Gemini Omni has no voice editing yet: keep your own voiceover

You cannot change the voice in a Gemini Omni clip today. Google's launch post says it is still testing how to bring editing of audio and speech in videos to users, and the Omni API guide lists voice editing and audio reference uploads as unsupported. The workable pattern is to treat Omni as the picture source and put a voice you recorded or licensed on top, using a timeline that takes your audio file as the spine. Sume's Timeline 1.0 does that assembly for $0.10 per output minute.

What Google says

Google's launch post says editing videos to change audio and speech is still being tested. The Omni guide lists voice editing and audio reference uploads among unsupported features, and also unsupported: system instructions, temperature, top_p, stop sequences, negative prompts, and YouTube videos as a source.

So with the API, the voice in a generation comes from the model and the prompt; the guide offers no audio reference to pin it to a recorded speaker.

The Sume assembly route

Generate the clips on the Video Router, then render with Timeline 1.0. The timeline takes audio.url, one gapless spine you provide, with audio.duration_seconds between 1 and 1800, plus ordered video[] slots. Audio and clips must be Sume-hosted, so import your voice file through POST /v1/media-imports first.

This route uses an audio file you supply, so bring a recording you have the rights to, or a voiceover you produced elsewhere.

Limits to plan around

The spine sets the output length, so cut your clips to fit it; coverage may trail the spine by at most 0.5 seconds. Check what happens to the original Omni audio in the render before you ship: the Timeline docs describe the audio spine, and they do not state whether source-clip audio is kept, so test one render and listen.

Run POST /v1/timeline-1.0/plan first; it is unbilled and returns the billable minutes, as covered in Jobs and results for the polling that follows a real render.