Google’s Gemini Omni video model spotted in early demos with impressive results

Google appears to be testing a new video generation model called “Gemini Omni” that could significantly expand the AI assistant’s multimedia capabilities. Early demos show the technology producing surprisingly realistic videos from text prompts, including complex scenarios like mathematical proofs and dining scenes.

Video generation has become one of the most competitive areas in AI, with companies racing to create tools that can produce realistic footage from simple text descriptions. Google’s existing Veo model has been part of this race, but Omni suggests the company is pushing further into integrated video creation within its main AI platform.

At least one Gemini user was prompted to “Create with Gemini Omni,” which Google describes as “our new video generation model” that can “remix your videos, edit directly in chat, try a template, and more.” The positioning suggests Google wants to make video creation as simple as having a conversation with its AI.

While it’s unclear exactly how Omni relates to Google’s existing Veo technology, metadata suggests Omni is built as an extension of Veo rather than a completely separate system. This approach would make sense for Google, allowing them to integrate proven video generation capabilities directly into the Gemini interface that millions already use.

The early demos show impressive results across different types of content:

• A mathematical proof demonstration where a professor explains trigonometric identities on a chalkboard, with the AI successfully handling both realistic human movement and accurate text rendering
• A dining scene with two men eating spaghetti at a seaside restaurant, complete with detailed environmental elements and natural human interactions

The spaghetti demo is particularly notable because it references the “Will Smith test” – a benchmark in AI video generation that stems from early, often comically bad attempts to show people eating. The fact that Omni appears to handle this scenario well suggests significant progress in the technology.

However, the new capabilities come with substantial computational costs. The two demo videos consumed 86% of a user’s daily allowance on Google’s AI Pro plan, indicating that high-quality video generation remains resource-intensive. This aligns with Google’s recent moves to implement more explicit usage limits across its AI services.

The timing of these demos is significant given the broader context of AI video generation. OpenAI recently discontinued its Sora video model, potentially leaving more market space for Google’s offerings. Google has previously stated that “video’s here to stay” in its AI strategy, suggesting a long-term commitment to the technology even as competitors retreat.

With Google I/O 2026 approaching, the company will likely use its flagship developer conference to officially announce Gemini Omni and detail how it fits into the broader Gemini ecosystem. The integration of video generation directly into Gemini’s chat interface could make the technology more accessible to everyday users, rather than requiring separate specialized tools.

This development represents Google’s continued effort to make Gemini a comprehensive AI platform that goes beyond text responses. By integrating advanced video generation capabilities, Google is positioning Gemini as a creative partner that can produce multimedia content on demand, potentially appealing to content creators, educators, and businesses looking for quick video production solutions.