Adobe has recently been dominating tech news with its myriad of AI-driven announcements. From innovations in Premiere Pro that simplify object addition and removal to pioneering text-based image creation in Photoshop, Adobe’s AI endeavors continue to push the boundaries of creative technology. The latest addition to this innovative streak is VideoGigaGAN, an experimental feature designed to revolutionize the way we upscale videos.
According to a report by The Verge, VideoGigaGAN sets a new standard in video upscaling, capable of enhancing video quality by up to eight times without succumbing to common issues like flickering or distortion. This development is a significant step forward from existing Video Super Resolution (VSR) methods, which commonly struggle with such artifacts. Adobe’s new technology promises not just to avoid these pitfalls but also to add unprecedented sharpness and detail to videos—a feat that has been challenging to achieve until now.
VideoGigaGAN surpasses other VSR technology by using a combination of strategies to improve video quality. Traditional methods often introduced unwanted artifacts and blurriness during the upscaling process, particularly when using General Adversarial Networks (GANs). In contrast, VideoGigaGAN incorporates temporal attention to minimize artifact accumulation over time, feature propagation to enhance details, anti-aliasing to smooth out the image, and a unique “HF shuttle” mechanism designed to transport high-frequency features effectively. This comprehensive approach allows the system to fabricate details with remarkable realism, including intricate textures like skin and fine hairs, even down to the delicate strands on a swan’s feather.
However, it’s important to note that VideoGigaGAN is crafting details from nothing. This synthetic enhancement means that, despite its impressive capabilities, it might not be suitable for applications requiring absolute veracity, such as forensic video analysis. Instead, its strengths lie in creative and entertainment contexts, where its ability to generate lifelike details can significantly enrich visual storytelling.
Adobe’s new upscaling tech draws upon the foundational work of GigaGAN, a sophisticated image upsampler, now adapted and extended for video. This adaptation is crucial, given that enhancing video fidelity is inherently more complex due to the added dimension of time. The potential applications for VideoGigaGAN are vast. Integrating this technology into professional video editing software like Premiere Pro or After Effects could dramatically elevate the quality of low-resolution footage, providing filmmakers and content creators with powerful tools to enhance their projects.
Despite the excitement surrounding Adobe’s announcement, the deployment of AI in enhancing video content, particularly concerning human subjects, remains a contentious issue. The ethical implications of altering human appearances with AI have sparked debate, especially in light of recent controversies involving AI-generated images in documentaries.
While Adobe has yet to confirm plans for the widespread implementation of VideoGigaGAN, the feature’s current capabilities demonstrate a promising step towards seamlessly bridged gaps between vision and reality in video production. As AI technology continues to evolve, so too will the possibilities for creative expression, with Adobe at the forefront of these advancements. Meanwhile, other major players in the tech industry, including NVIDIA, Microsoft, and Blackmagic Design, are also exploring AI upscaling, signaling a broader move towards smarter, more capable video enhancement tools that could redefine the future of digital media.
Source
