You can ask it to modify a scene, and it remembers what came before.
At its 2026 developer conference, Google introduced Gemini Omni — a model that accepts text, images, audio, sketches, and video as inputs and returns edited video through ordinary conversation. The announcement marks a meaningful shift in how machines participate in storytelling: rather than demanding that creators start over with each revision, the system holds memory across instructions, preserving character, physics, and context as a collaborator might. It is an early signal that the distance between imagining a video and producing one is narrowing, with consequences for creators, industries, and the nature of authorship itself.
- The old frustration of AI video tools — generate, discard, start again — is directly challenged by a system that remembers what you asked before and builds on it.
- Gemini Omni can simultaneously absorb a sketch, a photograph, an audio clip, and existing footage, synthesizing them into a single coherent output — a capability that compresses what once required a production team.
- Google is rolling the model out in stages across the Gemini app, Google Flow, and YouTube Shorts, with image and audio generation features still to come, keeping the full picture just out of reach.
- A digital avatar feature — a video version of yourself that speaks in your own voice — quietly raises the stakes for identity, authenticity, and what it means to appear on screen.
- Access is currently gated behind Google AI Plus, Pro, and Ultra subscriptions, meaning the democratization of video creation arrives, for now, on a tiered basis.
At Google's annual developer conference in May 2026, the company introduced Gemini Omni, an AI model designed to generate and edit video through ordinary conversation. Users can feed it text, images, audio, sketches, or existing video — alone or in combination — and receive edited footage in return.
What separates it from earlier tools is memory. Previous AI video systems required users to start fresh with each revision. Gemini Omni retains the context of prior instructions, so characters remain consistent across shots, physics hold from scene to scene, and changes — swap a background, add an object, alter an action — are made without erasing what came before.
The first release, Gemini Omni Flash, is live now through the Gemini app, Google Flow, and YouTube Shorts, with image and audio generation to follow. The model draws on Gemini's existing reasoning capabilities and extends them into video by applying knowledge of physics, history, science, and cultural context — the kind of grounding that makes generated content feel plausible rather than glitched.
Google also unveiled a digital avatar feature, allowing users to create a video version of themselves that speaks in their own voice — a tool with clear appeal for YouTube creators and significant implications for questions of identity and authenticity online.
Access is currently available to Google AI Plus, Pro, and Ultra subscribers. The company frames Gemini Omni as a tool for creators and marketers who want to make video without learning video software. How well it delivers on that promise will emerge as more people begin using it.
Google has moved deeper into the business of making video with machines. At its annual developer conference in May 2026, the company introduced Gemini Omni, a new artificial intelligence model built to take whatever you feed it—text, images, audio, video, even sketches—and turn it into edited video through ordinary conversation.
The shift matters because it changes how video creation actually works. Until now, AI video tools have asked users to write a prompt, wait for output, then start over if they wanted changes. Gemini Omni works differently. You can ask it to modify a scene, and it remembers what came before. Characters stay consistent across shots. The physics hold. If you want to change the background, add an object, or shift what a character is doing, you just say so—the system understands the context of your previous instructions and builds on them rather than erasing them.
The first version, called Gemini Omni Flash, is rolling out now through the Gemini app, a product called Google Flow, and YouTube Shorts. The company plans to add support for generating images and audio in the coming months. The model draws on Gemini's existing capabilities in reasoning and generation, but extends them into video territory by applying a deeper understanding of physics and context—the kind of knowledge that helps make generated content look less like a glitch and more like something that could actually happen.
What makes the system genuinely different is its ability to work across multiple media types at once. You can hand it an existing video, a photograph, a hand-drawn sketch, and an audio file, and ask it to synthesize them into a single piece. The model uses its training in history, science, and cultural knowledge to build explainers and visual stories, not just creative content. Google also introduced the ability to create a digital avatar of yourself—a video version that speaks in your own voice, which opens obvious possibilities for creators on YouTube and elsewhere.
The rollout is happening in stages. Gemini Omni Flash is available now to subscribers of Google AI Plus, Pro, and Ultra. YouTube Shorts and YouTube Create will get access as well. The company is positioning this as a tool for creators, marketers, and anyone who wants to make video without learning video software. Whether it actually works that smoothly in practice, and whether it changes how people actually make things, will become clear as more people get their hands on it.
Citas Notables
Characters remain consistent across scenes, edits retain context from earlier prompts and videos can be modified without restarting the creative process.— Google's description of Gemini Omni's capabilities
La Conversación del Hearth Otra perspectiva de la historia
What's the actual difference between this and just running multiple prompts through an older AI video tool?
The difference is memory. With older tools, each prompt is isolated—you ask for a scene, you get a video, you start completely over if you want to change it. Gemini Omni remembers what it just made. You can say "now make the character turn left" and it understands the character already exists, already has a face, already has a history in that scene.
So it's not restarting from scratch each time?
Exactly. It's building on what it made before. That sounds small, but it changes the entire workflow. You're not fighting the tool. You're having a conversation with it.
What about the avatar feature? That seems like it could get weird fast.
It could. You're creating a digital version of yourself that can speak in your voice. For creators, that's powerful—you can make content without being on camera. For everyone else, it raises obvious questions about authenticity and impersonation. Google hasn't really addressed that yet.
Is this actually available to regular people, or just developers and early adopters?
It's rolling out to paid subscribers first—the higher tiers of Google's AI service. YouTube Shorts will get it too, which is a much bigger audience. So it's not locked behind a developer program, but it's not free either.
What's the real limitation here? What can't it do?
We don't know yet. The announcement doesn't say. But any AI video tool has limits—it probably struggles with complex physics, with very long videos, with maintaining perfect consistency across many scenes. The real test is what happens when thousands of people try to use it for actual work.