Google appears to be preparing the launch of Gemini Omni, a new artificial intelligence (AI) video generation system that will integrate directly into the Gemini conversational interface, according to interface strings and limited user rollouts observed in the two weeks preceding Google I/O 2026. The annual developer conference is scheduled for May 19–20, 2026.
Pre-launch signals
The first public indication of the system surfaced on May 2, 2026, when <cite index="1-3,1-4,1-5">a user spotted an unusual UI string inside Gemini's video generation tab reading "Powered by Omni," with the model expected to be officially launched at Google I/O 2026 on May 19–20</cite>. <cite index="1-12,1-13,1-14,1-15,1-16">The "Omni" label appeared directly alongside "Toucan" — Google's internal codename for its existing Veo 3.1-powered pipeline — two weeks before Google I/O 2026, a pre-launch staging signal</cite>.
By mid-May, the rollout had widened. <cite index="1-17,1-18">Reddit users on the Gemini app began seeing the "Create with Gemini Omni" prompt themselves, sharing screenshots and early video demos, in what appeared to be either an accidental rollout or an intentional limited A/B test ahead of I/O</cite>.
Product positioning
Based on the leaked interface and early user-generated demonstrations, Omni is being positioned as a departure from Google's existing video pipeline. <cite index="1-8,1-9">Gemini Omni is described as Google's upcoming AI video generation model, integrated directly into the Gemini interface, and — unlike Google's existing Veo lineup, which focuses purely on video generation — is believed to be a true multimodal model capable of handling text, image, and video generation natively, without chaining separate models together</cite>.
That architectural shift would consolidate several previously distinct systems. Until now, Google's generative media stack has relied on Veo for video, Imagen for images, the Nano Banana image-editing model, and Lyria for music, with finished videos requiring users or developers to chain those models together.
The product emphasis observed in leaked clips centers on iterative editing rather than one-shot generation. <cite index="1-22,1-23,1-24">The editing capabilities are described as what separates Omni from competing systems: Seedance 2, Kling 3.0, and OpenAI's Sora 2 are generation tools, whereas Omni is being built as a generation tool that also edits — a difference in product philosophy</cite>.
Competitive context
The staged launch follows roughly a year in which large AI labs have raced to ship higher-quality video models. ByteDance's Seedance 2.0, Kuaishou's Kling 3.0, and OpenAI's Sora 2 all currently target professional-grade clip generation, and <cite index="1-7">early leaks show strong prompt adherence and advanced in-chat video editing, though raw generation quality currently trails ByteDance's Seedance 2.0</cite> according to observers reviewing the leaked clips.
Google has not formally confirmed Gemini Omni as of May 15, 2026, and has not commented publicly on the interface strings or limited rollout. The company is expected to use the I/O keynote on May 19 to detail the model's capabilities, pricing tier, availability across the Gemini app, Google Flow, and YouTube Shorts, and any developer or enterprise application programming interface (API) access. Outstanding questions include clip-length limits, whether the model will ship with synchronized audio output, watermarking provisions through Google's SynthID system, and how Omni will coexist with — or replace — the existing Veo 3.1 pipeline inside consumer products.
AIDB will update this brief following the I/O 2026 keynote.