Adobe continues to advance its artificial intelligence initiatives, following last year’s launch of Firefly, which was built on the previous SenseiAI platform.
This time, Adobe researchers have developed a new generative AI model called VideoGigaGAN, which can upscale blurry videos to eight times their original resolution with minimal distortion.
Adobe asserts that VideoGigaGAN surpasses other Video Super Resolution (VSR) methods. The company claims that through fine-tuning and specific precautions, the model upgrades images without distorting them.
VideoGigaGAN is described by researchers as a novel generative VSR model that delivers videos with high-frequency detail and temporal consistency.
VideoGigaGAN is built on GigaGAN, a large-scale visual amplifier. Generative Adversarial Networks (GAN), typically known for effectively upscaling still images to higher resolutions, often introduce flickering and disturbances when applied to videos.
In contrast, other video upscaling methods can avoid these issues. Adobe highlights that VideoGigaGAN merges the strengths of both approaches to achieve high video quality with significantly less shaking and distortion.
Adobe has shared numerous examples of VideoGigaGAN’s capabilities. While not every example is flawless and some artificiality is evident, the quality of most is impressively high.
In fact, some examples demonstrate such high quality that it’s difficult to detect the involvement of artificial intelligence. Currently, VideoGigaGAN remains a research project, and it is uncertain when or if it will be released commercially.
You may also like this content
- OpenAI’s AI Course for Educators Sparks Privacy and Security Concerns
- Microsoft Teams to Overcome Language Barriers with AI Translator Feature
- ESPN Introduces AI-Powered Announcer “FACTS” for SEC Nation
Follow us on TWITTER (X) and be instantly informed about the latest developments…
