Snap has developed a new technology, SnapFusion, and in the future, smartphones will generate images out of AI.

CTOnews.com, June 20 / PRNewswire-FirstCall-Asianet /-- the Snap Research division of social media company Snap recently announced that it has developed a new technology that can greatly accelerate the speed of AI to generate images, from minutes or even hours to less than two seconds.

This new technology, called SnapFusion, can quickly generate clear images based on text prompts entered by the user. Snap says this is the fastest speed ever published by academia. In a blog post, the company describes the principles and advantages of this technology:

By optimizing the network structure and denoising process, Snap Research makes the model very efficient while maintaining the image quality. As a result, you can now run the model to generate images based on text prompts and get the results on the mobile device in a matter of seconds, rather than waiting for a long time as in other studies.

Snap Research also published a paper (in PDF format) detailing SnapFusion technology and showing some of the artworks generated by the technology. The paper points out that:

Specifically, we propose an efficient UNet, which reduces the computation of the image decoder by identifying the redundancy of the original model and by data distillation. In addition, we enhance step distillation by exploring training strategies and introducing regularization without classifier guidance.

However, it will take some time for the new technology to be available to a wide range of users. But Snap says it "has the potential to provide a powerful driving force for high-quality AI-generated imaging experiences on mobile devices in the future." CTOnews.com speculates that this new technology is likely to be added to future updates to Snapchat mobile apps.