Who are AI video generators really for?

In late March, OpenAI killed Sora. The video generation app had launched six months earlier with enormous fanfare, shooting to the top of the iPhone App Store and peaking at around 3.3 million downloads.

Disney had committed a billion dollars to a licensing deal letting users generate clips featuring more than 200 characters from its catalogue. Six months later the whole thing had collapsed, and Disney found out less than an hour before the public announcement.

By February 2026, Sora's downloads had fallen 66 percent and daily active users had cratered below 500,000. Lifetime in-app purchase revenue reached just $2.1 million, against running costs of roughly a million dollars a day.

Each ten-second clip cost about $1.30 to produce. Bill Peebles, OpenAI's own head of Sora, publicly admitted the economics were completely unsustainable.

Tech products fail all the time. What makes this one interesting is that it came from the world's most recognisable AI company, backed by the world's largest entertainment conglomerate, and still couldn't last six months.

Follow these steps to redeem your free month:

Hurry, because this offer is capped at 25 claims. The code expires on 05/20. You will need to enter your card details and will be charged if you don’t cancel at the end of your free month.

I made our latest video because of my interest in visual AI. I’ve tested out a ton of video generator apps and really thought Hedra was amongst the best. Why? Because unlike Sora or Google Flow it gives you access to all the top models - not just its own (Hedra Omni is great though).

Using its platform you can generate video using models like Veo 3.1, Seedance 2.0 and Hedra’s own models, PLUS images with GPT Image 2 and Nanobanana, and generate avatars and audio. It’s an all in one creator suite.

I wouldn’t be giving this offer if I didn’t think it was a top creator tool. Using your month free is a great way to try out different video and image models yourself.

I've used Veo 3 and Midjourney a lot last year. Last year we tried to produce a prehistoric animals documentary using generative video, and it largely turned into a spoof of Veo 3’s limitations. Consistency was not ideal.

A giant dragonfly called Meganeura came out looking like an alien bat with a phallus. A Roman in lorica segmentata armour arrived with bizarre plumes and made no historical sense.

A 20 second speaking sequence recently took more than 200 generations to stitch four usable clips together (although something must be said of my lack of experience). With generative AI, can feel less like creative decision-making and more like a game of roulette. It’s really not as easy as it looks.

But the tools are improving quickly, though. Google's Veo 3 produces cinema-grade motion at 1080p with native audio and lip sync, and Veo 3.1 Lite launched in March at roughly half the previous cost. ByteDance's Seedance 2.0, embedded inside CapCut, offers director-level controls over camera, lighting and character consistency.

While consumer apps struggle, a different corner of the industry has been finding product-market fit. HeyGen and Synthesia generate corporate talking-head videos using AI avatars, and enterprise customers are paying for them.

The use case is training and internal communications. Any large company producing onboarding videos, compliance modules or executive updates in multiple languages was already spending serious money on production. Avatar tools cut that bill dramatically.

Gamma sits alongside them in a related category, turning prompts into polished presentations. The shared logic is workflow substitution. These tools replace existing corporate spend on slides, decks and training videos, so the value is measurable from day one.

That matters because it shows AI video does have a working business model. It just isn't the one OpenAI was selling. The winning products solve a more boring enterprise problem rather than chasing a consumer social feed.

The copyright wall still looms over the consumer side. Within 24 hours of Seedance 2.0's February launch, users had recreated shots from Stranger Things and Marvel films. Disney sent ByteDance a cease-and-desist, Paramount and Netflix followed, and the global launch was paused.

Every leap in quality on the consumer side triggers a matching escalation of legal and political opposition. For now, the clearest path to a sustainable AI video business runs through model curators and enterprise workflows, not through trying to be the next TikTok by topping the App Store charts.