In the 2026 AI video generation race, Seedance 2.0 has become the benchmark for virtually every new model released. Whether it’s Google’s Veo series or OpenAI’s Sora, each new release gets compared to Seedance 2.0. This kind of “being the standard” is itself a form of recognition.
But competition never stops. Just recently, Google—having been relatively quiet—seems to be preparing something big. Gemini Omni, an entirely new video generation model, accidentally appeared in the Gemini chat interface.
Gemini Omni Emerges
Early this morning, a user accidentally triggered this new video generation feature in Gemini. The interface directly displayed:
“Create with Gemini Omni—meet our new video generation model. Remix your videos, edit directly in conversation, try rich templates, and more.”
Although the feature disappeared when the user reopened the app, the leaked test videos were enough to cause a stir in the community.
Challenging Mathematical Blackboards: A Breakthrough in Text Rendering
The first leaked test case was a 10-second video: a professor writing mathematical proofs for trigonometric identities on a traditional blackboard while explaining the steps.
The prompt was just one sentence: “A professor writes out a mathematical proof for trigonometric identities on a traditional chalkboard, explaining the step he is currently on in the equation.”
Although some AI-generated artifacts could still be found in the footage, the video’s handling of the text content on the blackboard was quite precise—a major bug that plagued many previous video models. The overall realism of the video was also impressive.
A Classic Tribute: Will Smith Eating Spaghetti
The second test case followed this classic tradition—the “Will Smith eating spaghetti” has become one of the must-test prompts for virtually all video generation models.
Due to content restrictions, the tester described two men who looked similar to Will Smith dining at an upscale restaurant. In the video generated by Gemini Omni, the entire process—from the men approaching the table and exchanging pleasantries to starting to eat the spaghetti—was smooth and natural.
Side-by-Side: Gemini Omni vs Seedance 2.0
What happens when the same prompts are fed to both models?
Text Rendering Capability
In the mathematical blackboard test, Gemini Omni handled the blackboard text significantly better. The content generated by Seedance 2.0 had little connection to trigonometric identities, and there were still issues with text rendering on the blackboard.
Physics Logic Handling
In the spaghetti-eating scene, Seedance 2.0 performed more naturally. Gemini Omni showed some minor physics issues: when the characters were standing, there was no spaghetti on the plate, but it suddenly appeared after sitting down, and during eating, the spaghetti disappeared without being actually consumed.
Overall Assessment
Gemini Omni is indeed impressive in text consistency, but Seedance 2.0 still holds the advantage in video generation’s physics logic. Both products have their strengths, and actual performance depends on the specific use case.
Not Just Generation: Gemini Omni’s Video Editing Capabilities
This leak also highlighted another highlight—Gemini Omni demonstrated powerful video editing capabilities.
Users can directly replace objects in videos within conversation. For example, the spaghetti in the original video can be replaced with a bowl of Tom Yum soup. More practically, it can also remove watermarks generated by other AI tools—the test successfully removed the continuously moving Sora watermark while maintaining the original video’s basic consistency.
AI Video: Becoming a Mainstream Medium
The competition between these two products is just the tip of the iceberg. Recently, a wave of high-quality AI videos has been exploding across the internet.
Marko Slavnic, a technical artist at Runway, created an AI-generated pigeon video that garnered a million views. The 47-second clip showed no AI traces at all—perhaps the Disney animation team should be worried.
In China, content under the Weibo topic “AI Short Films: Mass-Producing Love, Death, and Robots” went viral internationally, reaching 12 million views after being reposted overseas. The creator revealed that the short film was made using Seedance 2.0 over 10 days and has already partnered with a film company for a theatrical release.
These cases demonstrate that AI video is evolving from a “lab toy” into a mainstream creative medium.
Is Seedance 2.1 Coming?
Google’s leak of Gemini Omni might be building hype for next week’s Google I/O conference. If Gemini Omni officially launches at the conference, the second half of the AI video generation competition will officially begin.
Meanwhile, ByteDance is also actively preparing. Sources say Seedance 2.1 is already in development and may debut alongside Veo 4’s release.
Regardless of the final outcome, for content creators, this is the best of times—tools are evolving, barriers are lowering, and possibilities are expanding infinitely.
Get Started with Seedance 2.0
Want to experience the power of Seedance 2.0 for yourself? Click the button below to start creating right away.