Google: Gemini Omni Flash brings native video generation from mixed inputs

At the Google I/O 2026 conference, Google officially launched Gemini Omni Flash, the first model from the new Omni family that natively generates and edits video content from mixed inputs. This is a significant step forward: the model does not simply accept text instructions, but simultaneously processes a combination of images, audio recordings, video clips, and text to create new video material or modify existing content.

What does “native video generation” mean?

Previous generative models mostly worked with a single type of input — text-to-video or image-to-video pipelines. Gemini Omni Flash introduces a genuinely multimodal approach: a user can simultaneously attach a reference image, an audio clip, and a short video, then describe the desired result in natural language. The model internally integrates all these signals and generates an output video that respects the style, motion, and context from each source.

This capability is especially powerful for iterative editing — the user can refine the result through multiple conversational turns without re-describing the scene from scratch. The model retains context across multiple revisions and consistently applies physical laws such as gravity, kinetic energy, and fluid dynamics.

SynthID: every generated video carries a digital watermark

The key safety component of Omni Flash is Google SynthID — an imperceptible digital watermark embedded in every generated clip. The watermark is neither visible to the naked eye nor audible, but can be verified through the Gemini app, the Chrome browser, and Google Search.

This mechanism directly addresses growing regulatory requirements around labeling AI-generated content — particularly relevant given the EU AI Act, which from August 2026 requires transparent labeling of synthetic media.

Availability: YouTube Shorts from day one

Google immediately integrated Omni Flash into YouTube Shorts and the YouTube Create app at no additional cost, meaning hundreds of millions of users today have access to native AI video generation directly within the platform. This is the broadest initial rollout of any Google generative model.

For advanced users, the model is also available through Google Flow and the Google AI Plus, Pro, and Ultra subscription tiers via the Gemini app. Developer and enterprise APIs are announced for the coming weeks, which will open up integrations in custom applications and production pipelines.

What comes next for the Omni family?

Google announced that Omni Flash currently supports audio references as the primary audio input, while other audio output types are marked as “coming soon”. Long-term, the Omni family is expected to expand to support direct audio and image output as well — which would position the model as a universal multimodal creative tool within Google’s ecosystem.

Also noteworthy is the model’s support for creating digital avatars and referencing the style, motion, and effects from attached material — opening possibilities for personalized video production at a scale previously unavailable to ordinary users. For content creators on YouTube and short-form platforms, Omni Flash could become a core daily-workflow tool as soon as this week.