The photo and video sharing platform, Snapchat, announced the development of a new artificial intelligence model for generating images that works directly in smartphones.
According to a team of researchers collaborating with the company “Snap”, which developed the “Snapchat” application, the new model is called (SnapGen), and it allows the generation of high-resolution images within seconds on advanced phones, and is distinguished by its small size compared to other common models, as it requires ( SnapGen is only down to 379 million parameters, making it about seven times smaller than similar models. It’s also smaller than Huawei’s PixArt-? model, which is lightweight. It is also intended for use in phones.
The new model is considered the fastest in terms of performance. When measuring the extent to which images match textual descriptions, SnapGen recorded a score of 0.66 on the GenEval scale, superior to SDXL, which recorded a score of 0.55, as it can create an image with a resolution of 1024 x 1024 pixels in only about 1.4 seconds. Via iPhone 16 Pro Max.
The research team relied on training the small model (SnapGen) with artificial intelligence models to ensure the accuracy and speed of generation. They also developed a special training process capable of recognizing difficult tasks for the small model, and modifying the teaching method to achieve the best results.
It is noteworthy that the company displayed the performance of SnapGen through a demo application for iOS, to highlight its practical capabilities in daily use.