Google Omni: The Leaked Video Model That Could Change Everything at I/O 2026

A Gemini user generated a video of a professor writing correct math equations. Then two clips used 86% of their daily limit. Here is what we know about Google's unified video model ahead of I/O.

• Google I/O 2026 starts tomorrow, May 19. But the biggest announcement may have already leaked. Over the past two weeks, users have spotted a new model called Gemini Omni inside the Gemini app, generated videos with it, and posted the results online. If this model ships as described, it would be the first top tier AI system to handle text, images, and video generation in a single unified model.
• Here is everything we know, what the early tests show, and what it means for the AI video landscape.

What Leaked and When

On May 2, an X user spotted a UI string inside Gemini's video generation tab that read: Start with an idea or try a template. Powered by Omni. TestingCatalog, a reliable tracker of Google AI leaks, picked it up immediately.

Then on May 11, a Reddit user gained actual access to the model. They got a popup saying Create with Gemini Omni and were able to generate videos. The first test was a professor writing trigonometric identities on a chalkboard. The result was remarkable: the math was correct, the steps followed a logical order, and the handwriting looked natural.

This matters because current top tier video models like Sora, Kling, and Seedance have all struggled to generate accurate text within videos. Creating video that simultaneously maintains visual coherence and semantic accuracy in written content is one of the hardest problems in AI video generation. Omni appears to have cracked it.

The second test was less impressive. A dinner scene with two men at a seaside restaurant showed spaghetti appearing out of thin air on empty plates. Typical AI video artifacts, but the overall motion quality was still strong.

Source: Verified from Android Authority, TestingCatalog, 9to5Google, and Reddit user Zacatac_391, May 2026.

Why Omni Is Different from Veo 3.1

Google currently runs a fragmented AI creative stack. Veo 3.1 handles video generation. Nano Banana handles image generation. Standard Gemini handles text. These are separate pipelines that do not share context or style.

Omni appears to merge all of these under one roof. The leaked UI text says: Meet our new video model. Remix your videos, edit directly in chat, try a template, and more.

That phrase edit directly in chat is the key difference. Every major AI video model today is generation only. You write a prompt, get a clip, and if you want changes you either regenerate from scratch or export to a separate editing tool. Omni would let you describe changes conversationally and apply them to existing footage.

Metadata found by Android Authority suggests Omni is an evolution of the Veo foundation rather than a completely new architecture. But the consumer facing experience would be fundamentally different: one model, one interface, one conversation for text, images, and video.

No current competitor offers this. Sora 2 is API only after shutting down its consumer app in April. Seedance 2.0 tops video benchmarks but only does video. Kling 3.0 generates over $20 million monthly in China but also only does video. None of them handle text, images, and video under one roof.

The Compute Cost Problem

The most important detail from the leak is the cost. The Reddit user who tested Omni reported that generating just two short video clips consumed 86% of their daily usage limit on the Google AI Pro plan.

This tells us two things. First, whatever architecture Omni uses is extremely compute intensive. Second, if Google launches this at I/O, it will almost certainly be gated behind the paid Gemini Advanced plan, not available on the free tier.

For context, Gemini Advanced costs $19.99 per month as part of Google One AI Premium. If Omni launches with similar restrictions to what the leaks show, heavy video generation users could burn through their daily allocation in minutes.

Google may introduce Flash and Pro tiers for Omni, following the same pattern they use for other Gemini models. A Flash tier with lower resource requirements would make casual video generation accessible. But the full quality Omni experience will likely remain a premium feature.

Source: Pricing verified at one.google.com, May 2026.

What to Expect at I/O Tomorrow

Google I/O 2026 runs May 19 to 20. Gemini and AI updates are confirmed agenda items for the keynote. Based on the leak pattern and Google's history of pre conference UI strings appearing two to four weeks before official reveals, an Omni announcement is highly likely.

Other AI announcements expected alongside Omni include Gemini 3.2 and possibly Gemini 3.5 focused on faster performance, a visual model codenamed Spark Robin, and new memory features internally called Teamfood for long term chat context.

For creators and developers, the practical question is whether Omni will be available immediately or follow Google's typical staged rollout. Veo 3.1 launched with limited access and regional restrictions. Omni could follow the same pattern.

If you currently use AI video tools, do not pause your workflow waiting for Omni. Seedance 2.0, Kling 3.0, and Veo 3.1 are all shipping today and producing professional grade output. But keep an eye on the I/O keynote tomorrow. If Omni delivers on the leaks, it could be the most significant AI creative tool launch of 2026.

We will update this post after the I/O keynote with confirmed details, pricing, and availability.

Source: Google I/O schedule confirmed at io.google, May 2026.