Gemini Omni 1.1 Flash turns Google’s video API update into a migration deadline

Google’s August documentation identifies `gemini-omni-1.1-flash` as the stable Gemini API model code for conversational video generation and editing. The documented interface adds video extension, first-and-last-frame interpolation, selectable output resolution, and natural-language refinement through the Interactions API. That makes the change more concrete than a general quality claim: developers can target a named API model and use defined control surfaces for continuation, transitions, and iterative edits. Google’s product announcement also places the model in Google AI Studio and its enterprise agent platform, with consumer availability described separately for Flow and the Gemini app.

The immediate operational context is the replacement of the earlier preview path. Google’s API changelog says `gemini-omni-flash-preview` is scheduled for deprecation on September 30, 2026 and names `gemini-omni-1.1-flash` as its recommended replacement. The current model reference lists text, image, and video inputs, video output, a 1,048,576-token context window, and generated clips of three to ten seconds at 24 frames per second. Teams that integrated the preview endpoint should therefore treat this as an API migration and regression-testing task, rather than assuming a preview alias will continue to work unchanged.

The practical distinction from the preview generation path is control at several points in a workflow. The guide documents extension requests for continuing a clip, two-image interpolation for connecting a first and last frame, portrait or landscape aspect ratios, and a `previous_interaction_id` pattern for editing a prior result in a subsequent turn. Resolution is explicitly configurable from 360p through 4K, with 720p as the documented default. This favors product teams building editors, storyboarding tools, or review loops, where preserving a usable intermediate result and changing one instruction may matter more than one-shot prompt generation.

There are limits in the same documentation that should temper expectations. The model reference describes short generated outputs, and the guide says that 1080p and 4K are produced through upscaling rather than presenting them as native high-resolution generation. Google’s pages establish the available interface and the vendor’s intended workflow, but they do not provide an independent comparison of visual consistency, edit fidelity, latency, or cost against competing video systems. Technical users should test their own source material, extension boundaries, and upscale quality before moving production workloads, especially where continuity across several clips or exact frame control is a requirement.