Gemini Omni: Everything Leaked Before Google I/O 2026
Gemini Omni is the codename for an unannounced Google video-generation product that surfaced inside the Gemini app's UI on May 2, 2026, eight days before Google I/O 2026 opens on May 19–20. The only confirmed public evidence is a single leaked string — "Start with an idea or try a template. Powered by Omni." — found in Gemini's video generation tab and surfaced by TestingCatalog. Google has not officially confirmed Omni as of May 11, 2026.
This article will be updated within 24 hours of any official Google announcement.
What is Gemini Omni?
Gemini Omni is, as of May 11, 2026, a product name that exists only inside Google's own staging code. No specs, no demos, no model card, no Google PR. The name implies a multimodal capability scope — "omni" suggesting image, video, and possibly audio in one model — but Google has neither published nor leaked any technical details. Everything beyond the single discovered UI string is analyst inference.
The reason the leak matters: Omni appears parked next to "Toucan" — the internal codename for Gemini's existing video-generation surface, which currently runs on Veo 3.1. Staging a new public product name next to an existing one is the standard pattern Google uses 1–2 weeks before swapping the underlying engine or relabeling the surface.
What was leaked, and where it came from
The leak is small in surface area but unusually clean in provenance. Here is what is publicly verifiable, and what is not.
The exact UI string
On May 2, 2026, a screenshot of the Gemini app's video generation tab surfaced containing the line: "Start with an idea or try a template. Powered by Omni." TestingCatalog published the screenshot the same day, and WaveSpeed followed with analysis on May 3.
The string had not appeared in any prior public build of Gemini. As of publication, it remains in test surfaces only — not in the production Gemini app most users see.
What this proves: Google has shipped a new product name — "Omni" — into a build pipeline that touches consumer-facing Gemini UI. What this does not prove: that "Omni" is a new model, that it does anything Veo 3.1 cannot, or that it will reach production users at I/O 2026.
Toucan, and how Omni fits beside it
"Toucan" is the internal codename for Gemini's existing video-generation tab — the surface users today see when they generate a clip inside the Gemini app. Toucan currently runs on Veo 3.1, Google's deployed flagship video model. Toucan was discovered earlier through similar UI-string archaeology and is not itself a model.
In the May 2, 2026 leak, "Omni" appeared inside the same UI surface area as the existing Toucan references. That co-location is what shifted the leak from "random string" to "imminent product launch." Google's standard pattern for replacing a consumer-facing model is to land the new product name in code 1–2 weeks before announcement, run parallel staging, and flip the switch on keynote day.
Three plausible interpretations of Omni
Analyst coverage since May 2 has converged on three competing readings of what Omni actually is. None is confirmed.
Theory 1 — A rebrand of the Veo pipeline
Omni is a new consumer-facing product name for the same Veo-powered backend. Toucan/Veo 3.1 continues to do the actual generation; only the label changes. This is the least disruptive reading and would explain why Google has not pre-briefed press: there is no new technology to brief on. Supporting evidence: Google has been moving toward consumer-friendly product names that abstract over backend models — "Toucan" itself never appeared as a user-facing label.
Theory 2 — A new Gemini-trained video model
Omni is a fully new video model trained inside the Gemini line, parallel to Veo. Under this reading, Veo remains the Vertex AI / Google Cloud product for enterprise, while Omni becomes the consumer pillar inside the Gemini app. This explains the "Omni" name — it would identify the model itself, not a wrapper — and aligns with Google's history of running separate consumer and enterprise model tracks.
Theory 3 — A unified omni-model for image, video, and audio
Omni is the first member of a new Gemini-trained model family that handles image generation, video generation, and synchronized audio in a single forward pass. This interpretation is the most ambitious: it would mean a single model replacing both the current Veo pipeline and the Nano Banana Pro image stack inside Gemini. The "omni" name is the strongest evidence for this theory, but no leak supports synchronized audio or unified image+video specifically — it is inference from the name alone.
When will Gemini Omni launch?
The most likely launch window is Google I/O 2026, May 19–20, 2026 — eight days from this article's publication. Google has confirmed Gemini and AI updates on the keynote agenda. The 1–2-week staging pattern observed in past Gemini launches makes a keynote reveal the highest-probability outcome.
Adjacent leaks support an I/O launch. The same UI sweep that surfaced "Omni" also surfaced references to "Gemini 3.2" and "Gemini 3.5" (@pankajkumar_dev, May 3, 2026) — suggesting Google may bundle Omni as the video-generation pillar of a broader Gemini 3.2 / 3.5 model release.
Two non-trivial risks to that read: (1) Google has previously staged UI strings for products that slipped 1–3 months past initial leaks, and (2) "Powered by Omni" could remain a developer-build label even if the consumer rollout is gated to a fraction of users at I/O. The defensible claim today is that an I/O 2026 announcement is the most likely outcome — not a certainty.
How Omni stacks up against the competition
Because Omni has no published benchmarks, capabilities, or sample outputs, any quality comparison is speculation. What can be compared is status and structural positioning — where each player sits in the May 2026 video-generation landscape.
| Model | Status (May 11, 2026) | Native audio | Max public clip length | Where to try today |
|---|---|---|---|---|
| Gemini Omni | Leaked, unannounced | Rumored, not confirmed | Unknown | Not yet available |
| Veo 3.1 (Toucan) | Live in Gemini, also on Vertex AI | Yes | ~15s with scene extension | lovegen.ai/veo3-1 |
| Sora 2 | Live since 2025 | Yes | Variable, supports editing | Sora app, lovegen.ai |
| Seedance 2 | Live, current public-benchmark leader | Yes | Varies by plan | lovegen.ai |
Distribution is the differentiator that matters now. Seedance 2.0 leads public video-generation benchmarks as of May 2026, but reaches creators primarily through dedicated tools. Sora 2 has consumer reach via its standalone app. Omni — if it launches inside the Gemini app — inherits Gemini's existing consumer user base by default. That is the dimension on which Omni is most likely to matter, independent of whether its raw output quality leads the field.
What we know vs. what we don't know
The single most important piece of context for this story is the gap between leak-confirmed facts and analyst extrapolation. Treat any row marked "Speculation" as a rumor until I/O 2026.
| Claim | Status | Source |
|---|---|---|
| A product named "Omni" exists in Gemini's build pipeline | Confirmed | TestingCatalog, May 2, 2026 |
| Leaked UI string reads "Start with an idea or try a template. Powered by Omni." | Confirmed | @testingcatalog post |
| Omni appears in Gemini's video-generation tab, alongside Toucan | Confirmed | TestingCatalog |
| Toucan currently runs on Veo 3.1 | Confirmed | TestingCatalog |
| Omni is a new model (vs. a rebrand of Veo) | Speculation | WaveSpeed analysis |
| Omni handles image + video + audio in one model | Speculation | Inference from the "Omni" name |
| Omni generates synchronized spatial audio | Speculation | Unsourced rumor |
| Omni outperforms Veo 3.1 | Speculation | TestingCatalog speculative line |
| Omni launches at Google I/O 2026 (May 19–20) | Reported / probable | Staging-pattern inference plus I/O agenda |
| Omni is paired with Gemini 3.2 or 3.5 | Reported | @pankajkumar_dev, May 3, 2026 |
| Pricing, availability tier, regional rollout | Unknown | No leak |
What creators should do until I/O 2026
The eight-day gap between this article and the I/O keynote is short, but real. Three reader scenarios — three different answers.
If you're mid-project on Veo 3.1
Finish the project on Veo 3.1. Do not pause work on the chance Omni ships at I/O. Two reasons. First, even in the most optimistic interpretation (a new model launched at the keynote), Omni will likely roll out to a fraction of users initially and may take weeks to reach general availability. Second, prompting techniques you build on Veo 3.1 transfer directly to any successor model in the Gemini family — Google's prompt grammar across Veo generations has remained largely stable.
If you're choosing between Sora 2, Seedance 2, or waiting
If your timeline is more than 30 days out, waiting one week for the I/O announcement costs almost nothing — and the announcement will materially change the decision. If you need output within the next seven days, the defensible options today are Seedance 2 (current benchmark leader) or Sora 2 (most mature consumer flow). "Powered by Omni" alone is not enough basis for a production commitment.
If you produce image + video with Nano Banana Pro
The image+video stack on Gemini today uses two models: Nano Banana Pro for images, Veo 3.1 for video. If Theory 3 is correct and Omni is a unified omni-model, your workflow simplifies — one model handles both. But until that's confirmed at I/O, treat the current Nano Banana Pro → Veo 3.1 handoff as the supported path and avoid restructuring pipelines on speculation.
FAQ
What is Gemini Omni?
Gemini Omni is the codename for an unannounced Google video-generation product that appeared inside the Gemini app's UI on May 2, 2026. The only confirmed public evidence is a single string — "Start with an idea or try a template. Powered by Omni." — discovered in the video-generation tab. Google has not officially announced Omni as of May 11, 2026, but its placement alongside Toucan, Gemini's existing Veo 3.1-powered video tool, suggests it is being staged for imminent launch at Google I/O 2026.
Is Gemini Omni officially announced?
No. As of May 11, 2026, Google has made no official announcement about Gemini Omni. The only public evidence is a UI-string leak surfaced on May 2, 2026 by TestingCatalog. Any feature claims beyond the leaked string — synchronized audio, higher resolution, unified image+video — are analyst speculation, not Google statements.
When will Gemini Omni launch?
The most likely launch window is Google I/O 2026, which runs May 19–20, 2026 — eight days after this article's publication. Google has confirmed Gemini and AI updates are on the keynote agenda. Two-week staging of a public-facing UI string is a standard pre-launch pattern, but no Google source has confirmed Omni will appear at I/O. A delayed reveal at a later event remains possible.
Is Omni replacing Veo 3.1 inside Gemini?
It might be, but it has not been confirmed. The leaked UI shows Omni appearing next to "Toucan," the internal codename for Gemini's existing Veo 3.1-powered video tool. Parking a new product name next to an existing one is the typical staging pattern before a swap, but Google could also keep Veo as a Vertex AI / Cloud product while branding the consumer Gemini experience as Omni.
What is Toucan in Gemini?
Toucan is the internal codename for Gemini's current video-generation tool — the tab inside the Gemini app where users today generate clips powered by Veo 3.1. Toucan was discovered earlier through similar UI-string leaks and is not a separate model from Veo. The May 2026 leak surfaced "Omni" sitting next to "Toucan" in the same UI surface area, suggesting a parallel or replacement product.
Will Gemini Omni do image generation too?
Unknown. The leaked UI string only references video generation. However, the name "Omni" implies multimodal capability — image, video, possibly audio in one model — which would be a meaningful break from Google's current setup where image generation runs on Nano Banana 2 and Nano Banana Pro. A unified omni-model is one of three plausible interpretations analysts have proposed, but none of it is confirmed.
How does Gemini Omni compare to Sora 2?
A direct comparison is not possible because Omni has no published benchmarks, no public outputs, and no confirmed capabilities. Sora 2, by contrast, has been publicly available since 2025 and ships with verified clip lengths, audio support, and editing controls. The most defensible claim today is that if Omni launches at I/O 2026, it will be Google's most serious consumer-facing answer to Sora 2 to date.
How does Gemini Omni compare to Seedance 2.0?
ByteDance's Seedance 2.0 leads public video-generation benchmarks as of May 2026. Omni has no benchmarks, so any quality comparison is speculation. The structural difference is distribution: Seedance 2.0 is an API-and-app product targeting creators; Omni, if launched inside the Gemini app, would reach Gemini's existing consumer user base by default — a much wider distribution surface than Seedance commands today.
Will Omni generate synchronized audio?
Synchronized audio is the single most-rumored Omni feature, but it has not been confirmed by Google or by the leaked UI string. The rumor stems from the "Omni" name itself, which implies multiple modalities, and from analyst extrapolation. If Omni does ship with native synchronized audio, it would close the most-cited gap between Google's video output and Sora 2, which has supported synced audio since launch.
Is Omni connected to Gemini 3.2 or 3.5?
Possibly. The same UI-leak sweep that surfaced "Omni" also surfaced references to Gemini 3.2 and Gemini 3.5, suggesting the next Gemini model generation is staging alongside Omni. The most defensible reading is that Google may announce Omni as the video-generation pillar of a broader Gemini 3.2 / 3.5 launch at I/O 2026. The exact pairing has not been confirmed.
How will I try Gemini Omni when it launches?
If past Google rollouts are a guide, the most likely first surface is the video-generation tab inside the Gemini app itself — the same tab where the "Powered by Omni" string was discovered. Wider access will probably arrive through Gemini Advanced, Google Workspace, and a Vertex AI API for developers. We will update this article with confirmed availability paths within 24 hours of the I/O 2026 announcement.