UCLA scientists use laser light to make AI image generation 90% more energy efficient

UCLA scientists have developed a new AI image generation system that uses laser light as a decoder instead of traditional digital computation, dramatically reducing energy consumption during the inference process. The breakthrough could make AI more sustainable while opening doors for energy-efficient wearable AI devices like smart glasses.

How it works: The system combines a shallow digital encoder with a diffractive optical decoder that uses light to generate images instantly.
• The heart of the system is a liquid crystal screen called a spatial light modulator (SLM) that imprints image information into a laser beam.
• By passing through a second decoding SLM, the system produces images in a “snapshot” rather than through thousands of iterative computational steps.
• This approach eliminates the heavy, iterative digital computation typically required during inference, dramatically cutting energy usage.

The results: Testing showed the optical system can produce both black-and-white images and Van Gogh-style artwork comparable to advanced diffusion models while using only a fraction of the energy.
• The system avoids the computational burden of traditional digital decoders that require thousands of processing steps.
• Light-based decoding also offers improved security and privacy by making content inaccessible without the correct decoder.

Why this matters: While individual AI queries generate only 2-3 grams of CO₂, the scale of usage creates significant environmental impact—ChatGPT users alone created over 700 million images in one week in March 2025.
• As AI adoption grows, creating more energy-efficient models becomes increasingly critical for sustainability.
• Current AI systems contribute to a steadily growing carbon footprint as platforms gain more users.

What experts are saying: The research represents a significant advancement in practical optical computing applications.
• “Our work shows that optics can be harnessed to perform generative AI tasks at scale,” said Aydogan Ozcan, UCLA’s senior author of the study.
• Alexander Lvovsky from the University of Oxford told New Scientist: “This is perhaps the first example where an optical neural network is not just a lab toy, but a computational tool capable of producing results of practical value.”

Looking ahead: While widespread integration isn’t expected in the near term, the technology shows particular promise for wearable AI systems where energy efficiency is essential.
• AI glasses and similar devices would benefit significantly from reduced power consumption.
• The research demonstrates there’s substantial room for sustainable improvements in AI technology.