With less than a week until Google I/O 2026 kicks off, a new AI video generation model codenamed “Gemini Omni” has already leaked online. From UI strings being discovered in the Gemini App interface last Friday (5/2), to actual generated demo videos going viral on X (formerly Twitter) this Monday (5/11), this model that could become Google’s next-generation flagship video generator is sparking heated discussions across the tech community.
Google’s All-New Video Generation Model Gemini Omni Leaked Early
Everything started on May 2nd. The well-known AI news site TestingCatalog discovered that some users saw a brand new string in the video generation tab of the Gemini App: “Start with an idea or try a template. Powered by Omni.”
This line of text appeared in the same UI area alongside “Toucan,” the internal codename for Google’s existing Veo 3.1 video generation feature. This is a key signal: while Veo has evolved from version 1 to 3.1, the brand name has remained unchanged throughout, but the first appearance of a brand-new product-level name in the Gemini App typically means Google is preparing for a major announcement at the upcoming I/O conference.
As of May 11, some Gemini AI Pro users had been prompted that they could “create content using Gemini Omni” and began sharing the videos they actually generated. Posted by AI journalist Chetaslua.A demo videoGained over 1.2 million views quickly
Holllllyyyyyyyy @GeminiApp cooked 😳😳
🚨 Gemini Omni: New video model
Here is the first output and see the text coherence , if this is not nano banana moment of video then what is ??
direct link for those who believes otherwise in comments pic.twitter.com/LUqJLXUxrf
— Chetaslua (@chetaslua) May 11, 2026
The prompt for the video was: “A professor writes out a mathematical proof of a trigonometric identity on a traditional blackboard and explains the step he is currently working on.” The results were stunning—the mathematical formulas like sin²θ + cos²θ = 1 that appeared on the blackboard not only featured smooth handwriting and natural writing animations, but the text was also completely accurate, free from the “gibberish text” problem commonly seen in previous AI video generation models. Many comments called this moment “the nano banana moment of video generation.”
However, some observers have pointed out that Omni’s lip-sync realism still falls short of Seedance 2.0 or Kling, and the audio has a noticeable “Google voice” style, with room for improvement in the blackboard texture and chalk physics effects.
What is Omni? Three possible interpretations
Google has yet to officially reveal the technical details of Omni, but there are three speculations about its true identity:
Type 1: New brand packaging for Veo 4The most conservative interpretation: Omni is simply a new product name for Veo 4 or Veo 3.2 in the Gemini App, with the underlying technology still being the Veo series. Similar to how Gemini 3.1 Flash named static image generation “Nano Banana 2.”
Second type: The brand new Gemini-native video model.Google may have trained a brand new video model called Gemini, running in parallel with or replacing Veo. This would unify the current multi-model strategy of “Veo for video, Nano Banana for images, and Gemini for text.”
Type 3: A Truly Omnipotent ModelThe most ambitious interpretation: “Omni” could be a single multimodal model capable of processing text, images, video, and even audio simultaneously. If true, this would be the first all-purpose model with native video output from a major AI vendor—far more significant than a simple version bump.
The ambition behind the multimodal approach is even evident in the name itself: “Omni” means “all” in Latin. This aligns with GPT-4o’s concept of integrating text, images, and audio into a single model, but Omni takes it a step further by adding native video output capability.
Google I/O 2026 is coming
According to leaked usage data, a user on the AI Pro plan used up 86% of their daily quota after testing just two Omni prompts, indicating that this model’s computational costs are quite high.
Google will hold its I/O 2026 developer conference on May 19–20, with Gemini Omni widely expected to make its official debut at the event. Given that OpenAI has recently announced plans to gradually wind down its Sora video generation business, Google’s decision to open Omni for testing on the eve of I/O carries particularly significant timing.
Source: KOCPC Chinese