The Netflix Anime, Aniplex USA and GKIDS YouTube channels are among the many whose videos have been harvested by AI company Runway. 404 Media shared an extensive, redacted spreadsheet detailing the YouTube channels the company used to train its AI video generation tool, despite it violating YouTube's terms of service.
In addition to these, anime-themed channels mentioned in the document include anime recap channels such as Anime Recap Time, Recap-kun, and AniClimax, as well as anime-centric video essayists such as Mastar and Anime Philosopher. Animenz Piano Sheets, Anime Luna, Utoi Anime Talk TV, and Anime Select also appear to have been scraped, as their channels were tagged with “anime.”
Besides YouTube channels, Runway's AI reportedly took down several pirate sites, including an Archive.org page featuring many Studio Ghibli films, as well as Aniwave, Animesuge, 9Anime, Aniwatch and a spinoff of the now-defunct KissAnime. Many independent anime channels were also affected.
Google had previously warned OpenAI that using videos from the platform to train its AI model, Sora, violated the platform's terms of service.
“From a creator's perspective, when they upload their hard work to our platform, they have certain expectations. One of those expectations is that our terms of service will be adhered to. We don't allow people to download transcripts or parts of videos. That's a clear violation of our terms of service. Those are the rules for content on our platform,” YouTube CEO Neal Mohan told Bloomberg.
According to a report from 404 Media, an anonymous former Runway employee revealed that a company-wide effort was underway to compile video information into spreadsheets for use in AI training. The company then used YouTube download tools, specifically YouTube-DL, to download videos in bulk. The company also used proxy services to avoid being detected by YouTube when downloading multiple videos.
“That spreadsheet channel was a company-wide effort to find good quality videos to build a model on. This was then used as input to a large-scale web crawler, which downloaded every video from every channel, using proxies to get around blocking by Google,” the former employee told 404 Media.
It has not yet been confirmed whether all of the videos from these channels were used to train Runway's AI model, but it is worth noting that the channels are categorized with tags that users can use as prompts when they want to generate AI videos in a specific style. Runway, which raised millions of dollars from backers including NVIDIA and Google last year, has not released a statement on the matter.
Featured Image: Sword Art Online: Alicization©2017 Reki Kawahara/KADOKAWA ASCII Media Works/SAO-A Project
