Researchers at the University of California, Los Angeles, led by Aydogan Ozcan, have introduced a fresh approach to generating images with AI that uses far less energy. Instead of the traditional diffusion method—which adds and then gradually removes noise from an image—this new system harnesses light to do the heavy lifting.
A digital encoder, trained on public datasets, starts the process by creating a static pattern while using minimal power. This pattern is then projected onto a laser beam through a spatial light modulator (SLM). A second SLM decodes the pattern, turning the laser’s light into the final image. As lead author Shiqi Chen explains, the optical generative model can produce countless images with almost no computing power.
The team tested the system on a range of images—from familiar celebrity likenesses to works reminiscent of Van Gogh—and found that the results not only matched those of conventional generators but did so with a much smaller energy footprint. If you’ve ever found traditional AI image generators to be both time-consuming and energy‑intensive, this could mark a welcome step forward.
With potential applications in virtual and augmented reality, as well as in smartphones and wearable devices like AI glasses, this approach may help considerably reduce the carbon footprint of digital image creation while maintaining performance.
