On May 19, Google announced Gemini Omni Flash at its I/O developer conference. The model, the first in a new multimodal video family, began rolling out that day to paying Gemini subscribers and to YouTube creators at no cost, replacing Veo inside the Gemini app and extending AI video editing into Google Flow and YouTube Shorts.
The release signals a shift from one-shot generation to multi-turn, conversational editing that responds to natural-language instructions. By embedding the capability directly in the assistant and Google’s creator toolchain, the company pushes generative video into the workflows of millions of consumers and youtubers, though full developer access requires an API that has not yet launched. It reflects a strategy to make generative video a routine tool inside Google’s most widely used services, rather than an isolated demo.
Gemini Omni Flash accepts text, images, video, and voice references as inputs. In demonstrations, Google showed how a user could upload a video clip and ask the model to alter objects, change scenes, adjust camera angles, or switch visual styles across multiple turns—iterative editing within a single dialogue, rather than a series of separate prompts. The company described this conversational flow as the core differentiator.
The rollout is global for Google AI Plus, Pro, and Ultra subscribers through the Gemini app and Google Flow. The same week, free access became available to users 18 and older with a Google account via YouTube Shorts Remix and YouTube Create. Google did not immediately publish a list of supported languages or countries beyond the broad “global” label. Developer and enterprise API keys are expected in the coming weeks; at launch, no programmatic access exists.
All Omni-generated videos carry SynthID digital watermarking and C2PA Content Credentials, a metadata standard designed to trace media provenance. Verification is possible through the Gemini app, and Google said that Chrome and Search support is available or coming soon. The company framed these as transparency tools, not as safeguards that prevent deepfake creation.
Omni Flash’s audio capability is narrow: only voice references are supported initially, with broader audio input planned later. Media reports and Google’s own demonstrations showed 10-second clips, but the company did not publish a universal technical limit. Independent performance benchmarks are absent; VentureBeat reported that Google did not issue public comparisons at launch. Competitors such as OpenAI’s Sora occupy the same space, but side-by-side quality data are not yet available. The model’s resolution, frame rate, and generation caps also remain unspecified. The lack of a live API means enterprise customers cannot yet evaluate the model on their own workloads.
Google described Omni as the successor to Veo inside the Gemini app, though the company did not say Veo is being retired from all Google surfaces. Veo had been the video generation model inside the assistant since late 2024; Omni now takes over that role, shifting the interaction from a single prompt to a conversation. For YouTube, the move puts AI-aided creation directly into two built-in editing tools, lowering the barrier for Shorts content. The launch comes as several AI labs race to embed video generation into everyday applications, though without independent benchmarks, any advantage remains a vendor claim. The launch also marks a deeper embedding of generative AI into Google’s subscription ecosystem, where video joins text and image capabilities.
With API access slated for the coming weeks, the commercial product will soon face the demands of developers and enterprise pricing. For now, the limited rollout tests whether conversational video editing can move from a laboratory promise to a utility embedded in Google’s paid subscription ecosystem. The controlled release also gives Google a chance to gather user feedback before broader distribution.