Google DeepMind has released Gemini Omni 1.1 Flash, updating its generative video model with expanded developer controls, multi-second temporal context for scene extensions, keyframe interpolation, and multi-tier resolution pricing. The release aims to transition video generation from single-prompt clips into programmable, multi-step production pipelines.
Temporal Scene Extension with 10-Second Context
A core challenge in generative video has been maintaining temporal consistency when extending video sequences. Previous iterations referenced only the final frame of an input clip, frequently leading to abrupt changes in character features, lighting, and environmental layout.
Gemini Omni 1.1 Flash expands this context window by analyzing up to 10 seconds of preceding video footage. The model allows developers to extend clips iteratively in 10-second increments, reaching a total cumulative sequence length of up to 40 seconds. By tracking motion vectors and subject positions over the entire 10-second buffer, the model preserves narrative continuity and visual coherence across extended shots.
Keyframe Interpolation and Multimodal Video References
Omni 1.1 Flash introduces explicit keyframe control. Developers can specify both starting and ending visual frames, prompting the model to generate continuous intermediate footage. This capability supports complex cinematic maneuvers, such as whip pans, continuous 360-degree orbital shots, and dolly zooms, without jump cuts or structural warping.
Additionally, the model accepts up to 3 seconds of external reference video alongside text and image prompts. Developers can reference specific motion patterns, character designs, or camera trajectories to condition the output without extensive prompt engineering.
Resolution Tiers and Serving Economics
To lower the cost of iterative testing and storyboard generation, Google added a 360p lightweight preview mode alongside standard and high-resolution upscaling options:
- 360p Draft: $0.03 per second of generated video (up to 60 percent faster generation speed for rapid iteration).
- 720p Standard: $0.10 per second.
- 1080p HD: $0.15 per second.
- 4K Upscaled: $0.30 per second.
The tiered architecture allows applications to generate multiple draft candidates in 360p before committing compute to final 1080p or 4K renders.
Availability and Ecosystem Integration
Gemini Omni 1.1 Flash is available through the Gemini API in Google AI Studio using the gemini-omni-1.1-flash model identifier, as well as on the Gemini Enterprise Agent Platform. Early production integrations include Adobe Firefly, Figma Weave, and Runway. End users can access the model via Google Flow and the consumer Gemini interface.
