คู่แข่งที่แข็งแกร่งที่สุดของ Seedance 2.0 ปรากฏตัวแล้ว
A major leak hit the AI video generation space recently. Google is reportedly testing a new video generation model codenamed Gemini Omni, and a user accidentally triggered access inside the Gemini app. Based on the leaked test results, Gemini Omni performs remarkably well in several key scenarios and could be the strongest rival Seedance 2.0 has faced so far.
What Is Gemini Omni
According to leaked screenshots, Gemini Omni is described as Google’s “new video generation model.” It supports not only generating videos from scratch but also video remixing, conversational editing, and template-based creation. In short, it aims to let users modify video content through chat-like interactions, not just “generate a clip.”
Google has not officially released it yet, and the leaked test access was quickly revoked. So the following comparison is mainly based on test clips that have surfaced online.
Test 1: Complex Math Blackboard Text Rendering
One long-standing pain point in AI video generation is text coherence. Many models produce blackboards or sheets of paper with garbled text, or the writing doesn’t match across shots.
In the leaked test, someone used a one-sentence prompt to generate a 10-second video with Gemini Omni: a professor writes a trigonometric identity proof on a blackboard and explains the current step.
The result showed that the mathematical formulas on the blackboard were generally readable and remained relatively stable throughout camera movement. Although careful inspection still reveals some AI artifacts, the improvement over previous models is clear.
For a fair comparison, we fed the same prompt to Seedance 2.0. Seedance 2.0 still excels in photorealism and lighting, but the mathematical accuracy of the board content was slightly off, with some text rendering remaining blurry.
| Comparison Item | Gemini Omni (Leaked) | Seedance 2.0 |
|---|---|---|
| Visual Realism | Excellent | Excellent |
| Text Readability | Good, formulas mostly legible | Average, some text blurry |
| Camera Stability | Stable | Stable |
This test shows that text rendering is becoming a new battleground among top-tier models. For instructional videos that need to display text, formulas, or interface operations, this capability is particularly critical.
Test 2: “Will Smith Eating Spaghetti”
This prompt has become a standard benchmark for almost every video generation model. The test focuses on whether character movements are natural, whether object physics are correct, and whether the eating process is coherent.
In the Gemini Omni-generated version, the character appearance and scene atmosphere were quite realistic. However, some viewers noticed a detail issue: there was no spaghetti on the plate before the character sat down, but it appeared after sitting; then during eating, the noodles inexplicably vanished. This indicates the model still has gaps in long-chain physical logic.
We entered the same complex prompt into Seedance 2.0. In terms of eating action naturalness, Seedance 2.0 performed more smoothly. The chewing rhythm, hand movements, and food interaction were more consistent with real-world physics.
| Comparison Item | Gemini Omni (Leaked) | Seedance 2.0 |
|---|---|---|
| Character Appearance | Photorealistic | Photorealistic |
| Scene Atmosphere | Strong upscale-restaurant feel | Excellent lighting and texture |
| Eating Action Coherence | Discontinuous, physics need work | More natural, complete action chain |
In this round, Seedance 2.0 edged ahead on action physics and logical coherence.
Gemini Omni’s Video Editing Capabilities
Beyond generation, Gemini Omni also demonstrated surprising video editing capabilities. For example:
- Replace the spaghetti in a video directly with a bowl of Thai tom yum soup;
- Remove a Sora watermark from a video while keeping the footage consistent.
If this capability remains intact at official launch, its practical value will be enormous. Currently, most AI video tools make secondary editing difficult after generation. Gemini Omni’s “conversational editing” approach could change that status quo.
AI Video Is Going Mainstream
Regardless of Gemini Omni’s final performance, one trend is clear: the quality of AI video is rapidly approaching professional production standards.
Recently, several shorts created with Seedance 2.0 have gained massive attention on social platforms. For example, an AI short dubbed “China’s Love, Death & Robots” was completed by its creator in just 10 days and has already received collaboration offers from film studios. A pigeon video made by Runway technical artist Marko Slavnic using Seedance 2.0 racked up millions of views in a short time, with many commenters saying they “couldn’t tell it was AI at all.”
These cases show that AI video is no longer a toy—it’s becoming a real creative tool.
Final Thoughts
The accidental leak of Gemini Omni has added more fuel to the AI video generation race. Google may officially unveil the product at next week’s I/O conference, and its real-world performance will be worth watching.
For everyday users, competition between models is good news—generation quality is improving, barriers to entry are dropping, and the choice of tools is expanding. Seedance 2.0 remains in the top tier for visual quality, action physics, and creative workflow. If you want to experience the current state-of-the-art in AI video generation, you can start using it right away.
Related searches: Seedance 2.0, Seedance tutorial, Seedance Chinese version, AI video generation comparison, Gemini Omni.