Google launches Gemini Omni for conversational video creation

Background

The Gemini app is the consumer side of Google's Gemini line: a chat-style interface where people ask for text, images and other output in ordinary language. Making video with AI has usually meant opening a separate tool built around its own controls and settings. Moving that work into a conversation changes who can do it and how the work proceeds — the user describes an idea, looks at the result, and says what to change.

What launched

On 19 May 2026, Google added Gemini Omni to the Gemini app, letting users create and edit videos through conversational prompts.^[1]

Omni combined text, photos and video as inputs, and it could build AI avatars.^[1] A user could start from scratch, from material already in the camera roll, or from preset templates.^[1] In practice that means one conversation can hold a written description, a still photo and an existing clip at the same time, with the model asked to blend or extend them; editing runs the same way, with the user saying what should change instead of operating a timeline by hand.

Access

Omni was available to Google AI subscribers aged 18 and older worldwide, with features varying by region.^[1] The age floor and the subscription requirement meant it launched as a paid feature rather than something open to every user.

Developers got their own route in on 30 June 2026, when gemini-omni-flash-preview appeared in the Gemini API.^[3] A preview label usually means a model is offered for testing and may change before wider release, and the flash name places it among Google's faster, lighter variants.

Where the output can be made, and how it is marked

Content created or edited with Omni can be made in the Gemini app, Google Flow or YouTube.^[2] The same capability therefore reached users in three places rather than one.

What comes out carries two provenance signals: SynthID watermarks and C2PA Content Credentials.^[2] SynthID is Google DeepMind's watermarking method, which embeds a pattern into generated media that a detector can later read to judge whether it came from an AI model. C2PA Content Credentials are a metadata record, following an industry standard, of how a piece of media was produced and edited. Neither stops anyone from making a synthetic video; they make it likelier that viewers and platforms can find out where it came from.

Why it matters

Two things stand out. First, the interface: video generation moved from a specialised console into a chat window, where the prompt is a sentence and the revision is another sentence, which lowers the amount of tooling a person has to learn before making a clip. Second, the route to users: a subscriber feature in the app in May, developer access through an API a little over a month later, with the same provenance marking attached in both cases. As synthetic video becomes easier to produce, the watermark and credential machinery is what lets a result be checked rather than merely believed.