Looking for a fun AI way to push your iPhone compute to its limits? Check out this impressive project that runs a 2GB Stable Diffusion model 100% offline on your iPhone.

Every year we get new iPhones that claim to be faster and better, but the incremental improvements rarely push the hardware to its limits. Stable Diffusion changes that — a full text-to-image generative model running entirely on-device, no cloud required.

The key challenge is fitting a 2GB model into the constrained memory of a mobile device while maintaining reasonable generation times. Through careful optimization of the model pipeline — including the text encoder, diffusion model, and image decoder — it becomes possible to generate images directly in your pocket.

This is a glimpse of where edge AI is heading: powerful generative models running locally on consumer devices, with no network latency and full privacy.

Read the full post