In short: Stable Diffusion is a phenomenal example of how a picture is worth a thousand words. In fact, by completely truncating the text prompt for image generation, the visual AI could be used to obtain a highly compressed, high-quality image file.
Stable Diffusion is a machine learning algorithm capable of generating oddly complex and (somewhat) believable images simply by interpreting natural language descriptions. The text-to-image AI model is incredibly popular with users, although online art communities have begun to reject AI-based imagery.
Aside from being a controversial example of machine-aided visual expression, stable diffusion may have a future as a powerful image compression algorithm. Matthias Bühlmann, a self-proclaimed “software engineer, entrepreneur, inventor and philosopher” from Switzerland, recently explored the possibility of using the machine learning algorithm for a completely different way of manipulating graphics data.
In its traditional model, Stable Diffusion 1.4 can create works of art thanks to its acquired ability to make relevant statistical associations between images and related words. The algorithm was trained by feeding the “AI monster” millions of internet images, and it requires a 4GB database containing compressed, smaller mathematical representations of the previously analyzed images, which upon decoding are extracted as very small images be able.
In Bühlmann’s experiment, the text prompt was bypassed entirely in order to get Stable Diffusion’s image encoding process to work. This process takes the small source images (512×512 pixels) and converts them to an even smaller (64×64) representation. The compressed images are then extracted to their original resolution with some pretty interesting results.
The developer highlighted that SD compressed images had “far superior image quality” at a smaller file size compared to JPG or WebP formats. The stable diffusion images were smaller and showed more defined detail and showed fewer compression artifacts than those produced by standard compression algorithms.
Could Stable Diffusion have a future as a higher quality algorithm for lossy compression of images on the web and elsewhere? The method used by Bühlmann (for which there is a code example online) still has some limitations as it doesn’t work that well with text or faces and can sometimes produce extra detail that wasn’t present in the source image. The need for a 4GB database and the time-consuming decoding process are also quite a significant burden.
https://www.techspot.com/news/96132-stable-diffusion-weird-visual-arts-boon-image-compression.html Stable Diffusion: Strange for Fine Art, Boon for Image Compression Algorithms?