Adobe's impressive AI upscaling project makes blurry videos look like HD

Adobe researchers have developed a new generative AI model called VideoGigaGAN that can upscale blurry videos to eight times their original resolution. Adobe claims in an article published April 18 that VideoGigaGAN is superior to other Video Super Resolution (VSR) methods because it can provide finer-grained detail without adding any “AI weirdness” to the footage.

In short, Generative Adversarial Networks (GANs) are effective at upscaling still images to higher resolution, but struggle to achieve the same with video without introducing flickering and other unwanted artifacts. Other upscaling methods can avoid this, but the results are not as sharp and detailed. VideoGigaGAN aims to provide the best of both worlds – the higher image/video quality of GAN models with fewer flicker or distortion issues in the output images. The company has provided several examples here showing its work in full resolution.

Some of the finer details in the demo clips provided by Adobe appear completely artificial, such as the skin texture and wrinkles in the example below, but the results appear impressively natural. It would be hard to say that generative AI was used to improve resolution, which could extend the “what is a photo?” debate to video.

This is just a research preview, so there is no guarantee that Adobe will make VideoGigaGAN available to consumers through Creative Cloud software like Premiere Pro. The company previously previewed a separate diffusion-based upsampling experiment called Project Res-Up during its MAX event in October 2023, which also improves the quality of GIFs and low-resolution video footage. And Adobe is not alone in this work, as both Microsoft and Nvidia have also developed their own VSR upscaling technology.