New signs about Google’s upcoming Gemini Omni video model surfaced over the weekend, with a Reddit user posting a screenshot of a revised Gemini interface that reveals new model cards. “Create with Gemini Omni: Introducing new video models, remix videos, edit directly in chat, try templates, and more,” the description reads, and seems to confirm the long-rumored unified approach Google is preparing for next week’s developer event. This development appeared to be a coincidence or part of a limited A/B test.
Sample video and initial feedback 👀
> I’m not going to lie, this is one of the best video models I’ve ever seen, and while it may not be the *best*, it’s a really strong performer. I was particularly impressed with how quickly the models responded (except for one shot where the centerpiece was missing). pic.twitter.com/sG58DXlswL
— 🚨 AI News | Testing Catalog (@testingcatalog) May 11, 2026
Users found a new Usage Restrictions tab within Settings, alongside the Model Card. Several people also reported that generating videos quickly consumed their credits. This suggests a pay-as-you-go system similar to what Google is testing across its Gemini surface. Early works received mixed reactions. When it comes to raw production fidelity, Omni seems to lag behind ByteDance’s Seedance 2, with viewers noting that its cinematic quality is a step behind the current benchmark leader. What made the model stand out was the editing. Removing watermarks, replacing objects within clips, and rewriting scenes with chat instructions all worked unusually well for a first release.
GOOGLE 🔥: Google’s upcoming Gemini Omni video model is expected to have even more advanced features in video editing, allowing you to perform tasks like removing watermarks and replacing objects in videos.
Also, Google is likely to release two versions of this model. https://t.co/OJUOBUXjOw pic.twitter.com/lT9sDlI8Lu
— 🚨 AI News | Testing Catalog (@testingcatalog) May 11, 2026
This pattern mirrors Nano Banana, which was launched as a native image model in Gemini. Nano Banana debuted with a middling gen score, but topped the editorial leaderboard and was later upgraded to the Frontier Image System. Google appears to be following the same strategy with video, prioritizing modality consolidation under Gemini over quality leadership at launch. There are also hints that Omni ships in phased variants (possibly Flash and Pro) and that the output currently in circulation is most likely from the Flash tier.
Google continues to prepare for the release of the upcoming Gemini Omni model.
> Gemini Omni models will also be available in the API
> Models are considered agents, similar to Deep Research in AI Studio
Right away? 👀
PS: Let me remind you that Nano Banana 1 was not better than that… pic.twitter.com/QnkbQ9WRQm
— 🚨 AI News | Testing Catalog (@testingcatalog) May 11, 2026
The timing perfectly coincides with Google I/O on May 19th and 20th, when the company has a track record of announcing its most ambitious AI shifts. The combination of a short pre-event window and controlled leaks gives Google room to gather reactions and shape the story before the keynote.
sauce
