Gemini Omni is an AI model from Google that makes and edits videos. You give it any mix of text, photos, video clips and a voice recording, and it turns them into one clip, which you then change by typing instructions such as "change the background". Google announced it on 19 May 2026 and made it available that day to subscribers of its paid AI plans.
What could Omni do in May 2026?
Google presented Omni at its I/O developer conference, together with agents such as Gemini Spark (see Google I/O 2026: Gemini Omni and always-on agents). It called Omni a model that can "create anything from any input", starting with video (Introducing Gemini Omni, 19 May 2026). The first model in the family is Gemini Omni Flash. According to Google:
- The input can be any combination of text, images, video and audio. For audio, only voice was supported at launch.
- The output is video. Google plans other outputs, such as images and audio, and says Omni will eventually be able to generate any output from any input.
- You edit by conversation. Each instruction builds on the previous one, and characters, physics and the scene stay consistent. In Google's example, a violinist is moved into the setting of a reference image, the violin is made invisible and the camera moves over the player's shoulder.
- Omni has a better grasp of gravity, kinetic energy and fluid dynamics, and draws on Gemini's knowledge, for example to make short explainer videos.
- With Avatars you make videos with a digital version of yourself that looks and sounds like you. Changing the audio and speech of other videos was still being tested.
- Every Omni video carries SynthID, Google's invisible watermark, which you can check in the Gemini app, Gemini in Chrome and Google Search.
How does it work?
Google does not describe the architecture. Its list of I/O announcements only says that Omni combines Gemini's intelligence with Google's generative media models. Omni is a form of multimodal AI: Google says it built Gemini to be multimodal from the start, able to take in several kinds of input at once, and Omni adds the ability to create video from such a mix.
Who could use it?
On 19 May 2026 Gemini Omni Flash rolled out worldwide to all subscribers of Google AI Plus, Pro and Ultra, in the Gemini app and in Google Flow, Google's AI studio for video and images. It was also free for users aged 18 and over in YouTube Shorts Remix and the YouTube Create app. Developers and enterprise customers were promised access through APIs in the following weeks.
What are the limits?
- At launch Omni made only video and accepted only voice as audio input.
- The claims about physics and consistency come from Google's own announcement, which gives no measurements.
- A model that makes realistic video of people can also be used for fake videos. Google started the avatar feature with your own likeness and marks every video with SynthID.
Example
Suppose a bicycle shop wants a short video of a new e-bike for its website. The owner uploads three product photos to the Gemini app and asks for a clip of the bike on a forest path. Then they adjust it with instructions such as "make it evening" or "lower the camera". The video carries a SynthID watermark, so anyone can check in Google's tools that it was made with AI.