Veo2 Has an API Now

Veo2 is on Vertex AI. You can call it from code. Today.

That's the whole thing, really. The model that was generating those uncanny slow-motion macro shots and the footage that made people argue about whether it was real — it now has an endpoint. You send a text prompt, you get video back. No waitlist page with a three-paragraph explanation of responsible AI. Just an API.

This is the part of the diffusion model arc that always feels slightly unreal when it arrives, even after you watched it happen with images and then with audio. The capability existed. It was impressive. People used the demo. And then one day there's a model parameter in a JSON payload and you're piping video frames into your application like it's nothing.

Google has been playing catch-up on the "ship it as infrastructure" part of this race — Gemini has had some rough months of being technically impressive and productively awkward to use. Veo2 on Vertex feels different. Vertex is where actual Google Cloud customers live, where the credit card is already on file, where the compliance team has already signed off on using GCP. The distribution is already there.

The interesting question isn't what you can make with a video model. It's what happens when the cost of a ten-second clip drops to somewhere between a coffee and free, and every app that currently uses a stock video API has to decide whether to rebuild.

That reckoning is now scheduled.