Video2 min17 November 2026

    Gemini Omni 1.2: more control, longer videos, and better handling of references

    Gemini Omni 1.2: more control, longer videos, and better handling of references

    Google has introduced Gemini Omni 1.2 Flash, an updated model for AI-powered video generation and editing. The main focus of this release is not simply generating videos from text prompts, but providing more precise control over the output, maintaining details across frames, and editing existing videos without having to start the generation process from scratch every time.

    Videos Up to 15 Seconds Long

    One of the major updates is the increase in maximum video duration from 10 to 15 seconds. The model can now generate videos ranging from 3 to 15 seconds, with one-second increments.

    This opens up more possibilities for creating advertisements, short stories, product presentations, and social media content. A single generation can now deliver a more complex scene with evolving action, camera movement, and a complete ending, without splitting it into multiple separate clips.

    Improved Reference Handling

    Gemini Omni 1.2 follows visual references more accurately. The model can use images of characters, products, and visual styles, short video clips of up to three seconds, as well as audio prompts.

    One particularly important improvement is the ability to preserve details throughout the entire video. Character appearances, clothing, product shapes, packaging text, and brand colors are designed to remain more consistent, even when camera angles change and the camera moves.

    This is especially useful for advertisements, virtual try-ons, product demonstrations, and videos where maintaining the recognizable appearance of a specific object is essential.

    Smoother Transitions Between the First and Last Frames

    The model has also improved its handling of the first and last frames. You can provide an initial and a final image, and Gemini Omni 1.2 will generate the motion between them.

    Rather than simply transitioning between two images, the model attempts to create a continuous scene with natural camera movement while preserving the main objects. This makes it possible to create cinematic zoom-ins, zoom-outs, transitions between locations, and more complex visual transformations.

    Conversational Video Editing

    Another important feature is multi-step editing. After the initial generation, you can continue working with the same video using text instructions.

    For example, you can change the background, lighting, weather, individual objects, or aspect ratio. There is no need to regenerate the entire video each time: the model attempts to preserve the elements of the scene that have already been created.

    Speech synchronization with lip movements, dialogue naturalness, and handling multiple voices within a single scene have also been improved.

    Quality, Speed, and Cost

    Gemini Omni 1.2 supports draft generation at 360p and 720p resolutions, with the option to upscale the results to 1080p or 4K.

    Conclusion

    Gemini Omni 1.2 Flash is an update that makes AI video generation more controllable. Longer videos, more accurate reference adherence, smoother transitions between the first and last frames, and conversational editing provide greater control over the final result.

    The model is particularly interesting for creating advertising content, video templates, and short-form videos where scene continuity, character consistency, and detail accuracy are essential.

    eng.png