Gemini Omni 1.1 Flash: Boosting Video Continuity, Control, and Efficiency

After a brief mix‑up involving the Ox Alpha model, Google’s DeepMind team clarified that the earlier speculation referred to Gemini 3.7 Flash, and then, in the early hours of the same day, announced the Gemini Omni 1.1 Flash video model.

The update focuses on three core aspects: continuity, controllability, and iteration efficiency.

Extended Scene Continuity

Omni 1.1 Flash can ingest up to 10 seconds of preceding video context and continue generation from there, allowing a single generation to reach a total length of 40 seconds. The longer context provides richer information about characters, scene relationships, camera motion, and narrative flow, which improves consistency for dialogue‑heavy, long‑take, or multi‑segment videos. Google’s demo shows a single scene maintaining stable character‑environment relations while performing actions such as dialogue, camera pull‑back, and spatial transitions, and also generating alternative camera plans (track‑push, rapid zoom, orbit).

First‑Frame and Last‑Frame Control

Creators can specify the exact start and end frames of a shot, letting the model fill in the intermediate motion. This capability benefits orbit shots, push‑pull movements, rapid cuts, and looping videos, giving finer control over the generated camera language.

360p Draft Mode for Faster Iteration

The model introduces a 360p draft mode whose throughput is roughly 60 % higher than the 720p mode, and its generation cost is about one‑third of 720p. This mode is suited for early‑stage experimentation, allowing creators to batch‑generate low‑resolution drafts, compare variants side‑by‑side, and then upscale promising candidates to higher resolutions.

High‑Resolution Output

Final outputs can be rendered at 1080p or 4K. The 4K option expands use cases to advertising, brand videos, product showcases, film pre‑visualization, and high‑resolution social media content.

Multimodal Reference Capability

Users may provide up to a 3‑second video clip as reference input. The model extracts motion, rhythm, and visual relationships from the clip and combines them with image and text prompts to generate new scenes, enabling actions from one video to be transferred onto different characters or settings.

Availability

Gemini Omni 1.1 Flash is now open to developers through the Gemini API and Google AI Studio, with enterprise access via the Agent Platform API. Consumer‑facing services include Google AI Plus, Pro, and Ultra, and the capabilities are integrated into Google Flow and the Gemini App.

Hands‑On Impressions

In internal tests, the model produced a “Three Animals vs. Lü Bu” clip using a screenshot from the “Ox Alpha” demo and a still of He Run‑dong as Lü Bu. While the speed was noticeably fast and the overall quality acceptable, the generated video did not fully preserve the original comedic style. Attempts to recreate the classic “Hua Qiang buying melons” segment failed, with Gemini refusing to generate the content for unknown reasons. Additional experiments with a popular prose excerpt and a human‑evolution video demonstrated detailed output, though occasional mismatches with reference images were observed.

Overall, Gemini Omni 1.1 Flash positions itself as a foundational video‑generation capability for developers and creative‑tool vendors, addressing continuity, shot‑level control, rapid prototyping, high‑resolution delivery, and multimodal reference.