At a Glance
- Researchers have developed an AI image generator that utilizes light instead of computation to drastically reduce energy consumption, thereby addressing the sustainability challenges associated with modern AI models.
- The system uses a digital encoder to create a noise pattern, which is then imprinted onto a laser beam and processed all-optically through custom layers.
- This optical generative model successfully created complex monochrome and multicolored images, including handwritten digits, celebrity faces, and artworks inspired by the works of Vincent Van Gogh.
- Results were comparable in quality to those of conventional digital AI generators, but were achieved with virtually no computing power, offering a scalable and energy-efficient alternative.
- The technology’s speed and low power needs make it ideal for future applications in augmented reality, wearable electronics, and other real-time generative tasks.
As generative artificial intelligence tools become increasingly integrated into daily life, their substantial energy consumption poses a significant challenge to sustainable technological development. Addressing this, researchers have developed a novel AI image generator that operates with almost no computing power, instead using the physics of light to perform its calculations. This new approach could drastically reduce the carbon footprint associated with large-scale AI content creation. The research, led by Aydogan Ozcan from the University of California, Los Angeles, was published in the journal Nature.
The new system is an optical generative model inspired by a standard AI process called diffusion. In typical digital diffusion, an AI is trained by adding statistical noise, or digital static, to images until they become unrecognizable, and then learns how to reverse the process. This new model starts with a shallow digital network that generates a noise pattern, which is then imprinted onto a laser beam using a device called a spatial light modulator, a type of configurable liquid crystal screen. This patterned light is then passed through a series of custom-designed optical layers that act as a decoder, all-optically transforming the noise into a new, coherent image.
By replacing power-hungry computer calculations with the passive transmission of light through optical components, the model synthesizes images while consuming virtually no energy beyond that required for illumination. The team successfully demonstrated the system by generating a variety of images, including handwritten digits, fashion products, and full-color artworks in the style of Vincent Van Gogh. The results were found to be comparable in quality to those from conventional, digitally based generative models, demonstrating the viability of this energy-efficient method.
This breakthrough in optical computing opens the door to scalable, high-speed, and environmentally friendly AI. “Our optical generative models can synthesize countless images with almost no computing power, offering a scalable and energy-efficient alternative to digital AI models,” said Shiqi Chen, the study’s lead author, in a statement to Phys.org. Due to its speed and minimal energy requirements, the technology can be integrated into applications such as real-time video generation for augmented reality displays or embedded in compact, low-power devices, including smartphones and AI-powered glasses.
References
- Arnold, P. & Phys.org. (2025, August 26). The AI breakthrough that uses almost no power to create images. Tech Xplore; Phys.org. https://techxplore.com/news/2025-08-ai-breakthrough-power-images.html
- Chen, S., Li, Y., Wang, Y., Chen, H., & Ozcan, A. (2025). Optical generative models. Nature, 644(8078), 903–911. https://doi.org/10.1038/s41586-025-09446-5
