© STABLE DIFFUSION
WHILE MOST OF THE TEXT-AND IMAGE-GENERATING AI MODELS we’ve discussed here run in massive throbbing data centers on racks of GPUs—indeed, Elon Musk purchased 10,000 Nvidia cards in March for an unannounced AI project within Twitter—there is one that can be run on your home PC.
Stable Diffusion is a deep learning text-to-image service that was released in 2022. It uses a diffusion model rather than a large language model, but the concept is broadly similar in that it’s probability based and is trained from images harvested from the wider internet, plus words associated with them, as well as versions of the same images that have been blurred. With this knowledge, Stable Diffusion is able to begin with a frame full of random image noise, and works to de-blur it to ‘reveal’ the image. The final result is based on the model’s training and the prompt it’s provided with.
If you go to stablediffusionweb.com, you’ll be able to use Stable Diffusion to create images, but it’s also possible to run it on your own PC as long as your graphics card has 6GB or more VRAM. It doesn’t matter if the card is from AMD or Nvidia, as it doesn’t use any technology that’s specific to a manufacturer. It’s the RAM allocation that counts, though it also helps if your card is a newer one. You’ll need to download a lot of data to make this work, but once you’ve got it, you’ll have your own image generation service at your fingertips. There’s no copyright on AIgenerated images, and those created by Stable Diffusion are fully public domain, so you’ll be able to use them in social media posts, college assignments, hilarious pranks, and even more.