Data Backup Digest

Graphic artists and designers have been obsessing over image quality for years. Even in the days before digital was the norm, photographers went to great pains – and great expense – to purchase the best quality film, lenses, and cameras that money could buy. While traditional photographers like this still exist, they operate side-by-side with the digital artists of modern times.

In either case, image quality is the key. Not only is it the key to capturing an iconic moment in time, but it’s can help others gain a greater appreciation for the art form, too.

Digital Image Quality

Given the prevalence of digital mediums today, it’s difficult to find an image, photograph, or other work of art that hasn’t been digitized in some way. In some cases, this results in a greater final product. When an expert uses technology to restore or enhance a photograph that is decades old, for example, it can result in a much clearer image overall.

However, the opposite can be true, too. In some cases, especially when storing digital images online, these files can become so compressed that the final image quality suffers as a result. While compression is used in almost all digital images, with the exception of those saved in a raw format, any downgrades in quality are usually unnoticeable to the human eye.

Regardless, some graphic designers, artists, and digital image enthusiasts still desire the best possible image quality – even if that requires significantly more disk space than a compressed file.

Enter Stable Diffusion

In a nutshell, Stable Diffusion is a type of AI-driven image synthesis that is used by modern text-to-image converters to transcribe words into image form. The process works by scanning millions of pre-existing digital images, pulled directly from the internet, and forming associative bonds between these images and any related words.

From there, the information is stored and assigned an appropriate weight – a specific mathematical value – where it’s effectively compressed into an area referred to as latent space. When using the latest iteration of Stable Diffusion, version 1.4, this results in a file that is approximately 4GB in size. However, the single file actually represents millions upon millions of separate, unique digital images.

Matthias Bühlmann, a software engineer from Switzerland, recently isolated the Stable Diffusion model and, instead of using it to convert text to an image, used it to encode and compress traditional digital images. He discovered that digital images using Stable Diffusion maintain a much higher image quality than some of the most popular digital formats, including WebP and JPEG.

Not only that, but the resulting file size was smaller when using Stable Diffusion than either WebP or JPEG. Considering the latter two formats have been considered a standard online for years, it’s easy to see how significant this breakthrough really is.

Unfortunately, there are some limitations when compressing traditional digital images via Stable Diffusion. Firstly, it doesn’t work well with faces or text, including letters and numbers. In certain scenarios, it even creates specific features that were never present in the original image. Finally, the process requires the 4GB weight file and a significant amount of processing time.