Researchers at the University of Surrey have created SD3.5-Flash, an AI image generator requiring just four processing steps-10 times fewer than conventional models. The breakthrough enables local execution on smartphones and laptops, eliminating cloud dependency for enhanced privacy, speed, and reduced energy use. Lenovo has licensed the technology for its Qira platform, with devices expected soon.
The system compresses diffusion process into larger leaps while preserving quality. 'Our model creates images entirely on-device with no data leaving hardware,' said researcher Hmrishav Bandyopadhyay. Tests show comparable results to traditional models despite drastic step reduction. Yi-Zhe Song of Surrey's SketchX Lab added, 'SD3.5-Flash puts creative tools in users' hands while keeping data private and cutting energy demands.'