Gemini Omni 1.1 Flash shifts video editing to a stable API surface

Google's August 27 Gemini API update identifies `gemini-omni-1.1-flash` as the stable Gemini Omni Flash model for conversational video generation and editing. The endpoint documentation lists text, image, and video as inputs, video as output, a 1,048,576-token context window, and 3-to-10-second clips at 24 FPS. This is a concrete API-surface change for teams building video workflows, rather than a broad claim of a new general-purpose language model.

The practical delta from the preview surface is specific. Google's changelog documents video extension, first-and-last-frame interpolation with up to two images, and a `resolution` control for 360p, 720p, 1080p, and 4K output; it also says 1080p and 4K are produced through upscaling. The same changelog sets September 30, 2026 as the deprecation date for `gemini-omni-flash-preview`, so integrations using that identifier have a defined migration window rather than an open-ended preview period.

The documentation also draws an important scope boundary. Google's model card describes the wider Gemini Omni Flash family, including text, image, audio, and video inputs and video with audio output, while the endpoint-specific 1.1 table lists only text, image, and video inputs. That difference does not establish that audio is unavailable everywhere, but it does mean developers should treat the endpoint table as the safer contract for a production integration until Google documents audio behavior for this exact model identifier.

For technical users, the next step is a focused migration test: switch fixtures to the stable identifier, test extension and interpolation separately, and validate the actual delivery format at each requested resolution. The source material confirms the API controls and preview retirement date, but it does not provide independent quality comparisons, throughput commitments, or cost guidance. Those remain deployment questions that need measurement against a team's own prompts, media assets, and latency budget.