Few products have managed to cause a media frenzy as intense as the one caused by Stable Diffusion. Depending on the background of the journalist, it will either lead to political Armageddon or to the downfall of the liberal arts scene.
Yours truly has worked as a photographer and does not consider the images produced by Stable Diffusion ready for prime time. Good examples for problems are hands and feet, or images containing text. Nevertheless, it has its uses. Landscape ‘shots’, for example, tend to turn out well. Due to the extreme computational demands, running Stable Diffusion locally is not for the faint of heart. It is for good reason that Microsoft has embraced it and uses the vast compute powers of its Azure cloud to make image generation more palatable. Of course, the service – only in preview currently, due to ethical concerns – will come at a significant, and as yet not fully disclosed, monetary cost.
Be that as it may, the following steps show a way to get Stable Diffusion running on a moderately equipped mobile workstation. It can then generate around five images in two minutes, which is a decent speed for home use. When set up, you can replicate the media frenzy from your sofa. Alternatively, it can generate wallpapers, avatars or other scenery that is usually bought from stock image agencies.
First things first: like many other AI algorithms, uses the Nvidia CUDA interface. Hardware acceleration can only be used on a workstation equipped with Nvidia graphics – if you have AMD or Intel, don’t