Modelo de vídeo Gemini Omni no Google I/O 2026: tudo o que sabemos até agora

Gemini Omni is a leaked Google video generation model spotted in Gemini’s UI ahead of Google IO 2026 (May 19–20). Evidence suggests it could be the first top-tier omni-model with native video output, potentially replacing Veo 3.1 and unifying image, video, and text generation under one Gemini system.

What Is the Gemini Omni Video Model Leaked Before Google IO 2026?

On May 2, 2026, an X user named @Thomas16937378 discovered a UI string inside Google’s Gemini video generation tab that read: ‘Start with an idea or try a template. Powered by Omni.’ TestingCatalog, a reliable tracker of Google AI leaks, quickly picked up the finding and published a report that spread across the AI community within hours.

O Gemini Omni video model reference appeared right next to ‘Toucan’ — the internal codename for Google’s current Veo-3.1-powered video generation pathway inside Gemini. Two weeks before Google IO 2026 opens on May 19–20, that placement is hard to dismiss as noise.

I have been tracking Google’s generative video strategy since the original Veo launch, and this is the first time a brand-new public-facing product name has surfaced in the video tab. Previous updates — Veo 2, Veo 3, Veo 3.1 — all kept the Veo branding. ‘Omni’ signals something structurally different.

‘If Google plans to release Gemini Omni for video generation, it would likely outperform Veo 3.1.’ — TestingCatalog

What Was Actually Spotted in the Gemini UI?

The entire body of evidence is a single screenshot — but the details matter. Two things elevate this beyond a random code reference:

• The string is visible to users, not buried in source code or hidden behind feature flags. UI copy that includes a brand name typically reaches that state only when a team is preparing for a public release.
• The placement is next to ‘Toucan’, Google’s known internal codename for the current Veo-backed video generation tool. New code parked next to an existing production pathway is the standard staging pattern before a swap.

A freshly created profile in Gemini’s video tab surfaced the ‘Powered by Omni’ line, suggesting the feature is in late-stage testing. This is not a developer build or an APK teardown — it appeared in the live interface.

Three Possible Interpretations: What ‘Omni’ Could Actually Be

We do not have architecture details, benchmark numbers, or an official statement. But the leak supports three readings, each progressively more transformative.

1. A New Veo Wrapper

The least disruptive explanation: Omni is simply the new product name for Gemini’s video tab, with Veo 3.x or Veo 4 still doing the actual generation work under the hood. Brand consolidation under a single Gemini-native name — similar to how Nano Banana sits on Gemini 3 / 3.1 Flash Image for stills — would explain why a public-facing string is appearing.

2. A New Gemini-Trained Video Model

Google may have trained an entirely new in-house video model under the Gemini umbrella to sit alongside or replace Veo. The current split — Veo for video, Nano Banana / Gemini for images — is architecturally awkward. Omni could be the result of unifying these pipelines, with its own architecture and benchmark profile distinct from Veo 3.1.

3. A True Omni-Model: Single System for Image + Video + More

The most ambitious reading, and the one the nome practically demands: a single Gemini omni-model that handles image generation, video generation, and possibly audio in the same system — the way GPT-4o is positioned for text-image-audio, but with native video output.

If option 3 is correct, Gemini Omni would be the first top-tier omni-model with video output from any major AI provider. That is a meaningful first. Current state-of-the-art video models — Veo 3.1, Seedance 2.0, Kling 3.0 — are all specialized video generators. They do not also handle image creation or text reasoning natively.

The leaked string cannot distinguish between these three possibilities. But only option 3 justifies a brand-new public name like ‘Omni’ rather than simply bumping Veo’s version number.

Will Google Gemini Generate Videos Natively?

Yes — Gemini already generates videos through its integration with Veo 3.1. The question Omni raises is whether Google is moving from a split-model strategy (Veo for video, Nano Banana models for images, Gemini for text) to a unified model that handles all modalities in one system.

Today, Gemini’s video generation flow is labeled ‘Powered by Veo 3.1,’ while image generation is tied to Nano Banana 2 and Nano Banana Pro. Google describes Nano Banana Pro as built on Gemini 3 and Nano Banana 2 as Gemini 3.1 Flash Image. The naming alone shows how fragmented the current approach is.

A unified Omni model could offer concrete advantages:

• Consistent quality across modalities — a single model trained on text, images, and video would maintain stylistic consistency when generating a storyboard image and then animating it into video.
• Simpler creative workflows — no switching between image and video models. One prompt could produce a cohesive image + video sequence.
• Better prompt understanding — a model that understands both visual and temporal concepts could produce more accurate video from complex text descriptions.
• Potential cost efficiencies — maintaining one large model instead of several specialized ones could reduce infrastructure costs at Google’s scale.

Gemini Omni vs the AI Video Generation Landscape in 2026

Video generation is the most competitive category in generative AI right now. If Omni launches at Google IO 2026, it enters a crowded and fast-moving field. Here is how the major players stack up as of May 2026:

ByteDance’s Seedance 2.0 currently sits at the top of most public benchmarks, with Fast and Turbo variants making cinematic AI video financially viable for high-volume production. Alibaba’s HappyHorse-1.0 briefly overtook Seedance on the Artificial Analysis Video Arena leaderboard with an ELO of 1411.

What would make Omni different from all of these is the ‘omni’ part. Every model listed above is a specialized video generator. None of them also handles image creation or text reasoning. If Gemini Omni truly unifies these capabilities, it occupies a category of one.

Where Is Google IO 2026 and When Does It Happen?

Google IO 2026 runs May 19–20, 2026. The event is Google’s annual developer conference, historically held at the Shoreline Amphitheatre in Mountain View, California. Gemini and broader AI updates are confirmed agenda items for the keynote.

The timing of the Omni leak — surfacing exactly two weeks before the event — follows a well-established pattern. Google’s pre-IO leak cycle typically involves UI strings and APK teardowns appearing 2–4 weeks before the keynote, giving the community time to speculate before the official reveal.

That said, UI strings have shipped without product launches before. The most defensible reading of this leak: Google has a video product called Omni in late-stage staging on Gemini, and the most plausible window for a launch is May 19–20.

Is Google Releasing a New Gemini 3 AI Model?

Additional leaks reported by Pankaj Kumar suggest Google is testing new Gemini versions — specifically Gemini 3.2 and Gemini 3.5 — focused on faster performance. These are separate from the Omni leak but part of the same pre-IO testing wave.

Other features spotted in testing include:

• ‘Teamfood’ memory feature — long-term chat context that persists across sessions
• ‘Spark Robin’ — a visual model codename that has appeared alongside Omni in testing references
• Gemini 3.1 Flash-Lite — already launched in General Availability as of May 8, 2026

Google’s model strategy in 2026 appears to be a two-track approach: incremental Gemini version bumps (3.1 → 3.2 → 3.5) for the core language model, and a potential architectural leap with Omni for multimodal generation.

Is Google Veo 3 Free?

Veo 3.1 — the current version — is available through Gemini but has been gated and region-locked. Free-tier access exists for basic video generation within Gemini, but higher-resolution outputs and extended generation times require a Gemini Advanced subscription.

If Omni replaces Veo 3.1, the pricing model could shift. Google has historically offered new AI features in a freemium structure: basic access free, premium features behind a subscription. We expect a similar approach for Omni if it launches at IO.

Third-party platforms like WaveSpeed AI already host Veo 3.1 and plan to add any new Gemini video model shortly after public release, with per-second pricing that can be more cost-effective for production workloads.

The Gemini Nano Controversy: A Cautionary Note on Silent AI Deployments

While the Omni leak generates excitement, Google’s AI deployment practices have also drawn criticism in 2026. A separate but related story: Google Chrome has been quietly downloading a 4GB AI model called Gemini Nano onto user devices without asking for permission.

The file, named weights.bin, lives inside Chrome’s user profile directory and powers on-device features like ‘Help me write’ and scam detection. Users cannot find any checkbox to opt out. When users delete the file, Chrome re-downloads the entire 4GB package automatically.

At Chrome’s global scale, the environmental cost of pushing 4GB to hundreds of millions of devices is estimated between 6,000 and 60,000 tonnes of CO2 equivalent emissions. Mobile data plans in many parts of the world treat 4GB as a month’s entire allowance.

The ePrivacy Directive prohibits storing information on a user’s device without prior consent. Chrome functions perfectly well without an on-device LLM, so no ‘strictly necessary’ exemption applies. This controversy is worth watching as Google prepares to announce even more ambitious AI features at IO.

How to Track and Analyze Gemini Omni Developments Efficiently

If you are following the Gemini Omni story across multiple sources — TestingCatalog, X/Twitter threads, tech blogs, Google’s own documentation — the volume of information can be overwhelming. I have been using iWeaver to aggregate and structure these leaks into a coherent timeline.

iWeaver is an AI agent for office workflows that handles text, images, and documents without requiring complex prompts. I feed it the raw leak screenshots, competitor articles, and official Google announcements, and it outputs structured summaries as doc/pdf files. For anyone tracking fast-moving AI news across dozens of sources, it eliminates the manual copy-paste-organize cycle entirely.

What a Gemini Omni Launch Would Mean for Creators and Developers

If Google ships Omni at IO 2026, the practical implications break down by audience:

For Content Creators

• A unified model means generating a thumbnail image and a matching video from the same prompt, with consistent style
• No more switching between Veo for video and Nano Banana for images
• Potential for audio generation in the same pipeline (if Omni truly covers all modalities)

Para desenvolvedores

• A single API endpoint for multimodal generation simplifies integration
• Platforms like WaveSpeed AI plan to expose Omni through the same OpenAI-compatible endpoint pattern they use for Veo 3.1, Seedance 2.0, and Sora 2
• A/B testing Omni against specialized models becomes straightforward

For Enterprise Teams

• Consolidated billing and model management under one Gemini umbrella
• Reduced vendor complexity if one model handles what previously required three
• Potential cost savings from unified infrastructure

When to Expect the Official Reveal

The timeline is tight. Google IO 2026 opens May 19. The Omni UI string surfaced May 2. A pattern of pre-IO leaks surfacing fresh public names is consistent with a keynote-stage reveal.

Based on previous years, we expect:

1. May 19 keynote — official announcement with demo reel
2. Same day or next day — developer documentation and API access for Gemini Advanced subscribers
3. Within 1–2 weeks — third-party platform availability (WaveSpeed AI, Oimi, and others)
4. Within 1 month — broader rollout including free-tier access

Treat all of this as speculative until Google says it on stage. The most defensible read: Google has a video product called Omni in late-stage staging, and the most plausible launch window is the next two weeks.

Perguntas frequentes

What is Gemini Omni?

Gemini Omni is a leaked Google AI model discovered in Gemini’s video generation UI in May 2026. It may be a unified model capable of generating text, images, and video in a single system, potentially replacing Google’s current Veo 3.1 video model and Nano Banana image models.

Where is Google IO 2026?

Google IO 2026 runs May 19–20, 2026, at Google’s annual developer conference traditionally held at the Shoreline Amphitheatre in Mountain View, California. Gemini and AI updates are confirmed agenda items.

Is Google Veo 3 free?

Veo 3.1 is available through Gemini with limited free-tier access. Higher-resolution outputs and extended generation require a Gemini Advanced subscription. If Omni replaces Veo 3.1, Google will likely maintain a similar freemium pricing structure.

Will Google Gemini generate videos?

Yes. Gemini already generates videos through Veo 3.1 integration. The Gemini Omni leak suggests Google may be building native video generation directly into the Gemini model itself, rather than relying on a separate Veo pipeline.

Is Google releasing a new Gemini 3 AI model?

Leaks suggest Google is testing Gemini 3.2 and Gemini 3.5 alongside the Omni model. Gemini 3.1 Flash-Lite already launched in General Availability on May 8, 2026. Major announcements are expected at Google IO 2026 on May 19–20.

How does Gemini Omni compare to Seedance 2.0?

Seedance 2.0 by ByteDance currently tops most public video generation benchmarks with over 90% commercial usability. Gemini Omni’s differentiator would be unified multimodal generation — handling text, images, and video in one model — which no specialized video generator currently offers.

What happened to OpenAI Sora 2?

OpenAI shut down the Sora 2 consumer app on April 29, 2026. The model remains available through API access only, with a Pro variant for higher resolution output.

What is the Toucan codename in Gemini?

Toucan is Google’s internal codename for the current Veo-3.1-powered video generation pathway inside Gemini. The Omni UI string appeared next to Toucan references, suggesting it may be a replacement or successor.