A few days ago, before Ox Alpha was revealed, the Google DeepMind Gemini team was so hot on X that it was thought that Ox Alpha was a new version of Gemini.
It turns out that the cattle were the Chinese model GLM-5.3-Flash.
Logan Kilpatrick later clarified that the comments that have led to speculation refer to the release of Gemini 3.7 Flash.
Google recently released a new video model: Gemini Omni 1.1 Flash.
While the Gemini Omni 1.1 Flash is generally hot in the shadow of Sunrise, the model is still pretty bullish from the standpoint of the list, and it's worth paying attention to.
In summary, Omni 1.1 Flash focuses on the process of video production.Continuity, controllability and iterative efficiency.
First of all,Scene extension capabilitiesOmni 1.1 Flash can now read the context of a pre-order video of up to 10 seconds and continue to generate subsequent images on this basis. It can be extended for up to 10 seconds, with a cumulative video length of up to 40 seconds.
With increased context length, the model can obtain more character status, scene relationships, camera movement and scenario information. This ability directly affects the consistency of the film for continuous contrast, long footage, narrative video and video projects that require multiple sections to be generated.
In the example shown by Google, the same scene can be used to continuously complete conversations, retracing, and spatial transformations, while maintaining a relatively stable relationship between character and environment. Creators can also continue to generate different lens designs from the same material, such as push-track focusing, rapid focusing, and around the lens.
Secondly, Omni 1.1 Flash is added.Head and tail controlThe creator can specify the start and end images of a lens, complemented by a continuous motion in the middle of the model. This ability applies to scenes such as surrounding the lens, pulling the lens, rapid rotation, and looping video, and further enhances the compilability of video generated in the lens language.
In terms of production efficiency, Google has provided Omni 1.1 Flash.360p Drafting ModeOfficial data shows that the system throughput in 360p mode is up to about 60 percent higher than 720p, and the production cost is about a third of 720p.
This model is better suited to the early stages of creation. AI video production typically requires repeated attempts at cue words, camera movements, character movements, and composition. Creators can first batch low-resolution versions, select directions from multiple programs, and then enter the high-resolution production phase.
In the presentation, Google demonstrates a set of workflows similar to the "Draft Room": multiple 360p versions are generated at a time, with each version adjusting a variable, and parallel comparisons made.
In terms of video output, Omni 1.1 Flash supports upgrading videos to1080p and 4KThe addition of 4K resolution enables the model to cover more full-film delivery scenarios, including advertising materials, brand videos, product presentations, video previews and high-resolution social media content.
Multimodal reference capacityIt is also an important part of this upgrade. Users can add up to 3 seconds of video references to the input, allowing the model to extract information such as movement, rhythm, visual relationships, and then combine images and text to generate new scenarios. For example, developers can use different dance videos as an action reference, and then map the action to specified characters and scenarios.
In terms of overall positioning, Gemini Omni 1.1 Flash is gradually forming a set of basic video generation capabilities for developers and creative tool manufacturers. Scenes extend to solve continuous content generation, end-user provides lens-level control, 360p draft mode enhances trial and error efficiency, 4K output assumes film delivery, and video reference inputs further enhance material replication and action migration capabilities.
At the moment, Omni 1.1 Flash is open to developers for access to the Gemini API and Google AI Studio, which can be used by enterprise customers through the Agent Platform API; while the consumer terminal covers the global Google AI Plus, Pro and Ultra users, with related capabilities available in Google Flow and Gemini App.
We also experienced the Gemini Omni 1.1 Flash on the Gemini website.
Here's a snapshot of the fountain with Hawthorne's version of the fountain, along with a very short tip, creating a "Three-Animal Battle Falls":
One thing is for sure, Flash is really fast. The effect is still good, but the video content does not strictly follow the reference graphics, losing the funny essence of the original "Bull Come".
Below we try to recreate the classic video clips of "The Chinese Power Buying Guinea", but unfortunately, whether based on reference graphics or purely visual video, Gemini refused to generate the guinea pigs and we don't know what strange guinea pigs we hit.
Next, we picked up a hot spot and intercepted the hot prose to generate a video and watch Gemini's live video performance:
Overall, the details are pretty good. Finally, we produced a video of human evolution, which is pretty good:
Have you experienced this new model and how does it work?
Also, when is the Gemini 3.5 Pro coming out?
