AI Learns To Create Images With Minimal Energy Use

A team of researchers led by Aydogan Ozcan at the University of California, Los Angeles (UCLA), has developed a groundbreaking technology that could transform the way generative artificial intelligence (AI) systems operate - by using light instead of electricity. Their findings were recently published in the prestigious journal Nature, Azernews reports.

Modern generative models - especially diffusion-based ones like DALL.E 2 or Stable Diffusion - require enormous amounts of energy. Training these models on large datasets can consume tens of thousands of kilowatt-hours. To put that into perspective, the average household uses around 3,000 to 10,000 kilowatt-hours of electricity per year. So training just one AI model can require as much energy as several homes consume annually.

Even generating a single image isn't energy-free. On traditional GPUs, a single full diffusion process can consume tens of watt-hours. Multiply that by 1,000 steps per image, and you're looking at hundreds of watt-hours - roughly equivalent to running an electric kettle for one or two hours.

The new optical approach, however, is radically different.

Instead of relying on energy-hungry processors, the system uses a beam of light. A digital encoder first creates a "noise circuit" - a pattern that requires almost no power. This pattern is then projected onto a laser beam using a spatial light modulator (SLM). The light, carrying this noise pattern, passes through a second SLM, which transforms it into a fully rendered image. In essence, the laser is doing all the heavy lifting, not the computer.

“Our optical generative models can synthesize countless images with virtually no computing power, offering a scalable and energy-efficient alternative to digital artificial intelligence models,” said Shiqi Chen, the study's lead author.

The implications of this technology are vast. Thanks to its ultra-low energy consumption and high processing speed, it could be integrated into lightweight devices - from virtual and augmented reality systems to smartphones, wearable gadgets, and even AI-powered glasses.

Researchers believe that by harnessing light, AI could finally become both environmentally sustainable and accessible on a much broader scale. The combination of speed, energy efficiency, and image quality may signal the beginning of a new era - one where powerful generative AI doesn't have to come with a massive carbon footprint.

MENAFN30082025000195011045ID1109995564