Google Unveils Omni Flash — AI That Edits Films Based on Your Description
On May 19, 2026, at the Google I/O conference in Mountain View, an event took place that promises to change the rules of the game in the video content market. The company introduced Gemini Omni Flash — the first public model from the new Omni family, capable of generating video from text, images, and audio, and then editing it in a conversational mode. A technology that until recently was accessible only to professionals with expensive software is now coming to the Gemini app and YouTube Shorts.What Is Gemini Omni
Gemini Omni is a new family of multimodal models that Google described as a step toward a “world model” — an AI that understands physics, history, and cultural context. Unlike the previous Veo model, which simply turned text into video, Omni is capable of mixing any input formats: you can give it a photograph, add a voice description, and ask it to generate a clip based on those materials.
The key difference is so-called “reasoning.” The model does not simply stitch frames together; it analyzes what should be happening in a scene from a physics perspective — how light falls, how objects move, how a character’s facial expressions change. During the I/O demonstration, DeepMind CTO Koray Kavukcuoglu showed how a simple prompt, “claymation explaining protein folding,” produces a finished clip with a voiceover in a stop-motion style.Editing via Chat
Perhaps the most impressive feature is conversational editing. A user can write something like, “Now remove that car from the background and make it a sunset” — and Omni Flash will redo the video, preserving the plot and characters. According to Google DeepMind Head of Product Nicole Brihiotova, the model understands the context of the scene and “remembers” what happened in the previous frame, solving the problem of visual coherence — the headache of early generative models.
Google separately emphasized the solution to the problem of “identity preservation” of a character — another historic bug of AI video. Omni Flash allows you to upload your photo and create a digital avatar that will then appear in different scenes without changing the face or voice. However, to avoid misuse, creating such an avatar requires a separate verification — filming yourself on camera and reading out numbers.What Is Available Right Now
At launch, Google is rolling out Omni Flash specifically — a lightweight and fast version. It generates clips with a duration of 10 seconds with synchronized audio. The company claims this is not a technical limitation but a deliberate decision to give as many users as possible access to the technology. A more advanced version, Omni Pro, for professional use in advertising and video production has been announced, but without a specific release date.
Access to Omni Flash opened on May 19 for Google AI Plus, Pro, and Ultra subscribers. Separately — and this is an important strategic move — the model is available for free through YouTube Shorts and YouTube Create. Essentially, any blogger with a phone in their pocket will be able to create complex visual effects without editing.
For commercial users and developers, Google promises to open the API in the coming weeks.Watermarks and the Question of Safety
An invisible digital SynthID watermark is embedded into every generated video, which can be verified through the Gemini app, Chrome, or Google Search. For now, the ability to alter other people’s voices or create full-fledged deepfakes in Omni Flash is deliberately limited — Google is clearly being cautious, mindful of the storm of criticism aimed at its competitors.
It is worth noting that Omni Flash is not Google’s only announcement in recent days. In parallel, the company introduced Gemini 3.5 Flash for search and chatbots, as well as the Spark agent, capable of independently executing chains of tasks. But Omni looks like the most tangible embodiment of generative AI coming into everyday life — not as an expensive toy for the few, but as a tool accessible to everyone.