Announced at Google I/O 2026, Gemini Omni Flash is the first model in Google's Omni family — a multimodal generator that takes text, images, audio, and video as input and outputs cinematic video.
Stronger physics, better scene consistency, and conversational editing in a single model.
Released May 2026 · Multimodal input · Conversational refinement · Real-world physics
● Overview
Gemini Omni Flash is the production-tier video model in Google's new Omni family. It supersedes Veo 3.1 as Google's flagship video generator and brings three core advances.
Gemini Omni Flash treats text, images, audio, and video as equal input modalities. Earlier video models were primarily text-driven with images bolted on; Omni Flash routes all four through the same architecture, so reference media actually steers the result instead of being a hint.
Google has measurably improved how the model simulates gravity, kinetic energy, and fluid dynamics. Falling objects fall correctly. Liquids pour correctly. Cloth and hair move with realistic weight. This is the difference between "AI-generated" looking and "actually plausible" looking output.
Once a clip exists, you can refine it by replying in natural language. The model remembers the scene across turns — characters stay consistent, lighting holds, composition is preserved. You're editing the same scene, not regenerating from a new prompt.
Because Omni Flash inherits Gemini's broader knowledge of history, science, and culture, prompts about real places, eras, and phenomena produce visuals that look researched. Asking for "a Tokyo izakaya in the 1980s" or "a hurricane viewed from low orbit" returns visuals closer to reality.
● Access
Google ships Gemini Omni Flash through several channels. Each has trade-offs.
What the model can do today, based on Google's launch disclosures.
Flash-tier clips are capped at 10 seconds at launch. This is a deployment decision by Google, not a model limit — extensions are expected.
Text + image + audio + video as input modalities, in any combination.
Output is high-resolution and suitable for ads, social, and professional distribution. Exact specs vary by plan and surface.
Every clip carries Google's imperceptible SynthID watermark — verifiable as AI-generated through Google's tools, but not a visible mark on the output.
Reply with natural-language edits to refine a clip without losing scene state.
Google announced an avatar capability — generating videos that look and sound like the user — but is holding back its broadest rollout pending safety review.
● Getting started
Three steps from sign-up to first clip.
No Google account juggling. Email sign-up takes 30 seconds and you land with starter credits.
Type a prompt, drop an image, or do both. Mix inputs — Gemini Omni Flash treats them all as steering signals.
Reply with conversational edits to iterate. Export as MP4 in your target aspect ratio, watermark-free, commercial rights included.
● FAQ
Common questions about Google's Gemini Omni Flash video model.
No Google subscription. No waitlist. Sign up with email and generate your first clip in under a minute.
