Adobe unveiled an impressive AI-assisted video scaling project

Adobe

Adobe researchers have developed a new generative artificial intelligence model, VideoGigaGAN, that can scale blurred video up to eight times the original resolution.

Here's What We Know

Adobe claims that VideoGigaGAN outperforms other Video Super Resolution (VSR) upscaling methods, delivering more detailed results without unwanted AI-induced artefacts.

Traditionally, generative adversarial networks (GANs) are effective for scaling still images to high resolution. However, they have difficulty with video, creating flicker and other defects. Alternative methods avoid this, but do not provide the same high quality.

VideoGigaGAN aims to combine the best aspects - high quality GAN results with minimal problems of artefacts in the output frames. Adobe has provided several examples demonstrating how the technology works.

Dopis vid @luokai

Revisit Threads

Some fine details in the demo video, such as skin texture and creases, appear to be fully generated, but overall the results look natural. It's hard to tell if the resolution is enhanced with generative AI, which raises questions about the nature of photo and video content.

For now, VideoGigaGAN remains a research project, and it's unknown if Adobe will incorporate the technology into its products like Premiere Pro. Earlier, the company announced another diffusion-based resolution enhancement experiment, Project Res-Up.

Source: The Verge