Google's 'Gemini Omni' Video Model Leaks Days Before I/O 2026

A new Google video model called Omni surfaced this week ahead of Google I/O 2026, after an X user spotted a UI string inside the Gemini app reading "Start with an idea or try a template. Powered by Omni." Early access demos from a Reddit user showed the model generating short video clips with chat-based editing and templated workflows — a clear signal that Google is preparing to launch a new video generation product at I/O on May 19-20.

The leak adds Omni to a long list of AI video tools competing for the same market, and it positions Google to put real distance between its consumer Gemini app and standalone video generators like OpenAI's Sora, Runway's Gen-4, and ByteDance's Seedance.

What the Leak Actually Shows

The original tip came from a UI string surfaced on X on May 2 by a user reverse-engineering the Gemini app. The string sits in Gemini's video generation tab, replacing what had been a Veo-3.1-powered flow. A Reddit user subsequently reported gaining early access to the same model inside the Gemini app and shared clip examples.

The demos suggest three capabilities that distinguish Omni from current Veo-driven video in Gemini. First, conversational editing — you can ask Omni to re-shoot a scene, change camera angles, or adjust lighting using natural language, similar to how ChatGPT can iteratively edit an image. Second, templates — pre-built prompt scaffolds for common use cases like product demos, ad creative, and short-form social clips. Third, possibly a unified image-and-video model that handles both modalities through one architecture.

The third capability is the most speculative. The Reddit demos don't conclusively show image generation, but the "Omni" naming and the way Google has been talking about multimodal models since Gemini 2 both point toward unification.

What Omni Probably Is

There are three plausible interpretations, in order of decreasing likelihood.

The first is that Omni is a new Gemini-trained video model that runs alongside Veo, with Veo continuing to power Vertex AI and the API while Omni powers consumer Gemini. This would match Google's recent pattern of differentiating consumer and developer products with different underlying models.

The second is that Omni is just a marketing rebrand of the same Veo-powered pathway, repackaged for the Gemini app's product surface. The UI changes alone wouldn't rule this out, but the Reddit demos showed capabilities — particularly the chat-based editing — that don't currently exist in Gemini's Veo flow. That cuts against the rebrand interpretation.

The third is that Omni is a unified Gemini omni-model that handles image and video generation in a single architecture, similar to what Google has hinted at with future Gemini releases. This would be the most technically ambitious option and the most strategically significant if true.

Early Reddit demos suggest Omni is strong at chat-driven editing, somewhat behind ByteDance's Seedance 2.0 on raw video fidelity, and burns roughly 43% of an AI Pro daily quota per clip — meaning serious cost on Google's end and serious credit consumption on the user's.

Why This Matters

Video generation has gotten weirdly competitive. OpenAI shipped Sora 2 in late 2025 and integrated it into ChatGPT. Runway just raised at a $5.3 billion valuation explicitly to pivot from video tooling into "world models." ByteDance's Seedance has been winning on raw clip quality. Meta launched Muse Spark in April. The category has too many players and too much money chasing roughly the same product.

Google's advantage is distribution. Gemini is on every Android phone, every Workspace account, and increasingly in every Google product. If Omni ships with reasonable quality and the conversational editing actually works, Google can put AI video generation in front of more users instantly than any standalone product can reach. That's the strategic reason this leak is more interesting than another Sora or Runway demo would be.

The competitive math also matters. Sora and Runway are products users pay for explicitly. Omni would be a feature inside a Gemini subscription people already have, or a free tier with limits. Bundling shifts the economics of the entire category if Google executes.

What's Coming at I/O

Google I/O is on May 19-20 — three to four days from the leak — and the company has confirmed Gemini and AI updates are central to this year's keynote. The timing of the UI string appearing now, with Reddit access already trickling out, suggests Google wants Omni to be an I/O announcement rather than a quiet rollout.

Expect a few things from the keynote. A public name and pricing tier for Omni, with availability inside the Gemini app and probably the Vertex AI API. A direct comparison or benchmark slide against Sora and other competitors — Google has done this consistently with recent model launches. Demo clips that show off the conversational editing in real time. And likely some integration with YouTube Shorts or Google Photos, since putting AI-generated video into existing video products has been a recurring Google theme.

What's less clear is whether Omni is the full model or a smaller preview. Google has been shipping Gemini in tiers — Flash, Pro, Ultra — and a video model might follow the same pattern, with a faster cheaper version available widely and a higher-quality version gated to AI Pro or Ultra subscribers.

Industry Reaction

Reaction in the AI video community has been guardedly optimistic. Several creators on X have noted that conversational editing is the feature they've been waiting for from Google specifically — the ability to iterate on a clip the way you can iterate on an image with ChatGPT has been missing from every major video model. Sora has some of this, Runway has some of this, but none of them do it as smoothly as image generation does.

The skepticism centers on quality. Seedance 2.0 has been pulling ahead on raw video fidelity, and Reddit demos suggest Omni isn't matching that yet. Whether Google ships an Omni that's good enough to win on integration despite a quality gap will be the early test.

There's also a recurring concern among developers that Google ships consumer features through the Gemini app first and only later through the Vertex AI API. If Omni follows that pattern, developers building on Google's video stack will be waiting for an API while Sora and Runway are already shipping product.

What's Next

I/O on May 19-20 is the obvious watch. Beyond the keynote, expect Google to roll Omni access out in phases — first to AI Pro subscribers in the US, then internationally, then to API users. The pacing will tell you how confident Google is in the model's quality and how much it wants to control the rollout.

The other thing to watch is OpenAI's response. Sora 2 hasn't gotten the consumer integration that Gemini provides, and OpenAI has been visibly thinking about how to compete. Don't be surprised if Sam Altman tweets something coyly competitive in the days after I/O.

The Bottom Line

The Omni leak suggests Google is about to make AI video generation a default feature of Gemini, not a separate product. If the conversational editing works as the demos suggest, that's a meaningful step beyond what Sora and Runway currently offer. Quality remains the open question — Seedance and others still have an edge on raw fidelity — but Google's distribution advantage means Omni doesn't have to win on quality to win on usage. Tune in to I/O on May 19 to see whether the leak's promises hold up.