Independent creators are beginning to license their content as data to train AI video models.
High production value publisher content accounts for most licensing activity for foundation video models. But high-quality creator video is also finding its way into the licensing mix and viewed as an equally viable and valuable source of the data that’s used to train AI video models.
Paths for creators to license video content to train AI models are nascent but forming alongside the fledgling market for content licensing for AI training. “Many creators don’t even know there are avenues today to potentially license their data,” said Moonvalley cofounder and CEO Naeem Talukdar.
Most creator video licensing activity so far has likely occurred through an emerging crop of startups, which facilitate content deals between rights holders and AI developers. Several such marketplaces, including Troveo, Avail’s Corpus and Protege Media (formerly Calliope Networks), are now sourcing high-quality video from creators, though professional content from production companies or distributors contributes a greater share of the total aggregated data.
Licensing creator video content to AI developers is one of the main ways agency talent are contemplating or starting to engage with generative AI, according to digital talent and strategy reps at two major agencies VIP+ spoke with on background. Even when deals haven’t progressed, reps are approaching some digital talent to discuss licensing as a potential monetization opportunity to consider.
In some cases, AI companies proactively negotiate licensing directly with content creators.
Those creators who are engaging in licensing may not always define as social media creators or influencers. For example, the vast majority of the training data licensed for Moonvalley’s video model Marey came from creators such as independent filmmakers or hobbyist videographers, some of which was sourced from data broker Troveo, said Talukdar.
“We work with some creators who have incredibly cinematic, high production value content,” said Troveo founder and CEO Marty Pesis.
VIP+ Unearths Gen AI Insights From All Angles — Pick a Story
Yet some creators considering the licensing opportunity are active on social platforms. Most notably, that has meant YouTubers, sources told VIP+. Licensing deals have occurred among a cohort of “very prolific VOD-centric creators,” such as “longtime YouTubers who have thousands of hours of content,” said Chris Wittine, a digital media agent at CAA.
Troveo’s network includes 1,300 licensors, many of which are YouTube, TikTok and Instagram creators, who Pesis estimated are contributing 25% of Troveo’s total 1 million hours of processed video to date. Though its primary focus has been film and TV, Protege Media general manager Dave Davis said the company has a “decent bench of YouTube creator video” through direct relationships with several YouTubers.
In addition to working with individual creators, Avail also sources creator video for Corpus through large YouTube creator networks such as Viral Nation.
While YouTube’s terms of service prohibit third-party scraping, the platform now offers some of its creators a toggle feature allowing them to opt into AI training by 18 third-party AI companies, though without any immediate way of being compensated (such as by licensing).
Perhaps more to the point is creators’ claim and the strong likelihood that AI developers, including Apple, Nvidia, Anthropic, OpenAI (transcripts for GPT-4, video for Sora) and Google itself, have already trained on YouTube content without consent or compensation to creators or publishers on the platform.
Like rationales among some publishers that have gone forward with licensing for AI training, creators are beginning to see it as a monetization opportunity in the face of likely scraping. “Licensing is definitely an opportunity that we’re seeing our talent wanting to explore,” a digital strategy rep told VIP+. “Especially for our YouTubers, the way we see it is these companies are training on your material anyway, so why don’t we help you unlock a fee? We see it basically as a source of passive income for some of our clients, and we’re starting to have these conversations more and more.”
Frequently, content creators are licensing raw, unused footage such as B-roll, multiple sources told VIP+. Bloomberg previously reported that OpenAI, Google and Moonvalley are licensing unused video footage from creators. In doing so, creators newly monetize cutting-room floor material and preclude third-party scraping altogether.
For some AI developers, unused material is valuable because this footage isn’t already publicly available (something many AI companies often already believe is fair use for training).
But the reasons are also practical: Unused footage is available in much greater volumes than what’s actually used in the final production and is also less likely to be affected by various production treatments or undesirable editing, such as fast-cuts or in-video text. “Often, it’s not their primary videos they distribute but could be any video they shoot, like B-roll that isn’t doing much for them but for us is super valuable,” said Talukdar.
Still, creators face some barriers to licensing, as AI developers tend to be selective about creator content. Developers building video models need training data to be high quality, available in large volumes (e.g., a minimum of 1,000 hours from an individual creator) and diverse in subject matter. Many creators will fall short of these criteria, sources said.
For example, some simply won’t have enough material to “move the needle,” or it may not be high enough resolution (e.g., 4K) or cinematically shot. Likewise, most foundation model developers aren’t prioritizing “talking head” videos of a person speaking straight to camera, which excludes many vloggers, although sources said talking heads have more value for video avatar companies.
A mismatch in content valuation expectations has also stalled interest in licensing among some creators, sources said. AI developers value aspects about content atypical in the creator economy, and a creator’s follower count or video views don’t factor into valuation at all.
Sources said rates have varied and the market hadn’t yet settled into a standard, though reported figures have ranged $1 to $2 or as high as $6 per minute of video, with premiums for professional-grade or more bespoke or specialized content versus generic or stock footage. For prominent creators, sums of a few thousand dollars may not move the needle enough for them to take on risks of supplying large amounts of material for AI training purposes, particularly if that content contains their own likeness.
Yet for those creators who fit the criteria, interest in licensing is the same as it would be for any rights holder who owns their own content. “We’re going to work with people who want to extract the value that’s there,” said Troveo’s Pesis.
