Gemini Omni Flash 1.1 Adds Start-End Frames and Video Extension
ElevenLabs presents Gemini Omni Flash 1.1 as a control-focused update to Google’s conversational video model, adding fixed start and end frames, a separate mode for extending uploaded footage, and lower-cost 360p testing. The company cautions that 1080p and 4K are upscaled exports from native 720p generations, while seamless extensions require users to select the dedicated Extend workflow and explicitly instruct the model to preserve continuity.
Video extension is a separate workflow, and continuity has to be requested
Gemini Omni Flash 1.1 adds a dedicated Video Extend mode for continuing an existing scene. A user can upload a clip up to 30 seconds long, add a prompt, and generate an additional three to 10 seconds. The model uses the full uploaded clip, or however much of it is supplied, as context for the continuation.
In ElevenCreative, extension is not activated through the standard Gemini Omni Flash 1.1 selection. Users must open the model picker and switch to “Gemini Omni Flash 1.1 Extend,” then upload the source video, select an extension length, and provide a prompt. The picker describes the Extend option as continuing an existing video from where it ends, up to 40 seconds; the standard model is listed separately with image and video references and exports up to 4K.
That distinction is operationally important: choosing the regular Omni Flash 1.1 model does not open the video-extension workflow.
Testing identified a potential failure at the boundary between the supplied footage and its continuation. When speech appears in the final second of the uploaded clip, the extension can slightly hallucinate or repeat those last words. ElevenLabs suggests that this may happen because the system regenerates part of the original ending to merge it with the new footage.
Continuity also needs to be made explicit in the prompt. ElevenLabs advises specifying that there should be no cuts and asking for a “seamless extension” or “seamless transition.” One example begins, “Seamless continuation of the same shot, no cuts,” and goes on to preserve the lateral dolly movement, a 50mm pace, long shadows, drifting leaves, and the characters’ next action. The instruction is not merely asking for more video; it tells the model which compositional and editing conditions must survive the handoff.
The higher-resolution exports are still 720p generations
Gemini Omni Flash 1.1 adds 360p, 1080p, and 4K output options, but its 1080p and 4K settings should not be read as native-generation resolutions. Google’s displayed wording is: “export in 1080p and 4K, upscale your 720p generations.”
The model generates video at 720p, then Google’s internal upscaler produces the larger export. ElevenLabs compares the arrangement to Flux 3, which also generates at a lower resolution before upscaling rather than producing its highest-resolution output natively.
The practical distinction is between image generation and delivery format. Selecting 1080p or 4K changes the exported file’s resolution, not the resolution at which Omni Flash creates the underlying video. ElevenLabs reports that the upscaled results have been “pretty good” in testing—good enough, in its view, that a user who did not know the backend process would probably not notice.
At the other end of the range, 360p is intended for testing prompts and experimenting at a lower credit cost before committing to a higher-resolution export.
Start and end frames give the prompt a constrained journey
The central control change is the ability to set a specific start frame, a specific end frame, or both. Users can provide two still images and direct the generated transition between them, rather than leaving the clip’s opening and ending conditions unspecified.
The frames establish the endpoints; the prompt specifies what happens between them. A creator can decide exactly how the video begins and finishes, then describe the action, movement, or transformation that bridges those conditions.
That fills a gap in Omni Flash’s earlier workflow. ElevenLabs describes the prior ceiling as 10-second, 720p generations: users could supply images and videos as references, but could not pin exact initial and final frames.
The feature is not new to video generation generally. ElevenLabs notes that start-and-end-frame controls are already common across most video-generation models, including Google’s earlier Veo 3 models. For Omni Flash, this is a catch-up feature rather than a new category of control.
The release changes control points, not the basic model experience
Google has said Omni Flash 1.1 is better at cinematic generations and kinetic typography. ElevenLabs nonetheless characterizes the release primarily as a feature update rather than a broader reinvention of the model.
The basic experience remains conversational editing with 10-second standard clips, the same reference limits, and the same physics behavior. What changes is the degree of control around that workflow: lower-cost 360p testing, upscaled 1080p and 4K exports, defined start-to-end transitions, and a separate mode for continuing an existing video.