Gemini's new video model, 'Omni,' detected before official launch; users praise audio quality

AIMPACT Message, May 11 (UTC+8): According to monitoring by Beating, with Google I/O 2026 still a week away, Gemini’s new video model “Omni” has already been accidentally discovered by users. Multiple Reddit users reported that over the past week, repeatedly opening the Gemini app triggered a new video generation entry labeled “Powered by Omni,” appearing alongside the existing Veo 3.1 (internal code name: Toucan). One user who actually tested it gave high praise, calling Omni one of the best video models he has ever seen, impressed by its prompt adherence and seamless multi-camera transitions. He particularly noted that the audio and ambient sound quality generated by Omni surpasses that of the Veo series by more than one level, even automatically adding background music suited to the scene. However, he also highlighted two clear issues: the rate limits are extremely strict—Pro subscribers exhaust 80% of their quota after generating just two videos; and celebrity likenesses are still blocked by safeguards, as the classic test of Will Smith eating pasta fails to work. Currently, Gemini’s multimedia generation is split: video relies on Veo 3.1, while images use the Nano Banana series. If Omni is a unified model, it suggests Google is integrating text, image, and video generation capabilities into a single architecture. DeepMind CEO Hassabis publicly stated last year that Gemini and Veo would be merged; Omni is likely the realization of this plan. Google has not yet officially confirmed the model, but it is expected to be announced at the I/O conference on May 19. (Source: BlockBeats)