AI Companies Pay Content Creators for Unpublished Videos to Train AI Models

TikTok, YouTube creators sell unpublished content for AI training

OpenAI, Google, and AI media company Moonvalley are paying YouTubers to purchase unused video footage in order to train their AI algorithms, Bloomberg reported. Here, "unused" footage refers to unpublished content, for which digital creators are being paid roughly $1-$4 per minute of footage depending on video quality and format. Besides YouTubers, creators on TikTok and Instagram have also turned to this opportunity as an avenue to diversify their income beyond advertising deals with brands, according to the news portal. This development is significant given Big Tech platforms were previously critiqued for training their AI applications on content without seeking consent or licenses. Recently, a lawsuit filed by a group of U.S-based authors alleged that Meta CEO Mark Zuckerberg approved the use of "pirated" versions of copyright-protected books to train its AI models. To explain, Meta used the dataset belonging to file-sharing project LibGen (currently involved in legal suits for online piracy) for training purposes. In July 2024, other AI giants including Apple, Nvidia, Salesforce, and Anthropic were under fire for using transcripts from over 170,000 YouTube videos without the consent of their creators and in breach of YouTube's terms of service. Such usage has been termed "free-rider behaviour", where organisations like OpenAI and Microsoft, among others use content without offering compensation and in violation of copyright laws. However, OpenAI and Meta have signed deals with news organisations and YouTube with major music labels to access their content. Further, even as companies like Google deployed features enabling website publishers to opt out of its crawlers scraping data to train AI models, concerns over crawlers scraping and indexing websites persist. Moving forward, as companies capitalise on the AI boom, experts now warn of data exhaustion for AI training purposes. OpenAI CEO Sam Altman and more recently xAI owner Elon Musk have also expressed concerns about human knowledge for such purposes being near depletion. According to a Data Provenance study, a rise in restrictions on crawling and training of AI models through data from web sources was found. The research, which audited 14,000 web domains, also concluded that such restrictions would also introduce bias in data used for training purposes toward older and "less fresh content", lacking in diversity. However, companies can explore alternate tactics for AI training amid rising constraints. For instance, Musk believes that synthetic data (created through AI) can 'fine-tune' new systems and train AI. However, this process is subject to risks like cybersecurity, bias propagation, and a surge in errors through hallucinations. Additionally, such agreements with digital creators to utilise their unpublished video footage to train AI models could help potentially aid this process. While most agreements prohibit AI companies from creating digital replicas or mimicking exact scenes, other questions persist. MediaNama reached out to OpenAI, Moonvalley, and Google with questions and is awaiting their response. We will update the story as soon as we hear from them.

AI Companies Are Paying Content Creators For Their Unpublished Videos

Content creators are having an unexpected boon: artificial intelligence companies are buying their unpublished videos for training purposes. It has been a little over two years since generative AI exploded into the world but already tech companies have burned through all the "available data" on the internet. In a bid to plug the gap and continue building AI video generators, Bloomberg reports that AI companies, including Google and OpenAI, are paying content creators for access to their unpublished videos. This content holds value precisely because it is material that never saw the light of day and is therefore unique. Sources tell Bloomberg that these AI firms are paying between $1 and $4 per minute of footage with the price depending on the quality or format. 4K videos, for example, will go for a higher price as well as specialist imagery such as that taken from a drone. For regular content creators posting to platforms like YouTube, Instagram, and TikTok, they can expect to fetch between $1 and $2 per minute. It could be a bonus revenue stream for creators who have filmed a video but for whatever reason decided not to post it. One social media agent tells Bloomberg that he thinks the window for selling content to AI companies won't be open for long. "It's an arms race and they all need more footage," says senior vice president of creators at Wasserman Dan Levitt. "I see a window in the next couple of years where licensing footage is lucrative for creators who are open to doing so. But I don't think that window is going to last that long." There are third-party agencies that mediate between IP owners and AI companies; such as Troveo AI and Calliope Networks which negotiate terms on behalf of content creators. It could be an opportunity for photographers and videographers willing to enter into a deal. "All the companies building video models we're either working with or are in our pipeline right now," says co-founder and chief executive officer of Troveo Marty Pesis, who adds his company has paid out over $5 million to creators. These deals between AI companies and creators generally have terms forbidding AI companies from creating digital replicas of the creators themselves or making exact replicas of the content. AI companies face a challenge to keep making bigger and better generative AI models which require huge amounts of data. The controversial practice of taking pictures, videos, and text from the internet without the copyright holders' permission is apparently over after everything has already been taken. During an X livestream last week, Elon Musk said the way to plug the gap is with synthetic data. "The only way to supplement [real-world data] is with synthetic data, where the AI creates [training data]," Musk says. "With synthetic data ... [AI] will sort of grade itself and go through this process of self-learning." However, this method is not totally proven. One study suggests that AI models trained on AI images start churning out garbage images with the lead author comparing it to species inbreeding. "If a species inbreeds with their own offspring and doesn't diversify their gene pool, it can lead to a collapse of the species," says Hany Farid, a computer scientist at the University of California, Berkeley. Nevertheless, Microsoft, Meta, OpenAI, and Anthropic are all reportedly using synthetic data to train AI models with.

YouTubers are selling their unused video footage to AI companies

YouTubers and other digital content creators are selling their unused video footage to artificial intelligence companies seeking exclusive videos to better train their AI algorithms, often netting thousands of dollars per deal. OpenAI, Alphabet's Google, AI media company Moonvalley and several other AI companies are collectively paying hundreds of content creators for access to their unpublished videos, according to people familiar with the negotiations. That content, which hasn't been posted elsewhere online, is considered valuable for training artificial intelligence systems since it's unique. AI companies are currently paying between $1 and $4 per minute of footage, the people said, with prices increasing depending on video quality or format. Videos that are shot in 4K, for example, go for a higher price, as does non-traditional footage like videos captured from drones or using 3D animations. Most footage, such as unused video created for networks like YouTube, Instagram and TikTok, is selling for somewhere between $1 and $2 per minute.