Show HN: A diffusion model that runs in pure CSS, no JavaScript (metrix187.github.io)

A new project, diffusion.css, has unveiled a fascinating diffusion model capable of generating images entirely in CSS without any JavaScript or WebAssembly involvement. By leveraging pure CSS techniques, users can press a button to initiate the process, where a "seed" value is transformed into noise and then denoised into a 16×16 sprite through a sequence of 12 steps. Remarkably, this innovative model requires a significant 5.2 MB of CSS and features 202,863 weights embedded as literals within the code, making it a unique approach to browser-based image generation. This development is significant for the AI/ML community as it showcases the potential of CSS for implementing machine learning concepts, challenging the notion that complex neural networks must rely on traditional programming languages. The model employs a series of small neural networks, trained on a limited dataset of hand-drawn sprites, and utilizes advanced techniques like DDIM sampling and keyframe animations for rendering images. As the algorithm operates entirely within the browser without any reliance on JavaScript, it opens new avenues for interactive and lightweight AI applications directly in web environments, illustrating a novel intersection between CSS styling and machine learning technologies.