From Midjouney to Bing Image Creator, there are plenty of prolific AI-based art generators. However, depending on the text prompt and other factors, such renderers can sometimes take a while to do their job.
But Snap, the parent company of Snapchat, has rolled up its sleeves to keep that time to a minimum, announcing that its Snap Research division has developed a new technique that is expected to speed up render time for prolific AI art makers.
Snap said in a blog post that the new method is called SnapFusion, and it reduces rendering time to less than two seconds through productive artificial intelligence. This was said to be “the fastest time ever published by the academic community”.
“Snap Research has made this breakthrough by optimizing the network architecture and making the noise removal process incredibly efficient while maintaining image quality. Therefore, as other research has revealed, it is now possible to run the model to generate images based on text prompts and get crisp and clear images on mobile devices in mere seconds instead of minutes or hours.”
The research paper details more about SnapFusion, along with a series of examples of artwork created with the new technique: “Specifically, we propose an efficient UNet that identifies redundancies of the original model and reduces the computation of the image decoder through data purification. We are also improving cascade distillation by exploring training strategies and introducing regularization from classifier-independent guidance.”
Unfortunately, Snap says it will take some time before this new method becomes generally available. Still, he adds, it “has the potential to power high-quality productive AI experiences on mobile devices in the future.” We can look forward to seeing this new technique added to a future update of the Snapchat mobile app.
51709
