An impressive new Gemini ‘Omni’ video model just leaked ahead of Google I/O

If you’ve been following the AI space lately, you know that video generation is the current frontier. While OpenAI has seemingly pumped the brakes on its Sora model, Google has repeatedly stated that “video is here to stay.” We’ve known about their Veo model for a while now, but according to a new report from 9to5Google, it looks like a brand-new iteration called Gemini Omni is already starting to show up for some users in the wild.

What is Gemini Omni?

According to the leak, at least one Gemini user was prompted to “Create with Gemini Omni.” Google describes the new video generation model as a way to remix your videos, edit directly in chat, and try out pre-made templates.

While it isn’t entirely clear how Omni fits into the broader context of Gemini and the existing Veo model, metadata suggests that Gemini Omni might actually be an extension or evolution of the Veo foundation. Regardless of the naming convention, the early results are looking pretty impressive.

Passing the spaghetti test

If you remember the early days of AI video, you probably remember the horrifying viral clip of Will Smith eating spaghetti. It became the unofficial benchmark for how bad AI video could be. Well, one of the leaked Omni demos specifically used a prompt for two men eating spaghetti at an upscale restaurant, and the results were incredibly realistic.

Another demo featured a professor writing out a mathematical proof for trigonometric identities on a traditional chalkboard while explaining the steps. While there are still some obvious AI “tells” in the final output, the model handles the text generation and physical movements remarkably well.

Know your usage limits

Of course, generating hyper-realistic video in a chat window takes an immense amount of computing power, and Google is apparently very aware of that. The user who spotted these Omni prompts also noticed a new “usage” tab on their account. Generating just those two video prompts managed to chew through 86% of their daily usage on an AI Pro plan.

Google’s intentions to add more explicit usage limits for Gemini have recently been spotted, and a resource-heavy model like Omni is likely the exact reason why. With Google I/O 2026 right around the corner, I would place a solid bet that Gemini Omni is going to get some serious stage time. We’ll be watching closely to see exactly how this rolls out and what it means for the future of video creation in Gemini.