The Video Model Wars Are Actually Two Models Fighting

Google updated Veo this week, and it's legitimately good — the kind of good where you watch an output and feel a small, quiet dread, the kind reserved for things that are progressing faster than your ability to have feelings about them.

So now we have a real fight. Sora on one side, Veo on the other — two models that can both produce video you would almost believe, each iterating fast enough that whatever I write here will be partially obsolete by the time you read it.

Grok Imagine exists. It is, by most accounts, the worst of the three on technical merit and the best of the three on willingness to generate things that will make you regret asking. This is not a small distinction — there is a meaningful market for uncensored generation, and xAI apparently decided that market was worth owning outright. Whether that's a business strategy or a character flaw is left as an exercise.

And then there's Meta. Meta has a video model. I am not going to describe it further because there is nothing further to say. It is present. It exists in the category. This is the entirety of its achievement.

The interesting thing about Sora vs. Veo is that OpenAI got there first with the cultural moment — the Sora reveal was the kind of demo that breaks tech Twitter for a day — and Google is doing what Google does, which is build the better engineering story quietly and then show up later with receipts. This has worked for them before and not worked for them before, in roughly equal measure, and there's no particular reason to believe 2025 will resolve the pattern.

What I keep thinking about is that none of these companies have fully decided what video generation is for yet. The demos are all the same: a golden retriever running through a field at sunset, a city street in rain, something cinematic and purposeless. Nobody has made the boring business case — the one where this actually replaces a workflow someone was paying for — at scale. When that happens, the winner won't be whoever made the prettiest golden retriever.

It'll be whoever made the fastest, cheapest, least-censored golden retriever that an ad agency can bill to a client without explaining what it is.