Google extends Omni video generation to 15 seconds

Google’s announcement introduces Gemini Omni 1.2 Flash for generating and editing videos up to 15 seconds, compared with ten seconds previously. Creators can request successive changes to a background or lighting instead of describing the whole scene again. Google presents this as a way to make variants of business videos. These are the developer’s claims, rather than results of an independent test.

The API guide explains the mechanism: an edit refers to the earlier interaction’s identifier, and each turn produces a new video. Disabling storage prevents later edits through that identifier. Limits also depend on the task. Interpolating between a supplied first and last image supports only three to ten seconds. Editing or extending uploaded videos is unavailable in the European Economic Area, Switzerland and the United Kingdom; editing or extending videos generated by the model is supported there. The longer generation limit therefore does not describe every operation.

The model card describes video with audio as the output and identifies safeguards used during development and production. Google says it uses synthetic captions during training to improve how concepts are represented, production filters to prevent unwanted outputs, and SynthID watermarks to identify generated content. It also explicitly restricts the model’s capability to change people’s speech. Conversational editing should therefore not be read as unrestricted voice editing.

MarkTechPost’s coverage connects the release announcement with Google’s cloud announcement and developer documentation.