Google had moved the video generation space forward with Veo, and it appears it could do an encore with a model named Gemini Omni.
The discovery came courtesy of X user @Thomas16937378, who spotted a UI string inside Gemini’s video generation tab that read: “Start with an idea or try a template. Powered by Omni.” A follow-up post from TestingCatalog showed new evidence of Gemini Omni appearing inside the Gemini mobile app as well, alongside the message: “Meet our new video model. Remix your videos, edit directly in chat, try a template, and more.”
Other users also discovered Gemini Omni’s model id on Gemini. Omni apparently currently has a 10-second video generation limit.
What Omni Could Actually Be
The name itself is doing a lot of heavy lifting. Three plausible interpretations have emerged in the community:
A rebrand of Veo. The least disruptive reading — Google may simply be retiring the Veo brand in consumer products and replacing it with “Omni” as a unified identity, much like it consolidated image generation under the Nano Banana name.
A new Gemini-native video model. A version of the Gemini architecture fine-tuned specifically for video output, separate from the Veo model family.
A true omni-model. The most ambitious — and most disruptive — scenario: a single Gemini model that natively generates text, images, and video within one unified system. This would make Gemini the first top-tier omni-model with video output, a meaningful first in the space.
The third interpretation is what the name implies and would represent a genuine architecture shift. Google currently runs a split-model approach: Veo 3.1 for video, Nano Banana 2 and Nano Banana Pro for image generation. A true Omni could collapse all of that into one consumer-facing product — something structurally different from OpenAI’s Sora, and from anything else currently available.
Impressive Early Outputs
Omni’s outputs seem to be quite impressive, in particular a clip of a professor writing math equations on a blackboard. The equations are correct, and the video also seems to be very lifelike.
Getting math right in AI-generated video is genuinely hard — it requires not just visual coherence but semantic accuracy. The fact that an early build is handling it has raised expectations ahead of a formal reveal.
The Competitive Context
The timing matters. Google’s Veo 3.1 is already considered one of the strongest video generation models on the market, capable of 4K output with natively generated audio. But the leaderboard has remained competitive, with Runway Gen-4.5 having previously edged out Veo 3 on Artificial Analysis benchmarks, and ByteDance’s Seedance 2.0 continuing to rank highly on public evaluations.
Omni, if it’s what the leaks suggest, would arrive as something qualitatively different — not just a better video model, but a model that handles video, images, and potentially audio within a single system. That changes the competitive framing entirely. It’s not just a fight for benchmark rankings; it’s a fight for which platform becomes the single destination for AI-native content creation.
Google I/O Is Ten Days Away
Google I/O 2026 is scheduled for May 19–20, and Omni’s consumer-facing UI string surfacing this close to the event is not likely a coincidence. Additional leaks, reported alongside the Omni discovery, point to new Gemini versions (3.2 or 3.5) focused on speed, a long-term memory feature internally codenamed “Teamfood,” and a visual model codenamed “Spark Robin.”
Google has not officially confirmed Omni. But given that the “Powered by Omni” label appeared in the live Gemini UI — not in a private alpha or buried developer log — it’s past the stage of idle speculation. The realistic outcome is a main-stage announcement at I/O, tied to a broader push across Gemini, Workspace, and Android.
If Gemini Omni turns out to be a genuine omni-model, the creator economy implications are significant. The current workflow — storyboard in one tool, still frames in another, video in a third — collapses into a single prompt. That’s not an incremental improvement, but a different product category entirely.