News

Credit: BW Marketing World

You've probably seen it: a video that looks exactly like a Hollywood blockbuster, starring a face that looks exactly like a movie star, except the film doesn't exist, and the star never signed on. It's on your feed right now, somewhere between a cooking video and a meme. It's fun. It's fresh. It's a little uncanny in the best way.

Here's the story most viewers tell themselves: some clever creator built this with a few prompts, the tools are free (or cheap), and the only people who might lose out are the actors whose faces got borrowed. Everyone else (the audience, the creator, the platform) comes out ahead.

That story is comforting. It's also wrong.

Where This Content Actually Comes From

Tools like ByteDance's Seedance and Seedream can generate video and images with startling realism, including, apparently, scenes that look a lot like existing films and famous faces. In February 2026, a two-line text prompt produced a viral clip of Tom Cruise and Brad Pitt in a fight scene neither actor ever filmed. Disney and other studios raised alarms almost immediately, and the Motion Picture Association (MPA) sent ByteDance a cease-and-desist letter accusing it of ignoring copyright law "on a massive scale." By August 2026, the two sides had signed a formal agreement to build stronger copyright safeguards into those tools.

That timeline (viral clip, legal threat, negotiated framework) is becoming a pattern. And it points to something the "fun and harmless" story leaves out: these outputs aren't created from nothing. The models generating them were trained on enormous libraries of existing film footage and publicly available images of real performers. The realism you're enjoying is, in a very literal sense, built out of other people's work and likenesses.

Video: Uproar Over AI-Generated Brad Pitt And Tom Cruise Fight Scene

Copyright Violations Didn't Start With AI: Here's What They Already Cost the Industry

Before generative AI entered the picture, the film industry was already losing enormous amounts of money to a much simpler form of copyright violation: piracy. Estimates vary depending on methodology, but multiple industry studies converge on the same conclusion: the losses are staggering. The Motion Picture Association has put global piracy losses at roughly $25 billion a year, and broader analyses covering film and TV losses put the figure as high as $90-plus billion annually when secondary markets and job losses are included.

The impact isn't abstract. It shows up as:

• Smaller production budgets, because projected revenue has to account for expected losses
• Fewer jobs on set, especially for below-the-line crew whose positions get cut first when budgets tighten
• Disproportionate harm to independent films, which rely much more heavily on box office and home-video revenue than big-budget studio releases with multiple income streams

The lesson from the piracy era is one worth carrying into the AI conversation: when someone's work is used without permission, the financial damage rarely lands only on the party being copied. It ripples outward to everyone whose paycheck depends on that work generating revenue in the first place.

Now AI Is Repeating (and Accelerating) the Pattern

If piracy was about distributing copyrighted work without permission, generative AI raises a new question: what happens when a copyrighted performance, or a person's own face and voice, becomes training material for something new? A few recent flashpoints show how this is already playing out.

The Scarlett Johansson / OpenAI "Sky" controversy. In May 2024, OpenAI demoed a ChatGPT voice called "Sky" that struck many listeners, including Johansson herself, as eerily close to her voice in the film Her, a role about an AI companion. Johansson said she had already turned down OpenAI's request to license her voice; when "Sky" launched sounding similar anyway, she hired lawyers and OpenAI paused the voice within days. OpenAI maintained the voice belonged to a different actress hired before any outreach to Johansson. Legal experts noted the case exposed a real gap in the law: the U.S. right of publicity, which protects people from unauthorized commercial use of their name, voice, or likeness, wasn't built with AI voice cloning in mind.

Credit: Getty Images

The 2023 Hollywood strikes. AI protections were one of the central demands in both the Writers Guild of America and SAG-AFTRA strikes that shut down productions for months. Estimates of the total economic toll on Southern California range from roughly $3 billion to $6.5 billion, with tens of thousands of jobs affected, not just actors and writers, but caterers, drivers, and hotel and restaurant workers who depend on productions actually shooting. The resulting contracts require studios to get an actor's consent before creating a digital replica of them, a direct response to fears that AI would let studios generate performances without ever paying, or asking, the performer.

Licensing done differently: McConaughey, Caine, and ElevenLabs. Not every AI-and-likeness story is a fight. Matthew McConaughey and Michael Caine chose to license their voices to AI company ElevenLabs on their own terms. It's a useful contrast: when consent and compensation are built in from the start, the same underlying technology looks a lot less troubling. But other performers who've licensed their likeness have described a messier reality: one actor said that after selling his likeness to a platform, he found himself in AI-generated ads for products he'd never agreed to endorse, without additional pay.

Credit: CBC

So Who's Actually Losing Here?

Go back to the assumption we started with: creators benefit, and only the actors are disadvantaged. Once you look closely, that framing doesn't hold up.

Actors are affected, clearly: their face and voice can end up in content they never agreed to and aren't paid for. But the list doesn't stop there:

• Working, non-famous performers: background actors and day players who don't have the leverage or legal teams to negotiate a licensing deal at all
• Writers, whose scripts and structures are part of the training data that makes AI-generated "movie-like" content possible
• Crew and support jobs, the same ones hit hardest during the piracy era and the 2023 strikes, whose livelihoods depend on productions actually being greenlit and staffed by humans
• Smaller studios and independent creators, whose original IP can be diluted or imitated at a fraction of the cost and time it took to make
• Audiences themselves, who increasingly can't tell what's a real performance and what's synthetic, a trust problem that gets bigger every time a "fun" AI clip turns out to be built on someone's unlicensed work

The creator who prompts the video isn't necessarily the villain either. Most people sharing these clips have no idea whose work or likeness fed the model. That's arguably the more uncomfortable truth: this isn't a clean story of one bad actor exploiting a victim. It's an entire ecosystem quietly shifting costs onto people who never got a say.

Why This Isn't Just Ethical, It's Necessary

The ByteDance/Seedance case is worth coming back to here, because it previews where this is heading. After the Tom Cruise/Brad Pitt clip went viral and the MPA pushed back with a cease-and-desist letter, ByteDance didn't dig in and fight it out in court. It negotiated a framework to build copyright safeguards directly into Seedance and Seedream. The reason that matters isn't just that one company backed off. It's what backing off protects, and what happens industry-wide if nobody does. To see that, it helps to trace each sector's harm all the way through to what it eventually does to the whole ecosystem, not just to the people directly affected.

Actors. If performances can be used or trained on without fair pay or consent, the return on being a working actor drops. Top talent has less reason to keep making films instead of pursuing better-paid or better-protected work elsewhere. Fewer skilled, well-directed human performances get produced. That starves the exact raw material generative AI tools depend on: fresh, varied, well-acted footage. AI models end up training on the same recycled catalog again and again, output quality plateaus, and eventually there's nothing new for either human productions or AI tools to draw from. The pipeline stalls for everyone.

Writers and creators. The same chain applies to scripts and stories. If original writing gets reused as training material without compensation, the financial incentive to spend years developing an ambitious original story shrinks. Writers gravitate toward safer, formulaic work, or leave the industry for other kinds of writing altogether. Fewer original, high-quality scripts enter production. AI training sets increasingly recycle the same material, audiences increasingly get derivative stories, demand softens, and financing new work gets harder to justify. The cycle compounds on itself.

Below-the-line crew. When productions shrink, whether from piracy, from AI tools undercutting original content, or from tighter budgets, crew jobs are usually the first cut. Experienced editors, colorists, and technicians leave the industry for gaming, advertising, or corporate video work instead, and that skilled labor pool doesn't come back easily. Remaining productions end up with less experienced crews and less reliable output, which makes the next ambitious project harder to finance, which shrinks the industry further still.

Studios and IP holders. If an AI tool can cheaply reproduce the look and feel of an expensive original film without a license, the financial case for funding new, original, costly productions weakens. Studios retreat toward safer bets: sequels, reboots, and existing franchises, instead of funding new stories. Less new intellectual property enters the market, and both the traditional industry and the AI tools trained on it end up drawing from a smaller, older, more repetitive pool of material.

There's a term researchers use for a version of this problem: model collapse, the tendency of generative models to degrade in quality across generations when they're trained increasingly on synthetic or recycled output instead of fresh, original, human-made work. That's the mechanism underneath all four examples above. Undermine any one sector's ability to keep producing new, high-quality material, and the damage doesn't stay contained. It cuts off the supply the entire industry, AI tools included, needs to keep improving.

That shared dependency is probably the real reason ByteDance chose to negotiate a framework with the MPA instead of fighting it out. Protecting these sectors isn't only about fairness to the people in them. It's about keeping the entire supply chain, human and AI, healthy enough to keep producing anything worth watching a decade from now. That's also why this is becoming less of a "should" and more of a "must": the consent-and-compensation requirements written into the 2023 SAG-AFTRA and WGA contracts, and the protections the NO FAKES Act would create at the federal level, are converting ethical best practice into real legal exposure for any company, AI or otherwise, that doesn't build these protections in from the start.

What Comes Next

A few developments suggest the industry and lawmakers are catching up:

• The bipartisan NO FAKES Act, advanced by the Senate Judiciary Committee in June 2026, would create a federal right protecting people's voice and likeness from unauthorized AI replicas, closing exactly the kind of gap the Johansson/OpenAI case exposed.
• The ByteDance/MPA agreement signals that AI companies and studios are beginning to negotiate copyright frameworks directly, rather than waiting for lawsuits.
• SAG-AFTRA's expanded AI provisions in its 2026 agreements point to consent and transparency becoming baseline expectations, not optional extras.

None of this fully resolves the tension. But it does suggest the "harmless fun" framing has a shelf life: the legal and financial reckoning is already underway.

Before You Share That Next AI Movie Clip

Next time one of these videos shows up on your feed, it's worth pausing on where the realism actually came from. Whose performance made this possible? Who consented, and who didn't? Do you think platforms and studios are moving fast enough to protect performers, or is legislation like the NO FAKES Act coming too late?

Want to explore AI-powered production without the copyright headaches? Contact Storyverse to learn more about our AI production services.

Sources

1. NBC News: ByteDance signs AI copyright pact with Hollywood trade group
2. PetaPixel: ByteDance and Hollywood Reach Agreement Over AI Video IP Rights
3. Variety (via report): Scarlett Johansson statement on OpenAI's "Sky" voice, covered by NBC News
4. Georgetown University: OpenAI v. Scarlett Johansson? Law Professor Answers Legal Questions on AI-Generated Content
5. CBC: Matthew McConaughey, Michael Caine license likenesses as generative AI debate deepens
6. Fordham Journal of Corporate and Financial Law: WGA Strike Resolution: AI Restrictions and Implications
7. Gecić Law: The 2023 Hollywood Strike: Labor, Scripts and the AI Showdown
8. UCLA Anderson Forecast: Hollywood Restructuring and the Economic Impact of the Writers and Actors Strikes
9. Reuters (via AOL): Industry video game actors pass agreement with studios for AI security
10. Kaufman & Canoles: The NO FAKES Act Moves Forward: How AI Voice and Likeness Rights Could Reshape Entertainment Deals
11. Indian Journal of Law and Legal Research (ResearchGate): Internet Piracy In The Film Industry: An Economic Analysis
12. Bytescare: How Much Does Piracy Cost The Film Industry? - 6 Effects