Google announced the first product in its Gemini Omni series—Gemini Omni Flash—at Google I/O 2026 on May 19, and officially released its technical overview on May 22 on its official website. The initial integrated platforms include the Gemini app, Google Flow, and YouTube Shorts.
Confirmed Core Features of Gemini Omni Flash
Conversational video editing: Users edit videos with natural-language instructions, where each instruction builds on the previous one. The model maintains consistent character identity, reliable physical effects, and scene memory. It supports changing the background, style, angle, or specific details without needing to regenerate the entire clip.
Advanced physical-engine simulation: Omni’s intuitive understanding of gravity, momentum, and fluid dynamics strengthens the realism of scenes. It allows users to create more accurate physical effects, such as dynamic scenarios involving object collisions, liquid flow, and Chain Reaction, among others.
Multimodal input generation: Omni can process any combination of inputs (images, text, video snippets, audio) as a single instruction and generate unified output content. In the initial stage, audio input supports speech referencing; other audio input types will be introduced later.
Knowledge integration and conceptual visualization: Drawing on Gemini’s knowledge of historical, scientific, and cultural context, Omni goes beyond simple pattern matching. It can generate explanatory content based on brief prompts—for example, using a clay animation to explain complex scientific concepts like protein folding.
Digital virtual avatar (Avatar) functionality: Users can create a digital version that includes their own voice, generating videos whose appearance and voice both resemble the user. Audio and voice editing features are still in testing and have not yet been opened to all users.
SynthID watermark: Confirmed AI content transparency mechanism
All videos created through Gemini Omni automatically embed SynthID digital watermarks. This is an invisible watermark technology developed by Google DeepMind; after embedding, it does not affect the video’s visual quality. Users can verify whether a video was generated by Gemini Omni through three confirmed channels: the Gemini app, Gemini in the Chrome browser, and Google Search. Google says the SynthID verification tools are designed to help users understand how content is created and edited on the web, as part of its responsible AI development policy.
Confirmed access channels and release schedule
Available now: Google AI Plus, Pro, and Ultra paid subscription users, via the Gemini app and Google Flow
Within this week: YouTube Shorts and YouTube Create app users, free
Within the next few weeks: Developers and enterprise customers, via the Gemini API and Agent Platform API
Common Questions
What technical differences exist in the “world model” positioning of Gemini Omni Flash compared with general video-generation models?
Google positions Gemini Omni as a “world model,” meaning the model not only performs input-to-output generation mapping, but also has causal inference capabilities based on a real-world knowledge base trained through Gemini (including physical laws, cultural context, and historical and scientific knowledge). For example, it can predict how objects in a scene will behave next, apply real-physics-engine effects, and convert language descriptions into semantically meaningful visual content. This differs at the architectural level in design objectives from video diffusion models that are purely based on pattern matching.
Can the SynthID watermark be removed or bypassed?
Google’s official documentation confirms that the SynthID watermark is invisible (does not affect the video’s visual content) and is embedded in the video’s digital structure, which can be verified using Google’s official verification tools. Google has not disclosed the specific technical implementation of the watermark in official documents. As of now, there is no publicly available independent technical assessment record regarding SynthID’s reliability and tamper resistance.
What input formats does Gemini Omni Flash currently support, and what output types will be expanded in the future?
Confirmed input support: text, static images, video snippets, and voice audio (initial stage). Google confirmed on its official blog that other audio input types “will be added soon” as a supplement. In terms of output, the current Omni Flash version focuses on video outputs. Google says that in the future, the Omni series will support image and audio output modes, but the specific release timeline has not been confirmed in this announcement.
