A new entrant has just swaggered into the always roiling AI video fray: lo and behold, a startup named CraftStory started by the brains who created OpenCV has unveiled a system it claims can churn out five minute long videos – which should put even more of its flashy AI competitor’s to shame.
This leap is fuelled by a “parallelised diffusion architecture” – reported in their profile in VentureBeat, that allows the model to generate whole sequences at once, rather than one piece at a time.
That’s brazen, big and actually kind of thrilling. “What used to take camera crews, tweakers and weeks of postproduction,” he said in an email, “could soon consist of a still image paired with a driving video clip with motion data points – and then pay your hourglass ensemble system to animate, lip-sync and gesture.”
CraftStory says this strikes the right balance for corporate training videos and product demos, rather than just stupid short clips for social networks.
Why does this matter? Because while the giants – OpenAI (which recently released Sora) and Google (with Veo, according to reports) – race toward the video frontier just as fast as they can, CraftStory’s saying: we’ll take a deep lap in one lane (long-form, human-centered), not try to be everything.
The company raised a smallish ~US$2 million seed round, yet lean when compared with the billions floating through larger labs. But sometimes lean means nimble.
If I’m honest, I had two thoughts when watching this. One: finally someone tackles the “long-form video” problem – when a 10-second clip won’t work, and you need minutes of sustained ‘coherence’, consistent character performance, movement, context – this model can handle that.
Two: gee whiz, what does this mean for traditional production workflows when this is real? Studios, editors, motion capture pros – they’ll need to pivot or innovate quickly.
There are some caveats though. CraftStory’s demo hints at applications able to be made in B2B (training videos, product explanations) as opposed to consumer facing content just yet.
And high-quality video still requires good “driving” footage, actor motion capture and the like – so it’s not yet “text prompt → blockbuster video” for anyone with a laptop (though such functionality might one day arrive).
Plus, as the Spanish startup news outlet Ecosistema Startup notes, there is already a crowded arena of competitors in that model includes Runway, Stability and Pika or are doing so.
Another interesting storyline: the fact that the people who work on OpenCV – they’ve been entrenched in computer vision and motion understanding for a really long time – are now taking all of that expertise into generative video.
Victor Erukhimov, CraftStory’s founder, stresses that it isn’t just about nicer visuals, but also knowing how humans move, understanding how facial dynamics roll out over time and creating video that looks natural.
What does that mean for creators, or even marketers, educators? “If you are working in the content department, then you may want to ask and wonder: How fast can I resize a set of tools that do more of the heavy lifting?” Where does the human creativity still dazzle?
What processes should I hang on to so I don’t get caught unprepared when automation arrives?
My bet, or a hope rather: the value going to move toward ideation, brand voice, human directing and making sure you understand why you’re creating a video-not just how.
In other words: At a time when we’re in the midst of another startup chasm (raise some cash, produce a model and then launch), CraftStory’s launch is more than that.
It’s a harbinger of where AI video generation is going: longer, more coherent, more human-focused and more enterprise-grade.
Whether or not CraftStory ends up being the name in long-form video, rather than simply another bid to become so, it is at least a contender worth consideration.