UPDATE: Gemini Omni Launches at I/O — Now Live for Subscribers

Google officially launched Gemini Omni at I/O 2026 on Tuesday, confirming weeks of leaks with a keynote demo from DeepMind CEO Demis Hassabis. The new video-generation model rolled out the same day to Google AI Plus, Pro, and Ultra subscribers, marking Google's most aggressive push yet into generative video.

This is the story Krasa.ai previewed when UI strings leaked on May 16. The official launch confirms most of those rumors and adds important new details about pricing, availability, and what Google chose to hold back.

From leaked screenshots to a working product

The leak that surfaced last weekend hinted at a model that could combine Gemini's reasoning with new video capabilities. Today's keynote went further. Hassabis described Gemini Omni as a fusion of three existing DeepMind systems: Veo (video generation), Nano Banana (image editing), and Genie (the world-model research project that powers interactive 3D environments).

The pitch is straightforward. Most generative video tools today are one-shot — you type a prompt, you get a clip, and you start over if you want to change anything. Gemini Omni is designed for conversational editing. You can ask it to change the lighting in a scene, swap a character's outfit, or add a new element to a background, all using natural language while the model preserves continuity from your earlier edits.

What was actually demoed

On the I/O stage, Google showed Omni generating a claymation-style explainer video about protein folding. The model produced consistent characters, smooth camera movement, and accurate text overlays — three things current video models routinely flub.

A second demo edited a user's selfie video by adding a new background environment and inserting visual elements on command. The presenter spoke the edits aloud: change the sky, add snowfall, place the subject on a mountain ridge. Each change rendered in seconds, and earlier edits stayed intact.

The first variant rolling out is called Gemini Omni Flash — the faster, cheaper sibling to the full-fat model still in testing. Hassabis confirmed the full Gemini Omni will arrive later this summer.

Why this matters for the AI video race

Generative video is the most competitive corner of the AI market right now. OpenAI's Sora, Runway's Gen-4, Kuaishou's Kling, and ByteDance's Seedance 2 have all shipped major updates in the last six months. Google was conspicuously quiet — until today.

Omni's combination with Genie is the strategic kill shot. Genie creates playable, controllable 3D worlds from a single image. Bolting that capability onto a consumer-facing video model means Gemini Omni isn't just generating clips — it's generating environments you can move through. That edges Google closer to the "world model" thesis Hassabis has been advocating publicly for over a year.

For enterprise customers, the immediate impact is on marketing teams, e-learning producers, and product video studios. The conversational editing workflow could cut iteration time on a 30-second ad from days to hours.

What Google held back

TechTimes reported that Google deliberately omitted Omni's most controversial capability: photorealistic human generation with voice and lip-sync. The company has been spooked by deepfake controversies that hit competitors earlier this year, and is rolling out face-and-voice generation only through restricted enterprise pilots with watermarking enforced.

That's a measurable retreat from the leaks, which had suggested Google was preparing to ship a full talking-head feature at I/O. The cautious posture mirrors recent statements from Google's Trust & Safety lead, who said in April that the company would "ship slower than competitors on synthetic humans, on purpose."

Expert reactions

Researchers and creators on X were quick to call out Omni's editing demo as the most important shift. Analyst Uttam, writing on Medium, argued that "conversational editing is the unlock — every other video model still treats each generation as disposable. Omni treats them as drafts."

Wave Speed AI's blog noted that the Omni branding is significant: "Google is signaling that this isn't a niche tool. It's the umbrella for all of Gemini's media work going forward."

How to access it

Gemini Omni Flash is live today in the Gemini app and in Flow, Google's creative tool, for all Plus, Pro, and Ultra subscribers worldwide. There's no separate paywall — if your subscription includes Gemini Advanced features, Omni Flash is in your model picker.

The full Gemini Omni is slated for "later this summer" and will be Ultra-tier only at launch. Google has not confirmed whether developer API access will arrive at the same time.

Bottom line

The leaks turned out to be mostly right. Gemini Omni Flash is shipping today and Google is moving fast on conversational video editing — but pulling its punches on synthetic humans. For creators, the immediate value is in iteration speed: stop regenerating clips from scratch and start editing them like documents. The bigger question, which Google didn't fully answer today, is whether Omni's world-model heritage from Genie will eventually let users build full interactive scenes, not just videos. If it does, this won't just be a Sora competitor. It'll be the first real product hint of where Google thinks the next decade of AI media is going.