For its AI videos, Google is lengthening clips and lowering the price with Gemini Omni 1.2 Flash

To generate and edit videos, Gemini Omni 1.2 Flash goes up to fifteen seconds. The API price drops 30%, to $0.07 per second in 720p.

In a demonstration of Gemini Omni 1.2 Flash, two armored characters appear successively in a forest, a stable, a fair, and in front of a castle. The changes in setting are requested throughout the conversation, starting from the same clip. The result shown by Google is meant to illustrate a central capability of this version: reworking an existing video while preserving the scene one wants to keep.

The model announced by Google brings video generation and editing to fifteen seconds, up from ten for Omni 1.1 Flash. It also produces synchronized audio, with voices, music, and sound effects. The changes can involve a product, lighting, the background, or the image format. Fifteen seconds, with an exception for first and last images

The generation duration is set in one-second increments, between three and fifteen seconds. The horizontal 16:9 and vertical 9:16 formats are supported. A single request can therefore aim for a short product shot or a longer sequence, with several framing changes like in the example of a beehive rendered in four shots.

Google Cloud’s technical documentation nonetheless specifies that generation guided simultaneously by a first and a last image remains limited to between three and ten seconds. The fifteen-second cap therefore does not apply to all creation methods. Video extension remains a separate operation, with added segments of three to ten seconds.

Conversational edits can apply to a complete fifteen-second clip. The user keeps a video as a starting point and specifies the following modifications, without having to restate the entire scene in a new request. This can be used to prepare several variants of the same content, for example by changing the background or the product shown. References must survive camera movements

Fidelity to the provided elements plays an important role in this version. References can describe a character, a garment, a product, or a visual direction. Google announces better preservation of faces, textures, and text when the camera or the characters move.

One demonstration combines a model’s photo with front and back views of several outfits to produce a video try-on. Another starts from architectural references to create a continuous camera movement in the Pantheon. These examples show the results selected by the editor; they do not provide a success rate for a series of requests.

Dialogue is also part of the announced improvements, with more natural speech, more precise lip-sync, and better attribution of voices to characters. Audio references remain subject to prior authorization, as specified in the Google Cloud blog. Their use is therefore not open to all accounts.

Fast sports movements and complex physics are still among the development areas cited by Google. The continuity shown in the demonstrations does not mean that every movement or every interaction between objects will be accurately reproduced. Less-defined drafts before the final render

The announced API price drops to $0.07 per second of 720p video, or 30% less than for Omni 1.1 Flash. At that rounded rate, fifteen seconds amount to about $1.05 for the video output. This amount is not the full cost of a sequence of tests and revisions.

The Cloud pricing table distinguishes input data, text outputs, and generated video. Resolution upscaling also has its own billing. Each variant or rework must therefore be taken into account when calculating the cost of the final work.

The proposed workflow consists of preparing drafts in 360p or 720p, then upscaling the retained versions to 1080p or 4K. The model supports all four resolutions; going through a less-defined draft is a working option, not a requirement. Access depends on the application used

The launch covers Google AI Studio, the Gemini API, and Google Cloud, where the model is offered for production use. In Flow, the current rollout covers scene direction and successive edits. In the Gemini app, fifteen-second clips are announced for Google AI Ultra subscribers, with a gradual rollout.

Generation of fifteen-second clips is also coming to Google Vids, announced at no extra cost in the service’s access terms. This mention should not be confused with free API access. On Google Cloud, outputs carry invisible SynthID watermarking, and the service offers availability commitments as well as reserved processing capacity for enterprises.