidden behind advanced mode. This enables access to the model manager... which leads me to the second recommendation: Download alternative models. SD Turbo is terrible, and even in low-VRAM environments, using something like DreamShaper v8 will produce far superior images.
Amuse includes Image to Image, Paint to Image, Inpainting, Upscaling, feature extractor, and even video generation, but I think simplifying image generation on AMD hardware is its strongest point. From a technical standpoint, Amuse 2.0 is in beta, however, I haven't found any major errors so far. As always, the only elements that "suffer" are the graphics card and the internet connection: The suggested amount of VRAM is "all", and some models are enormous.
Official website and download: Click here
