Gemini Omni 1.1 Flash shifts video generation toward controllable iteration

Google has documented Gemini Omni 1.1 Flash as an update to its video-generation API, with controls for extending scenes, interpolating between a first and last frame, using reference video, and choosing output resolution. The published endpoint is `gemini-omni-1.1-flash`, and Google positions the model around text, image, audio, and video inputs rather than a text-only video prompt. The change is concrete for developers because the controls are expressed through the Interactions API and response-format parameters, not only through a consumer product interface.

The most consequential change is temporal context during scene extension. Google says Omni 1.1 can analyze up to 10 seconds of prior video context, whereas previous models referenced only the final second, and it permits 10-second extensions to a cumulative 40 seconds. That does not establish that continuity will be reliable for every subject or edit, but it does make the API better suited to iterative shot construction than a workflow that treats each generated clip as an isolated asset.

The documentation also makes the cost-quality tradeoff explicit. The API defaults to 720p, offers 360p and 720p generation, and lists 1080p and 4K as upscaled outputs; Google says 360p previews can be generated up to 60% faster and at one-third of the cost of its standard 720p output. It supports 9:16 and 16:9 aspect ratios, plus first-and-last-frame interpolation, so a product team can test composition and transitions before committing to a higher-resolution result.

For technical users, the practical pattern is to treat Omni 1.1 as a staged media pipeline: prototype with 360p, use frame or video references when continuity matters, then upscale selected outputs. The limitations matter as much as the new controls. Google's speed, cost, and production-readiness statements are provider claims, the higher resolutions are explicitly upscaled, and the cited material does not provide an independent comparison of visual consistency, latency, or failure rates against competing video models. Teams should therefore validate the controls on their own footage and prompts before relying on them in a production workflow.