Google Unveils Gemini Omni at I/O 2026: A "World Model" That Rewrites Video Editing

A World Model, Not a Clip Generator

Google unveiled Gemini Omni at I/O 2026 on May 19. DeepMind CEO Demis Hassabis described it as a "world model" that integrates Gemini with Veo, Nano Banana, and Genie. Unlike Sora, Runway, or the original Veo — which generate clips from text — Omni understands physical environments, predicts cause and effect, and applies edits while holding full scene context.

Conversational Editing Sets It Apart

The headline feature is conversational video editing. Users can say "change the background to a sunset" or "pull the camera left," and Omni applies the change while preserving characters, motion, and environment consistency. Persistent inconsistency after edits is a known weakness of current AI video tools; Gemini Omni addresses this by maintaining scene context across all modifications.

• Multimodal input: Processes text, audio, images, and video simultaneously
• Physical simulation: Environment understanding, not just style transfer
• Sequence consistency: Characters, backgrounds, and motion remain coherent across edits
• YouTube integration: YouTube Shorts and YouTube Create support rolling out this week at no added cost

Available Now, Replacing Veo

Gemini Omni Flash launched immediately for Google AI Plus, Pro, and Ultra subscribers through the Gemini app and Google Flow. It replaces Veo inside the Gemini app. Full Gemini Omni is targeted for later in 2026. In a crowded AI video market that includes OpenAI Sora, Runway, and Kling, Google's "world model" framing emphasizes understanding over generation — a distinction that will be tested as developers put it through its paces.

Source: Google Blog — Introducing Gemini Omni