Google Vids now gives you three free 15-second AI videos a month with Omni 1.2 Flash

What's going on

On November 5, 2026, Google unveiled its new video model Gemini Omni 1.2 Flash and is building it into Google Vids. Vids is Google's browser-based video app, a kind of presentation tool for video, with projects saved in Google Drive. The biggest news for regular users: if you have a personal Google account or an eligible Workspace plan, Google says you can generate up to three AI videos a month, each up to 15 seconds long, at no extra cost.

The rollout is gradual. Google expects the feature to show up in supported Workspace accounts within one to three days. On desktop, you get started at vids.new. Keep in mind: three videos is a hard cap, not an unlimited free offer. Exactly how much you get can vary by account.

What Omni 1.2 Flash does better

According to Google, the new model improves in three main areas:

• Longer clips: You can set the length anywhere from 3 to 15 seconds, in one-second steps according to TechRepublic. Its predecessor Omni 1.1 was capped at 10 seconds. Tech site The Tech Buzz instead claims the old limit was 5 seconds and calls it a tripling. That contradicts Mixed and TechRepublic, which both say 10 seconds.
• More consistency: If you provide reference images, such as pictures of a character or product, the model should keep people, packaging and brand colors consistent more reliably. Mixed explicitly notes that this is Google's claim and wasn't tested independently.
• Better speech: Spoken passages should sound more varied. TechRepublic also mentions more natural dialogue and better lip sync, where mouth movements match the spoken words.

Also new, according to TechRepublic, is multi-turn conversational editing. You can use follow-up instructions to change the background, lighting or framing without starting over. Google itself admits weaknesses: fast-moving sports scenes, rapid scene changes in the script and complex physics still give the model trouble.

Hands-on: fast, but doesn't always listen

A writer on the platform note.com tried the feature on November 6 with a personal Japanese Google account in Chrome on a PC. The process: open vids.new, choose "Create AI video" and describe the video in the prompt box. The settings defaulted to the "Omni" model, 720p, landscape and 10 seconds. The interface didn't show the "1.2 Flash" version number. The 10-second clip was ready in about a minute and a half.

His test case was the children's corner of a public library: wooden shelves of picture books, small chairs in a semicircle, soft afternoon light, a slow camera pan to the right, no people, no text, no logos. The model nailed the location, lighting, camera movement and the absence of people. But it ignored two instructions: lots of text-like patterns appeared on the book spines and covers, and the chairs weren't arranged in a semicircle but around a round table. The tester suspects that, to the model, picture books simply come with titles on the cover, and that small details easily get lost in a long description. He stresses, though, that this is a single result.

His advice is a fixed five-part structure: location, what should be visible (two things at most), lighting and mood, exactly one camera movement, and finally what shouldn't appear. He also warns about a trap: a finished clip is only saved once you click "Insert". Close the tab before that, and the video is gone. Google doesn't say whether a lost or repeated attempt counts against your three-video quota. The upscale button was grayed out in the test.

Gemini app, API and pricing

Access works differently depending on the service:

According to TechRepublic, the model is generally available through Google Cloud and the Gemini API. The price of $0.07 per second is 30 percent lower than before. You can generate drafts at low resolution and then upscale the good clips to 1080p or 4K. For businesses there's Provisioned Throughput, meaning paid, reserved compute capacity. Generating videos directly in Gemini's business chat is slated to come later.

Watermarks and safeguards

According to TechRepublic, all generated videos carry an invisible SynthID watermark. That's Google's technology for letting machines identify AI content as such later on. Google DeepMind's Model Card also reveals that the model can, in principle, alter what people say in existing videos. For now, though, Google is restricting that capability. Using audio as a reference is only available to allowlisted users. TechRepublic advises companies to set up fixed review processes and labeling rules before publishing anything.

The bigger picture

Google is betting on offering AI video where people already work: in Workspace, next to Docs and Slides. The Tech Buzz sees it as a shot across the bow of paid generators like Runway and Pika, and as pressure on Microsoft, which sells its Copilot features for $30 per user per month. That said, The Tech Buzz overstates things in several places. It calls the offer "completely free" without mentioning the three-video cap, and it doesn't back up its figure of "more than 3 billion Workspace users" either.

For individuals, this is mostly a chance to experiment. Three short clips a month are enough for a club announcement or a quick social media video, but not for regular production. The API price cut for developers and businesses is the more interesting part. But as TechRepublic rightly points out, what really drives cost isn't the per-second price but the number of attempts it takes to get something usable. And the note.com test shows that some instructions simply get ignored. Google hasn't published independent comparisons with rivals like OpenAI's Sora so far.