AI Video Hit an Unbroken 30 Seconds. The Tech Passed. The Law Didn't.

Last week, ByteDance's AI video model Seedance 2.5 entered public release with a trick no competitor has pulled off: generating a single continuous 30-second clip, with no seams and no stitching.

Here's why that's a threshold. Until now, AI video models generated only about 8 to 15 seconds in one pass (Runway, Google's Veo, and OpenAI's Sora all sat in that range). To go longer, you stitched short clips end to end. The cost of stitching (splicing several short videos into one longer piece) is easy to see: the joints jump, a character's face quietly changes between segments, the lighting flickers, and the whole thing reads as machine-made. Seedance 2.5 claims to shoot the entire stretch in one breath, holding faces, lighting, and camera motion steady from start to finish, at up to 4K.

Thirty seconds doesn't sound like much, but it happens to be the length of a TV commercial. Under 15 seconds you can't finish a story or hold a face; past 30, AI video reaches the "usable" bar for the first time. For years, "AI video isn't ready yet" rested on one obvious tell. That tell just disappeared.

But the same piece of news has a second half, and the second half is the part written for people running a business.

Seedance's previous version got into trouble precisely because it was too convincing. When it launched in China this February, a fabricated clip of Tom Cruise and Brad Pitt fighting racked up millions of views within 24 hours. Days later, Disney, Warner Bros., Paramount, Sony, Netflix, and Universal all sent cease-and-desist letters, and the Motion Picture Association — representing every major Hollywood studio — issued its first such notice to a major AI firm, in blunt terms: this kind of counterfeiting is the product's selling point itself. ByteDance voluntarily paused the global rollout and added watermarking and filters to block real faces and copyrighted characters. As of today, none of those legal disputes are resolved.

Put the two halves together and you see the thing that actually matters if you run a business: once the visual-quality barrier falls, the questions a buyer should ask change entirely. The more real the model looks, the less "does it look real?" means, because the answer is now "yes." The questions worth asking, and the ones a demo reel will never answer, are three others: can I use this clip commercially, and who owns it; if it happens to resemble a real person or an IP, who pays when something goes wrong; and the brand assets, scripts, and customer footage I feed in, where do they end up stored, and who controls them?

That last one is the easiest to miss. Seedance is built by ByteDance, a company that sits within reach of China's National Intelligence Law. Which means the moment you click "sign up," you've actually made two decisions at once: one about picture quality, and one about where your data goes and who can touch it. Both got bundled into the same button, but they're two different things.

I build things with AI every day, and the longer I do it, the more one instinct hardens: the stronger a capability is, the bigger the mess it can make on your behalf. A model that can conjure any human face from a single sentence is powerful and dangerous in the same stroke. So when I look at an AI video tool, I stopped caring long ago about how gorgeous the demo looks. The demo is the one face the company works hardest to show you, which makes it the least important part of the decision.

An unbroken 30-second shot is a genuine feat, and I'll use models like this. But the moment a clip goes out with a brand's name attached, those three boring questions — can I use it, who's liable, where's the data — are the entire decision. The best-looking model can quietly be the worst business choice.