News
·
3 MIN
Google Omni: announced, shipped, and enterprise-ready are three different things
Gemini Omni turns any input into video, but as of mid-2026 there is no API and no EU-residency path for enterprises.
LOCATION
Worldwide
SERIES
Design, Generative Media & Creative Pipelines
AUTHOR
Sayan Sinha
PUBLISHED
For an enterprise buyer, the real story about Google Omni is the gap between three words: announced, shipped, and available. Its actual name is Gemini Omni, and it is a striking model that turns text, image, audio, or video into video. Google unveiled it at I/O in May 2026. But it launched in consumer surfaces only, with no developer API, no Vertex AI path, and no enterprise data-residency guarantee yet. The interesting part is the strategic bet behind it; the part you can actually buy today is still Veo 3.1.
01.
What Gemini Omni actually is
Get the name right first, because the marketing blurs it. The family is Gemini Omni and the first model is Gemini Omni Flash. It accepts text, image, audio, and video as input, and at launch it produces one output: video, in clips of about 10 seconds with natively synced audio. Google describes it as a step toward a world model, meaning it tries to keep physics, gravity, and spatial coherence consistent across a generated scene, and it runs on Google's 8th-generation TPUs split into training and inference variants.
One honest correction matters for a technical reader: Omni is marketed as any-to-any, but as it shipped it is any-input-to-video. Image and audio outputs are on the roadmap, not in the product. So it is not a real-time conversational omni-assistant, and it is not yet the universal any-to-any system the name implies. It is a very capable input-flexible video generator, which is a narrower and more useful thing to plan around.
02.
Announced, shipped, and available are three different things
The single most useful distinction for a buyer is that these three states are not the same, and Omni sits in different ones depending on who you are. It is shipped to consumers and creators, and merely announced for everyone who builds on an API.
Surface | Omni status (June 2026) |
|---|---|
Gemini app, Google Flow, YouTube | Live, on consumer plans |
Gemini API / Vertex AI | Not available; Google says coming weeks |
Pricing for developers | Not announced |
EU data-residency guarantee | None Omni-specific yet |
The practical reading for an EU company is direct. If you need GDPR-grade EU residency for video generation today, Omni cannot give it to you, because its enterprise path does not exist yet. The supported option on Google's own stack remains Veo 3.1 on Vertex AI, which you can pin to an EU region. This is the same pattern we keep seeing with new model launches, where the demo and the deployable product are months apart.
03.
What a German enterprise should do now
Treat Omni as a signal to track, not a tool to adopt this quarter. Three concrete moves.
Use Veo 3.1 region-pinned for production video today. It has a real Vertex API, EU regions, and an SLA. Omni has none of these yet, so it does not belong in a production pipeline now.
Plan the provenance story early. Omni embeds a SynthID watermark on every clip. For regulated or public-facing content, build the expectation that AI-generated video is labelled and traceable into your process, regardless of which model wins.
Watch two risks, not just the capability. Reviewers have reproduced recognisable trademarked characters in Omni output, which is a real IP exposure for commercial use, and visual consistency still drifts on fast motion. The 10-second cap is a UX limit, not a model ceiling, so expect it to move.
The durable point is the one the launch makes by accident: a model you can see in a keynote is not a model you can deploy under contract. Separate the two, and the hype becomes a roadmap item instead of a procurement mistake.
