From drafting emails and writing code to creating art and developing new medicines, generative AI tools are becoming increasingly vital in both professional and personal settings. But this growth comes at a cost—AI models typically demand massive amounts of computing power, memory, and energy. Researchers are now exploring ways to reduce that burden, and one new approach could be a game-changer.
A research team led by Aydogan Ozcan at the University of California, Los Angeles has unveiled an image generation system that requires virtually no power. Their findings were recently published in Nature.
How AI Image Generators Normally Work
Most AI art tools use a technique called diffusion. During training, they take a dataset of images and gradually add random noise—like digital static—until the pictures vanish completely. When prompted later (for example, “draw a house”), the system begins with a noisy, static-filled screen and works backwards, stripping away the noise step by step until the final image emerges.
This method can produce stunning results but is computationally expensive. Generating millions of images through diffusion requires vast amounts of processing time and energy.
A Light-Powered Alternative
The UCLA team’s method still begins with a digital encoder that generates the noise pattern, but from there the process is radically different. Instead of relying on billions of calculations, the pattern is projected onto a laser beam using a spatial light modulator (SLM). A second SLM then decodes the beam, producing the finished image.
In essence, the system uses light to perform the image synthesis instead of digital computation. “Our optical generative models can synthesize countless images with almost no computing power, offering a scalable and energy-efficient alternative to digital AI models,” explained lead author Shiqi Chen.
Tests showed the setup could generate images similar in quality to standard diffusion models, including portraits, butterfly photographs, and colorful Van Gogh-style artworks—yet with a fraction of the energy consumption.
Future Potential
Because it’s so fast and efficient, the optical system could open the door to new applications. It might be integrated into lightweight devices such as smartphones, smart glasses, or wearable electronics. It could also power content for augmented and virtual reality environments without draining energy resources.
Beyond practical uses, this approach could also make AI more sustainable, cutting down the carbon footprint associated with large-scale generative tasks.
thetechtrident.com 