Adobe researchers have made a significant breakthrough in AI innovation with the introduction of VideoGigaGAN, a generative model designed to upscale low-resolution videos by up to eight times their original resolution. This cutting-edge technology boasts superior results compared to traditional video upscaling methods, delivering higher detail without the common issues of glitches and artifacts.
The secret to VideoGigaGAN’s success lies in its clever approach, combining the power of Generative Adversarial Networks (GANs) with smoother results. GANs are known for enhancing still images, but struggle with video, causing flickering and distortion. Adobe’s VideoGigaGAN tackles this issue head-on, producing remarkably natural results that raise interesting questions about distinguishing between genuine footage and AI-enhanced content.
The potential for consumer use in Adobe’s Creative Cloud software is substantial, following earlier experiments like Project Res-Up. Microsoft and Nvidia are also developing similar video upscaling technology, making VideoGigaGAN a significant player in this field.
How Adobe VideoGigaGAN Works
VideoGigaGAN is based on GigaGAN, a type of AI that specializes in upscaling generated content and real photos. However, this technology faces challenges when improving video quality without introducing artifacts.
To address this, Adobe researchers employed various techniques, including:
- Flow-guided propagation module: Ensures consistency among video frames
- Anti-aliasing: Reduces artifacts
- High-frequency feature shuttle: Compensates for sudden drops in detail
These techniques contribute to VideoGigaGAN’s ability to upscale low-quality videos without sacrificing stability or important aspects of the source material. Demo clips on the project’s website showcase its capabilities, transforming blurry 128×128 pixel resolution videos into footage running at 1,024×1,024 pixels.
Future Possibilities of Adobe VideoGigaGAN
Although VideoGigaGAN is still in development, its potential for consumer use is significant. Adobe’s focus on implementing artificial intelligence in its software, as seen in the launch of Firefly and Acrobat’s new assistant, suggests that VideoGigaGAN may be integrated into future products.
While VideoGigaGAN faces obstacles, such as processing videos beyond 200 frames and rendering small objects, its development is an exciting step forward in video upscaling technology. As the field continues to evolve, we can expect to see more innovative applications of AI in video enhancement.
