Google I/O 2026 is still a week away, but Gemini's new video model "Omni" has already been spotted by users. Multiple Reddit users report that over the past week, opening the Gemini app has repeatedly shown a new video generation entry, labeled "Powered by Omni," sitting alongside the existing Veo 3.1 (internal codename Toucan).
One tester who actually used it gave high marks, calling Omni one of the best video models he's seen. He was impressed by its prompt adherence and the coherence of multi-camera transitions. He specifically noted that the speech and environmental audio generated by Omni are a clear step above the Veo series — it even automatically adds context-appropriate background music. But he flagged two clear issues: rate limits are extremely tight — just two video generations used up 80% of his Pro subscription quota — and celebrity likeness guardrails are still in place, blocking the classic Will Smith eating pasta test.
Right now, Gemini's multimedia generation is split across separate models: video from Veo 3.1, images from the Nano Banana series. If Omni is a unified model, it means Google is merging text, image, and video generation into a single architecture. DeepMind CEO Demis Hassabis said publicly last year that he wanted to merge Gemini and Veo, and Omni looks like the realization of that plan. This isn't the only Gemini update in the pipeline — developers recently got a heads-up about a breaking change to the Gemini Interactions API that must be addressed by June 8. Google has also been rolling out multimodal search to the Gemini API's File Search tool, allowing developers to search images and text together using natural language. Google has not officially confirmed the Omni model yet; an announcement is expected at its I/O developer conference on May 19.
