Enhancing YouTube Shorts with AI-generated Descriptions
Bite-sized mobile videos are incredibly popular today, with YouTube Shorts being viewed over 50 billion times a day. However, these short videos often lack descriptions and titles, making them difficult to find through search. To address this, YouTube has partnered with us to introduce Flamingo, our visual language model, which generates video descriptions based on the initial frames of Shorts videos. This metadata is stored in YouTube’s database, allowing for better categorization and improved search results.
Collaboration with YouTube to Enhance the Platform
We have partnered with YouTube to enhance their platform and provide a better user experience. YouTube’s mission is to give everyone a voice and show them the world, and we are working together to optimize safety, reduce latency, and enhance the experience for viewers, creators, and advertisers.
Improving Video Compression with AI
Video compression is becoming increasingly important, especially with the surge in video consumption during the COVID-19 pandemic. Together with YouTube, we have explored the potential of our AI model, MuZero, to enhance the VP9 codec, which compresses and transmits videos over the internet. Since implementing MuZero on YouTube’s live traffic, we have achieved an average 4% reduction in bitrate across a wide range of videos. This improvement allows for faster loading times, better resolution, and reduced data usage for millions of viewers worldwide.
Ensuring Brand Safety for Creators and Advertisers
Since 2018, we have been working with YouTube to educate creators on ad-friendly content and ensure that ads are placed alongside suitable videos. Together, we have developed a label quality model (LQM) that accurately identifies videos aligning with YouTube’s ad-friendly policies. This improvement enhances trust among viewers, creators, and advertisers.
Automating Chapter Creation with AI
Creators have started adding chapters to their videos to make content navigation easier. To simplify this process, we collaborated with YouTube to develop an AI system called AutoChapters. This system automatically processes video transcripts, audio, and visual features to suggest chapter segments and titles. Currently available for 8 million videos, AutoChapters will be expanded to over 80 million videos in the next year. This feature saves viewers time searching for specific content and allows creators to dedicate more time to other aspects of video creation.
Continued Improvement Through Collaboration
As technology and society continue to evolve, we remain committed to enhancing Alphabet technologies and products through our AI research. Our collaboration with YouTube has already made significant improvements, and we look forward to making more positive changes in people’s lives moving forward.