New AI Creates Images in 4 Steps, Coming to Your Phone

Your next smartphone might create stunning AI images in seconds without ever connecting to the internet.

Researchers at the University of Surrey and Stability AI have developed SD3.5-Flash, an artificial intelligence system that generates high-quality images using just four processing steps instead of the usual 30 to 50. That's a tenfold improvement that makes the technology fast enough to run locally on phones, tablets, and laptops.

Most AI image generators today work like this: they start with random digital noise and slowly refine it into a coherent picture through dozens of iterations. Each step requires serious computing power, which is why popular tools like DALL-E and Midjourney run on massive server farms filled with powerful graphics processors.

The new model takes giant leaps through that refinement process instead of tiny steps. Think of it like jumping across stepping stones instead of carefully walking across each one.

Lenovo has already licensed the technology and announced on March 4 that it will integrate SD3.5-Flash into upcoming devices. That means this breakthrough will soon be in your hands, literally.

"Our SD3.5-Flash model allows users to create images from text descriptions entirely on their device, with no data leaving their hardware," said Hmrishav Bandyopadhyay, the doctoral researcher who developed the model during an internship at Stability AI.

The Ripple Effect

Beyond convenience, running AI locally creates meaningful benefits for privacy, speed, and the planet. When your device handles the work, your prompts and images never get sent to remote servers where they could be intercepted or misused.

Speed improves dramatically too. Without waiting for data to travel to distant servers and back, image generation becomes nearly instant.

Then there's the environmental angle. Large AI data centers consume enormous amounts of electricity and water for cooling. A recent report found that training a single large AI model can emit as much carbon as five cars over their entire lifetimes. Lightweight models running on devices you already own slash that energy demand.

The technical challenge was maintaining image quality while compressing the process. "Achieving this level of efficiency is technically challenging, as it requires compressing a diffusion model to run in only a few steps while maintaining quality," Bandyopadhyay explained.

The researchers detailed their approach in a study uploaded to the arXiv preprint database on September 25, 2024. Their work represents a shift in how we think about AI: instead of always bigger and more powerful, sometimes smarter and more efficient wins.

Yi-Zhe Song, director of the SketchX Lab at the University of Surrey, emphasized the broader mission is making AI more accessible and practical for everyday use.

Powerful AI tools are moving out of distant server farms and into the devices already in your pocket.