Researchers at the UCLA Samueli School of Engineering have developed a new technology that uses photonics to generate images, potentially reducing the energy and computational resources needed for generative artificial intelligence. The research team’s findings were published in Nature.
Traditional generative AI systems, such as those powering popular chatbots and image generators, require significant energy and water to operate due to their reliance on extensive digital computation. These demands contribute to a large carbon footprint and raise sustainability concerns about current AI infrastructure.
The UCLA researchers’ approach leverages optical computing, using light instead of electricity to process information. Their system combines a digital encoder with an optical decoder, generating images in a single step after initial encoding. This differs from existing models that often need hundreds or thousands of computational steps per image.
“Our work shows that optics can be harnessed to perform generative AI tasks at scale,” said Aydogan Ozcan, senior author of the study and Volgenau Professor of Engineering Innovation at UCLA Samueli. “By eliminating the need for heavy, iterative digital computation during image inference, optical generative models like ours open the door to snapshot, energy-efficient AI systems that could transform everyday technologies.”
The team demonstrated their technology by producing new images across various datasets, including handwritten digits, fashion items, butterflies, human faces, and Van Gogh-inspired artwork. According to standard image quality metrics used in the field, the outputs from their optical model were comparable with those generated by advanced digital diffusion models.
One feature of this system is its security capability: multiple patterns or images can be encoded using different wavelengths of light and only decoded with matching decoder surfaces. This “key-lock” mechanism restricts unauthorized access to generated content and has potential applications in secure communication and anti-counterfeiting measures.
The researchers also see possibilities for integrating this low-power technology into portable devices such as smart glasses or augmented reality headsets for real-time use. Broader applications could include biomedical imaging and diagnostics as well as edge computing where data processing occurs locally rather than in remote servers.
The study’s first author is Shiqi Chen; other co-authors include Yuhang Li, Yuntian Wang, and Hanlong Chen—all affiliated with Ozcan’s research group at UCLA Samueli. The research received support from UCLA Samueli’s V. M. Watanabe Excellence in Research Award.
Aydogan Ozcan also serves as associate director for entrepreneurship at the California NanoSystems Institute at UCLA and is a professor with the Howard Hughes Medical Institute.
###
