Gemini Omni leak: the clearest clues about Google’s next AI video model

The Gemini Omni leak is interesting because it appears inside Google’s own Gemini video experience, not because anyone has a verified model card or public benchmark. The safest reading: Google may be preparing a new Gemini-native video workflow ahead of Google I/O 2026, but the public evidence still rests mostly on app strings and secondhand reports.

What actually leaked

The strongest reported evidence is a line of UI copy in Gemini’s video generation tab: “Start with an idea or try a template. Powered by Omni.” The discovery was attributed to X user @Thomas16937378, and multiple leak roundups say it was later picked up by TestingCatalog and other watchers of Google AI features.

That wording matters for two reasons. First, it places “Omni” directly in a video-generation flow, not in a generic settings menu or unrelated experiment. Second, the phrase “try a template” suggests a more guided creative workflow than a plain text-to-video prompt.

A follow-up report cited additional Gemini mobile app copy: “Meet our new video model. Remix your videos, edit directly in chat, try a template, and more.” If accurate, that points to Omni as a product experience for creating, remixing and editing video inside Gemini, rather than only a back-end model name.

There are also reports of early demos and viral clips. Gadgets360 says early demos showed more realistic motion, cleaner text rendering and improved scene composition, while a user-generated YouTube discussion described two clips circulating on X with metadata purportedly pointing to “Google Gemini Omni Mode.” Those clips are much weaker evidence than the UI strings: Google has not officially announced Omni in the provided sources, and the clips are not independently verified as Omni outputs.

What Omni may offer if the strings are accurate

The leaked copy points to four possible user-facing features:

• Templates for video creation. The “start with an idea or try a template” wording suggests Google may be preparing preset formats or guided workflows for users who do not want to prompt from scratch.
• Video remixing. Reported mobile app copy explicitly mentions remixing videos.
• Editing directly in chat. Multiple reports describe Omni as a tool that could create and edit AI-generated videos within Gemini’s chat interface.
• Tighter Gemini integration. The leak appears inside Gemini’s video-generation experience, which suggests Omni may be positioned as a native part of Gemini rather than a separate standalone video app.

That is the meaningful product angle. The leak does not yet verify frame length, resolution, API access, prompt limits, generation speed, audio quality, safety behavior or pricing.

How this compares with Veo 3.1

Veo 3.1 is the confirmed baseline. Google released Veo 3.1 and Veo 3.1 Fast in paid preview through the Gemini API, Google AI Studio and Vertex AI, and said the models also became available in the Gemini app and Flow. Google described Veo 3.1 as an update with richer native audio, more narrative control and improved results when generating video from images.

Google has also continued extending the Veo 3.1 family. In January 2026, it said Veo 3.1 could create more expressive videos from images, generate vertical videos for platforms such as YouTube Shorts and upscale to 1080p or 4K across products including Gemini, Flow, the Gemini API, Vertex AI and Google Vids. In March 2026, Google introduced Veo 3.1 Lite as its most cost-effective video model, saying it cost less than 50% of Veo 3.1 Fast while running at the same speed.

Against that official Veo 3.1 backdrop, the Omni leak does not prove a simple “better specs” jump. The clearest difference is workflow: templates, in-chat editing and remixing inside Gemini. Reports claiming better motion, cleaner text or improved composition may be true, but they are not yet supported by a Google model card, benchmark or reproducible public test.

Is Omni a rebrand, a new model or a unified multimodal architecture?

All three possibilities remain open.

One possibility is that Omni is a new label for an existing or upgraded Gemini video path. WaveSpeed’s report says the “Powered by Omni” string appeared near “Toucan,” described there as the internal name for Gemini’s current Veo 3.1-powered video tool. If that placement is accurate, Omni could be a replacement path, a test flag or a UI-facing name for a new generation pipeline.

A second possibility is that Omni is a genuinely new video model. The reported app copy says “Meet our new video model,” and Gadgets360 describes Gemini Omni as a model that may let users create and edit videos directly within Gemini.

A third possibility is that Omni is part of a broader multimodal system. Some leak roundups speculate that Omni could unify text, image, video and audio generation or reasoning under a single Gemini architecture. That is plausible as a product direction, but it is still speculation in the provided sources. Google has not confirmed whether “Omni” is a public product name, an internal codename, a model family, a UI layer or a broader architecture.

What testers noticed about limits and compute costs

There is not enough verified evidence to say what Omni costs to run or use. The provided reports do not confirm Omni pricing, latency, quota limits, generation length, model size, API availability or compute requirements.

The best comparison point is Veo, where Google already segments the family by cost and performance. Veo 3.1 Lite was introduced as a lower-cost option at less than half the cost of Veo 3.1 Fast, while keeping the same speed. That shows Google is thinking about video generation economics, but it does not tell us whether Omni will be expensive, limited, premium-only or developer-facing.

For now, claims that Omni is slow, costly, restricted to internal testers or limited to short clips should be treated as unconfirmed unless Google or a verifiable tester publishes evidence.

How Omni compares with Runway, Pika and Sora

No supported head-to-head ranking is possible from the current evidence. The provided sources do not include comparable benchmark data for Runway, Pika or OpenAI Sora, and the Omni evidence itself is not enough to judge realism, controllability, generation length, temporal consistency, safety systems or cost.

The only defensible comparison is product positioning. If the leaked Gemini copy is accurate, Google may be trying to compete not only on video quality but on workflow: prompt from Gemini, choose a template, remix a clip and edit through chat in the same interface. That could be a meaningful differentiator against standalone AI video tools, but it is not proof that Omni outperforms Sora, Runway or Pika on output quality.

What to watch at Google I/O 2026

The key questions for I/O are straightforward:

• Does Google announce Omni by name?
• Is Omni a new public model, a Veo successor, a Gemini UI layer or an internal codename?
• Does Google publish a model card, API details, safety notes or pricing?
• Are templates, remixing and in-chat editing actually part of the launch?
• Does Google provide verified comparisons with Veo 3.1, including duration, resolution, audio, text rendering and image-to-video quality?

Until those answers arrive, the Omni leak should be read as a credible signal of Google’s likely video direction, not as a confirmed specification sheet. The app strings are the story; the rest is still waiting for Google to make it real.