On the eve of Google I/O 2026 (May 19-20), a leak report from TestingCatalog has stirred wide attention in the AI community: Google is testing a new video generation model called Omni, to be opened to consumers as a Gemini built-in feature.
The leaked screenshots show the Gemini video generation interface's branding silently changed from "Powered by Veo 3.1" to "Powered by Omni." This change is far more than a brand name swap: it may signal Google officially bidding farewell to the "multi-model division of labor" strategy that has run for two years — Veo handling video, Imagen/Nano Banana handling images, each operating independently. Omni is expected to unify video and image generation under the same model architecture.
This strategic pivot is not hard to understand. OpenAI has long used Sora to unify the video-generation brand exit, while Google has maintained a three-brand four-tool dispersed layout — for ordinary users, it's hard to tell where Gemini's video feature "lives." If Omni lands as leaked, Google will for the first time have a more concise consumer-grade multimodal generation naming system than OpenAI.
But the current information is still quite limited. The leak only reveals brand attribution, and it's not yet clear whether Omni is a brand-new model based on Veo 4, or an existing combination packaging of Nano Banana series and Veo. Key details like pricing, API opening time, and output duration are also not disclosed.
With Google I/O approaching, is Omni a big move from Google in the multimodal generation space, or just a product-level brand integration? The answer is about to be revealed.