Google added three specific editing controls to its generative video model through the Gemini API. Omni 1.1 Flash lets developers extend an existing clip from where it left off, generate footage that runs between a chosen first and last frame, and upscale a finished render up to 4K. Google published the update on August 27 on its developer blog.
The distinction matters more than the release headline suggests. A model that takes a text prompt and returns a finished clip is a generation tool: useful, but the output is whatever the system decided to produce. A model that accepts a starting frame and an ending frame, or continues a specific existing shot instead of starting over, behaves like an editing tool. The creator sets the boundary conditions and the model fills in what happens between them.
Scene extension shows the shift most clearly. Google says earlier versions of the model could only reference the final second of a clip when continuing it, which capped how coherent a longer sequence could look. Omni 1.1 Flash reads back further, up to ten seconds of preceding footage, and clips can now be built out in stages toward a combined length of forty seconds. That treats a video as something assembled in passes, not generated once and thrown away if it misses.
Keyframe interpolation runs the same logic in reverse. Rather than continuing forward, a developer supplies two static images, and the model fills the motion between them. Google positions this for camera orbits, zoom transitions and clean loops. Specifying start and end points to generate the motion in between is standard technique in traditional animation and editing software; it has mostly been missing from text-to-video tools, where a creator describes a scene and hopes the render matches the description.
A faster draft mode and the 4K upscale round out the release. Google says a lightweight 360p preview renders roughly 60 percent quicker and costs a third as much as the model’s standard 720p output, framed as a way to test a storyboard before committing to a full render. Two other stories in this issue, fal’s H3 Max model and an LMSYS inference benchmark, already cover raw generation speed in this category. The more durable news here is the direction of the controls, not the throughput figures Google is citing about its own system.
Google is also pointing to early adopters. Figma’s Weave and Adobe’s Firefly have both built Omni Flash into their products, and a Figma creative director is quoted in the post saying the tool moves teams from generating clips to actually directing them. That framing comes from Google and a named partner, not from independent testing, and the announcement includes no benchmark comparison against rival video models.
Access runs through Google AI Studio and the Gemini Enterprise Agent Platform for developers, with the consumer version reaching Google Flow and scene extension arriving in the Gemini app for Plus, Pro and Ultra subscribers. For product teams building on generative video, the test to run over the next quarter is not whether Omni 1.1 Flash renders sharper clips than a competing model. It is whether frame-level control lets them swap out a piece of an existing editing pipeline rather than just swapping the camera.
Google detailed the Gemini Omni 1.1 Flash update in an August 27 post on the Google blog, authored by DeepMind product managers Anish Nangia and Alisa Fortin.
