Two Roads for AI Video: Google Bet on A24's Taste, ByteDance Bet on Replacing It

The A24 road: arm the artist, keep the judgment human

What is striking about the Google deal is what it is not. It does not give Google A24's film library, its data, or the right to train on its catalog. It is a research partnership in which DeepMind builds tools alongside A24's filmmakers, who keep full creative control. The studio's tech lead, Scott Belsky, drew the line clearly, saying the tools will not resemble the prompt-and-generate AI that makes people uneasy, and that the better uses are the ones that preserve creative control and support risk-taking rather than just making films cheaper and faster.

This matters because A24 is the least likely studio in the world to sell out craft. Its whole brand is taste, auteurs, and films that should not work on paper. Even its embrace of AI is conditional, aimed at the unglamorous parts of production, the planning and the workflow, while the human stays in charge of every choice that an audience actually feels. AI cleaning up an accent in post, or roughing out a storyboard, is a tool. It is not the thing audiences come for.

It is not a frictionless move. A24's audience skews young, and younger viewers are among the most wary of AI in creative work. The director of the studio's biggest recent hit has publicly called generative AI a kind of creative rot. A24's defense, that it would rather have a seat at the table than watch the technology get built without it, is the sound of a company trying to take the infrastructure without ceding the judgment that makes it what it is. That tension is the whole point. The bet is that taste and human authorship are not obstacles to clear out of the way. They are the product.

The Seedance road: generate it, and sort out the rest later

Now hold that against the other headline. Seedance 2.5 produces a continuous 30-second clip in native 4K from a single prompt, with synchronized audio and up to 50 reference inputs, and ByteDance's enterprise video business reportedly already runs at $2 billion a year. One analysis pointed out that this is the kind of output brands once paid between $50,000 and $500,000 to produce, now compressed toward zero.

That is a real capability, and for the right jobs it is a gift. But the Seedance road comes with a cost the price tag hides. Its earlier version drew cease-and-desist letters from Disney, Warner Bros, Paramount, and Netflix, plus a viral celebrity deepfake that prompted a complaint from the Motion Picture Association and a rebuke from the actors' union. ByteDance had to bolt on watermarking and copyright filters and, alongside the new model, announce an IP-licensing platform. This is the generation-first road in miniature. Astonishing output, arriving with a question attached about where it came from and whether anyone can stand behind it.

Why the fork matters for your brand

Strip away Hollywood and the same fork sits in front of every founder making a video. You can generate something, fast and cheap, or you can make something, with a human deciding what it means. The industry's most credible craft brand and one of the world's largest AI labs just agreed, with $75 million on the line, that the future worth building is the one where the tool serves the artist rather than replacing the judgment.

For a startup, the translation is direct. Use generation tools for what they are good at, the volume work, the social cut-downs, the quick tests. But the film that has to make a stranger trust your company is the A24 problem, not the Seedance one. It depends on a real founder on camera, a real decision about what to show and what to cut, a story that someone with taste shaped on purpose. That is the part no model generates, and it is the part that earns belief. In a feed about to flood with near-free 4K clips, the human-made film does not get less valuable. It gets rarer, and rare is where trust goes.

How we think about it

This is the philosophy our studio already runs on. We use modern tools where they genuinely help, in planning, in post, in the unglamorous middle of production. What stays human is everything an audience actually responds to: the direction, the story, the founder in the frame, the judgment about what matters. AI at the table, not at the wheel.

The results follow from that line. For Speakology AI, a brand film made during their raise lifted their investor reply rate by 86 percent and helped close the round at 120 percent above the ask. For Flowy AI, a launch built around a real human moment drew more than 350,000 views and helped bring over 45,000 demos and customers onto the platform with no paid spend. Fuse AI used their film in a successful $3 million seed raise, and the brand film we made for the Horological Society of New York premiered at a gala that raised $1.2 million. None of those came from a prompt. They came from people making deliberate choices, which is the same bet A24 just made in front of the whole industry.

If you are deciding how to make the video that has to earn trust, we are based in San Francisco and Los Angeles. Get in touch.

Sources

1. Reporting on Google DeepMind's roughly $75 million investment in A24, announced June 22, 2026: the research partnership structure, exclusion of A24's library and data, filmmaker creative control, and remarks from A24's Scott Belsky and DeepMind's Eli Collins and Demis Hassabis (Variety, Deadline, IndieWire, TechRepublic, LAmag, Memeburn).

2. Coverage of fan backlash and A24's "seat at the table" defense, including director Kane Parsons's criticism of generative AI (Variety, Deadline, ScreenDaily, AV Club).

3. Reporting on ByteDance's Seedance 2.5 (30-second native 4K, 50 reference inputs, $2B platform ARR) and the copyright and likeness disputes around its predecessor (The Next Web, TechTimes, FourWeekMBA).

Horizon Studios is a brand film and cinematic video studio for technology companies, based in San Francisco and Los Angeles. Recent work includes Speakology AI, Flowy AI, Fuse AI, and HSNY. See the full portfolio at horizonstudios.us/projects.