OpenAI IP promises ring hollow to Sora losers

analysis OpenAI's new Sora 2 video generator has become the most popular free app in Apple's App Store since launching last week. It has also drawn ire from Hollywood studios and anyone whose characters and storylines appear in the user-generated content without their explicit permission. Now CEO Sam Altman says rightsholders will be getting greater control over how their properties are used - and may even be paid.

Altman said in a recent blog post that OpenAI was making two big changes to Sora 2, which was released on Tuesday September 30. He noted that this backtrack occurred after "taking feedback from users, rightsholders, and other interested groups."

"First, we will give rightsholders more granular control over generation of characters, similar to the opt-in model for likeness but with additional controls," Altman said.

The likenesses Altman mentioned seem to refer to Sora's new "cameo" feature, which allows people to grant permission for others to use their face and voice in Sora videos. According to OpenAI's help page, users who want to be featured in cameos have to record video and audio of themselves for Sora to ingest. Users who chose to hand their likeness over to OpenAI can restrict the use of their appearance to only their own videos, or to those made by approved people, mutual followers, or anyone on the app.

"We are hearing from a lot of rightsholders who are very excited for this new kind of 'interactive fan fiction' and think this new kind of engagement will accrue a lot of value to them, but want the ability to specify how their characters can be used," Altman wrote. "We want to apply the same standard towards everyone, and let rightsholders decide how to proceed."

Given that this is generative AI, and no one really knows how the black box makes its decisions, Altman admitted that "there may be some edge cases of generations that get through that shouldn't."

The OpenAI CEO didn't specify exactly how these new features would work with copyrighted material - for example, if copyright holders would be able to grant limited use rights, how permission would be requested and granted, nor how rights would be enforced in the case of complicated copyright chains of ownership.

Altman also didn't mention reports that all copyright holders would be required to opt out of having Sora use their data. Then there's the question of whether Sora would use copyrighted material for training, even if rightsholders banned their material from being used in the final output.

The devil is in the details, and a lot of details have simply not been disclosed. Given these omissions, Altman's blog post feels a lot like an attempt to squeeze the words "opt-in" near "rightsholders" in the same sentence without actually connecting the two.

We reached out to OpenAI to get an explanation of whether IP holders would still have to formally opt out of having their properties used by Sora, but didn't get an answer by press time.

Please don't take that IP away: We'll pay you ... someday

A lot of Hollywood power-players are unhappy about the fact that their properties can now be - and reportedly have been - spun into AI videos doing things they'd never allow if they were in control. According to reports from the days following Sora 2's release, Disney has opted out of having its intellectual property appear in Sora 2. Talent agency WME, which represents a number of film, TV, music, and sports A-listers, has opted all of its clients out of Sora 2 as well.

In an effort to stem the tide, Altman also floated the idea of revenue sharing for IP holders who let their content be used by Sora - although, again, with very few details.

"We are going to have to somehow make money for video generation," Altman said, saying that users were creating far more videos than the AI giant expected. The company has yet to turn a profit, but once it does, IP rightsholders are going to get a slice, he said.

"The exact model will take some trial and error to figure out, but we plan to start very soon," Altman added. "Our hope is that the new kind of engagement is even more valuable than the revenue share, but of course we want both to be valuable."

Overall, the approach seems akin to YouTube's in its early years - let people post a flood of content, copyright be damned, then sort out the mechanisms for payment and control once the horse is out of the barn. But given how little profit the AI industry is actually making as a whole, that revenue sharing promise rings particularly hollow. At least when Google bought YouTube, it had its search advertising business humming along, and one could theoretically see how YouTube would slot into its ad platform and make money someday. OpenAI has no such cash cow yet. ®