Researchers from the National University of Singapore have developed the MagicAnimate neural network, which is able to animate images. According to the creators, it largely exceeds the efficiency of similar solutions. To create an animation, #MagicAnimate takes the original image and a sequence of frames from the DensePose library. Then, using a diffusion model, it generates frames of the future animation based on the reference image. The result is "glued" into a finished clip. The developers note that the neural network itself completes the missing elements of the object, and the user can specify a refining query for generation. Examples of MagicAnimate's work show that it effectively "animates" even non-existent characters created by the #DALLE3 neural network, and also copes well with animating group photos. For free, users can upload up to 5 photos per day and up to 30 photos in total from 1 account. Photo processing takes from 30 to 60 minutes depending on the workload of the service, with priority given to users with a subscription.