YouTube’s Digital Fingerprints: How Content ID Protects Videos
A top YouTube engineer gave Streaming Media Europe attendees a peek at how the video giant operates.
YouTube's Content ID creates a fingerprint of uploaded reference material, then compares it not only to newly uploaded videos, but to every video ever uploaded to the video sharing powerhouse. And it does it in minutes.
"It uses a hell of a lot of computers to do that," said Oliver Heckmann.
Heckmann, the engineering director for YouTube Europe, gave the second day keynote address at Streaming Media Europe. In a rare peek inside YouTube's operations, he gave stats on the site, explained how the Content ID system works, and gave a glimpse at the upcoming video ad formats YouTube will use to keep both advertisers and viewers happy.
YouTube currently receives 48 hours of uploaded video each minute, Heckmann revealed. It serves 3 billion video views per day (a number that's been growing by 50 percent year-after-year) to 500 million unique users worldwide. Besides being the largest video site in the world, it's also the second largest search engine (after parent Google) - and it's not even a search engine.
Money for Videos
"How does YouTube make money?" is a question Heckmann often gets, so he provided conference-goers with some insight into YouTube's financials. Advertisements are key.
"Almost all the money we make comes from us showing ads." Heckmann said.
YouTube makes the most from the seven ad spots available on its homepage, which gets 60 million impressions per month (and is visited by one in three Americans each month). Search pages and watch pages also do big traffic.
On the watch pages, YouTube only shows pre-roll ads on partner-uploaded or -owned videos, revenue that it splits with the partners. That sharing creates an incentive for content owners to upload more material.
YouTube Content ID
YouTube began work on Content ID over four years ago, Heckmann said, as a way to completely automate takedown requests. As the site grew, it found the task of manually removing copyrighted material onerous. With Content ID, content owners are able to remove all copyrighted works posted past and present with no need for assistance - or they can leave them up and make money from them.
The system is flexible enough to identify videos only partly comprised of copyrighted material, or where the dimensions or colors are altered. Content owners can create rules to block all pirated material, allow it and track its use, or allow it and monetize its views. Heckmann initially didn't see the value of the allow and track option, but said many content owners were thrilled to get free data on where in the world their content was popular.
Content owners need to upload low-resolution reference files for YouTube to check against. Once YouTube has created a reference file, it compares that to its library of millions of videos, and does it in minutes so that permitted videos can quickly go live. Over 1,000 partners worldwide, including CNN, use the system.
YouTube is moving into TrueView advertising, where advertisers pay only when their videos are watched. Heckmann showed off two existing TrueView ads and two that are in progress.
"We strongly believe in the importance of user choice in video ads," said Heckmann.
The two existing YouTube TrueView ads he showed were in-search (which show up in search listings) and in-stream (where viewers get the option to skip a pre-roll ad after a few seconds of watching it). Still in development are in-display (which show up in a standard display slot) and in-slate (where users are given the choice between three ads).
Scroll down to view the entire video.
Keynote: YouTube: A Peek Inside
Oliver Heckmann, Engineering Director, YouTube Europe
Oliver Heckmann is the Engineering Director for YouTube in Europe. He leads a team of 70 software engineers and researchers and is responsible for many partner and user facing projects, including Content ID, Annotations, Insight and Analytics, Video Ingestion and the GData API, as well as large parts of YouTube monetisation, Ads Targeting and Ads Quality. Before working on YouTube Oliver was a manager on the Gmail team. Prior to joining Google in 2006, he led a research team at the Multimedia Communications Lab in Darmstadt, Germany. In 2004, Oliver won an award for the best German Computer Science Dissertation, followed by multiple other awards, including a Google Founders Award.
The announcement that many saw coming is now official: YouTube is transcoding to WebM, but still supporting H.264.
No special camera required: 2D videos becomes 3D with one click, thanks to beta tool.
Reorganization emphasizes channels, letting viewers quickly find new videos they'll like.