What is Gemini Omni?

Gemini Omni is Google's model family for generating and editing short videos with native audio. The current version, Gemini Omni 1.2 Flash, makes clips of 3 to 15 seconds per generation from text, images and video, and lets you refine or extend a clip over several conversational turns. Google's announcement showcases audio cues, but the current Gemini API does not accept uploaded audio or audio references and does not support voice editing. Developers use it on the paid tier of the Gemini API and on Google Cloud; Google also brings it to Flow, Google Vids, the Gemini app and Google Ads.

What you can make with it

Typical jobs are short and repeatable: a 10- to 15-second product spot from a few product photos, a social ad in several aspect ratios, a short explainer, or a virtual try-on clip that keeps one model and garment consistent. Reference images, up to six, help keep a character, product or style the same across shots.

Editing in conversation is what separates it from a one-shot generator. After a first clip you can ask for a different background, lighting, product or aspect ratio and keep the rest of the shot, according to Google. You can also extend a clip at the end by 3 to 10 seconds per turn, up to 30 seconds in total, so 15 seconds is the limit of one generation, not of the finished video.

What changed in version 1.2 Flash

Google announced Gemini Omni 1.2 Flash on November 5, 2026. Clips can now run up to 15 seconds in a single generation, up from 10 seconds in Omni 1.1, and you can choose a length from 3 to 15 seconds. Google says the model holds references more closely, drifts less between the first and last frame, produces more natural multi-speaker dialogue and lip-sync, and handles long camera moves with fewer unwanted cuts. The announcement also mentions audio cues as inputs across Google's products; in the current Gemini API, Google's documentation says uploading audio inputs or audio references is unsupported and voice editing is not supported.

Those quality claims and the example clips come from Google and its partners; AppFor.ai has not tested the model. Google itself says development continues on fast sports action, rapid script changes and complex physics. From December 3, 2026, API requests to gemini-omni-1.1-flash will be routed to 1.2 Flash and billed at its rates.

Where you can use it

Developers can prototype in Google AI Studio and call the model through the Gemini API, on the paid tier. Businesses can use it in the Google Cloud console, where Google says it is generally available with Provisioned Throughput. This Cloud offering is separate from the Gemini API, where Google's documentation says provisioned throughput is not supported.

In the EEA (including Poland), Switzerland and the UK, the API does not support editing or extending uploaded videos; model-generated videos can be edited or extended. Uploading or editing images containing minors is also unsupported there. For non-developers, Google lists Flow, Google Vids (15-second clips at no additional cost), the Gemini app (rolling out to Google AI Ultra subscribers) and Google Ads; availability of each consumer surface still needs checking in your account.

What a clip costs

On the API you pay per second of video output, converted from tokens. Google's pricing page lists 5,792 tokens per second of 720p video at $12 per million tokens, about $0.07 per second, so a 10-second draft costs about $0.70 and a 15-second clip about $1.04. It lists about $0.023 per second at 360p and, with upsampling, about $0.104 at 1080p and $0.209 at 4K. These figures cover video output only: input tokens (text, images, video) cost $1.50 per million and text output $9 per million on top. There is no free API tier for this model.

A practical way to control cost, which Google also suggests, is to draft at 360p or 720p, fix details with follow-up edits instead of full regenerations, and upsample only the final take. Prices are as listed on November 7, 2026 and may change.

A first workflow to try

Pick one real brief, such as a 10-second product clip from two product photos. Generate a 720p draft, then make two conversational edits, for example a new background and a different aspect ratio. Check whether faces, logos and on-screen text stay consistent, whether the generated audio matches the picture, and what the whole sequence cost including input tokens. If you are in the EEA, work from model-generated clips rather than uploaded footage. Outputs carry a SynthID watermark, so plan for AI disclosure where your platform requires it.

What stands out

Clips of 3 to 15 seconds per generation, with native audio output
API inputs: text, images and video; up to 6 reference images for consistency
Multi-turn editing and extension via the Interactions API, up to 30 seconds total
Draft at 360p or 720p, then upsample to 1080p or 4K
Imperceptible SynthID watermark on outputs

Before you choose

One generation is 15 seconds at most; extensions reach 30 seconds total.
In the EEA, Switzerland and the UK, the API cannot edit or extend uploaded videos.
No audio upload, audio references or voice editing in the current API.
Paid tier only; input tokens are billed on top of video output.
Quality claims and examples come from Google and partners, not independent tests.

Plans & pricing

Current pricing on ai.google.dev

Go to the source

Sources & further reading

Based on the official documentation below. Features and access can change; check the provider before making a decision.

Something out of date or incorrect?Report a correction

Was this page useful?

Help us make our coverage more useful. This is feedback on the page, not a rating of the tool.

Go a little further

Gemini Omni