The most aggressive move in AI video this month didn't come from a video company. It came from the company that owns the plumbing.
On September 23, Google folded free AI video generation into Vids, its browser-based video editor, for every Google account holder. The engine is Gemini Omni 1.1 Flash — the same family that powers generation in YouTube Shorts and YouTube Create. Personal accounts get a monthly allowance (reports put it at six videos a month, resetting on the first), with 1080p output and SynthID watermarking baked in.
This is the third step in a sequence. In April, Google opened a limited number of free Veo 3.1 generations to account holders. In June, it expanded free access to AI avatars. Now the broadest release yet. Each step widens the same funnel: get people generating inside Google's tools, where the upsell to a paid AI plan is one click away.
The price war underneath
The consumer freebie sits on top of developer pricing cuts. Google dropped the Veo 3.1 Lite API rate to $0.05 per second for 720p and $0.08 for 1080p — undercutting developer-facing rivals while Chinese competitors like Kling and Hunyuan Video push from the other side. Model costs are falling fast across the industry, and Google is choosing to pass the decline through as market share rather than margin.
Why free wins here
Standalone video apps have a discovery problem: every new user has to decide to download something. Google doesn't. Vids sits next to Gmail, Drive, Docs, Slides, and YouTube — software people already open every day. The product advantage isn't a better model. It's that the model shows up where the work already happens.
The contrast with OpenAI is hard to miss. Within days of Google's announcement, OpenAI shut down the Sora API entirely, redirecting compute toward enterprise coding tools. One company is giving video away to acquire users; the other decided video wasn't worth the GPUs. Both moves tell you where each company thinks the money is.
The catch
Free tiers are funnels, not charities. Six videos a month at modest quality is enough to learn the workflow and not enough to do serious work — which is exactly the point. And "free" still costs Google real inference spend, a bet only companies with hyperscale infrastructure can afford to make. Everyone else has to charge.
Sources
- [1] finance.biggo.comRead source
- [2] nolowiz.comRead source