Making a generative AI image might take less than two seconds on a smartphone in the futureJuly 14, 2023
Snap, the parent company of Snapchat, has announced a new technique called SnapFusion that aims to speed up the process of generating AI art. According to the image custom software development company’s blog post, SnapFusion can produce an image using generative AI in less than two seconds, making it “the fastest time achieved by the academic community so far”.
Snap Research achieved this breakthrough by optimising the network architecture and denoising process, making it highly efficient while maintaining image quality. This means that the model can generate clear and detailed images based on text prompts within seconds on mobile devices, unlike other methods that may take minutes or even hours. Currently, there are various art creators like Midjouney, Adobe Firefly and Bing Image Creator, all enhanced with AI, but they can be slow depending on the input and other factors.
An excerpt from the research paper which goes into more detail on SnapFusion, along with a number of artworks created by the new technique, the company states:
“Specifically, we propose an efficient UNet by identifying the redundancy of the original models and reducing the computation of the image decoder via data distillation. Further, we enhance the step distillation by exploring training strategies and introducing regularisation from classifier-free guidance”.
Although this new method is not yet available to the general public, Snap believes it has the potential to greatly enhance generative AI experiences on mobile devices for Android and iOS applications. It is possible that this technique will be incorporated into a future update of the Snapchat mobile app, but it may still take some time before that happens.