Gemini Omni Flash: Google's AI Video Model, Explained

Google now has two AI video models, and it wants you to reach for the newer one first. Gemini Omni Flash is the model behind AI video in the Gemini app, and since June 30, 2026 it is available to developers through the Gemini API. It makes short clips from a prompt, an image, or another video, and it can edit those clips through conversation: you refine a clip by talking to it.

Gemini Omni Flash is a generative video model. You give it a text prompt, and optionally an image or a short video to steer it, and it returns an MP4 clip. It is the first model in Google's Gemini Omni family, and its model ID in the API is gemini-omni-flash-preview. The consumer version arrived first, on May 19, 2026 at Google I/O, and the developer preview followed on June 30, 2026 alongside a new image model called Nano Banana 2 Lite (gemini-3.1-flash-lite-image), the budget sibling to Google's Nano Banana Pro image model.

On paper it looks like any other text-to-video tool. What sets it apart is the editing that comes after the first render.

Most video generators are one-shot. You write a prompt, you get a clip, and if it is wrong you regenerate from scratch and lose whatever was already good. Omni Flash works differently. Through its Interactions API you pass the ID of a clip you already made, describe what you want to change, and the model applies the edit while keeping the rest. Google's docs put it plainly: describe what you want to change, and the model applies the edit while preserving the parts of the video you want to keep.

The limit is three. You can stack up to three sequential edits on a clip before you start a fresh generation. It is the feature Google leans on to explain why Omni Flash is now its default video model, and it is the one thing here that older one-shot generators cannot do.

The no-code way is the fastest way to try it. Omni Flash powers AI video inside the Gemini app and Google Flow for Google AI Plus, Pro, and Ultra subscribers, and it is free for users 18 and older in YouTube Shorts Remix and the YouTube Create app. You type a prompt, generate a clip, then tell it what to change.

The developer path runs through the Gemini API and Google AI Studio.

One thing to know going in: system instructions, temperature, and top_p are not supported on this model, so you steer it with the prompt itself rather than sampling settings.

• Length and resolution. Clips top out at 10 seconds and 720p at 24 fps. Google frames the 10 second cap as a current choice that may grow later.
• No audio input or voice editing. The current preview does not accept audio references and cannot edit voice. Google's docs state that uploading audio references is unsupported in the current version. The wider Omni family is pitched as multimodal with audio, but that is not in this preview, so you bring your own sound.
• Editing is bounded. You get up to three sequential edits, and character consistency across scene changes or panning movement has limits.
• Some inputs are not handled. Scene extension, video extension, and multi-video referencing are unsupported, and video references of about 3 seconds are not processed correctly yet.
• Regional limits. Editing uploaded videos is not available for users in the European Economic Area, Switzerland, and the United Kingdom, though editing model-generated video is.
• Watermarked and preview only. Every clip carries a SynthID watermark with no opt-out, English is the only fully evaluated language, and the model ID ends in -preview, so behavior can change.

Google's own guidance settles most of this. Its video docs say to use Gemini Omni Flash as your default model for video generation, citing coherence, multi-input reasoning, and conversational editing, and to use Veo 3.1 for scene extension, last-frame control, or integration with legacy pipelines. Both run $0.10 per second at the fast tier, so the choice is about the task, not the price.

If you are choosing a video tool more broadly, our AI video generator comparison covers the wider field, and our Google Veo 3 guide walks through Veo end to end.

Omni Flash bills at $0.10 per second of video output, so a full 10 second clip is about $1. That is the same rate Google lists for Veo 3.1 Fast. One thing to plan for: each edit produces new video, so budget every edit as more seconds of output. On the consumer side there is no per-clip charge, the free access sits in YouTube Shorts Remix and the YouTube Create app, and the Gemini app and Google Flow include it with a paid Google AI plan.

Gemini Omni Flash fits three jobs today: short social clips, quick prototyping and storyboards, and iterating on footage you already have through conversational editing. If you need clips longer than 10 seconds, higher than 720p, or with generated audio, this preview is not the tool, and Veo 3.1 or another generator is the better call. It is also a public preview, so the specs and availability here can move. Treat it as a fast, cheap way to make and refine short clips, and check Google's docs before you wire it into anything that has to keep working.

• Google: Generate and edit videos with Gemini Omni Flash
• Google: Gemini Omni Flash model page
• Google: Video generation (Omni Flash vs Veo 3.1 guidance)
• Google blog: Start building with Nano Banana 2 Lite and Gemini Omni Flash
• Google: I/O 2026 announcements (consumer launch)

What is Gemini Omni Flash?

Gemini Omni Flash is Google's AI model for generating and editing short video from text, image, and video inputs. Its model ID is gemini-omni-flash-preview. It reached the Gemini API in public preview on June 30, 2026, and a consumer version launched on May 19, 2026 at Google I/O through the Gemini app, Google Flow, and YouTube Shorts.

Is Gemini Omni Flash free?

Through the Gemini API it costs $0.10 per second of video output, which is about $1 for a full 10 second clip. On consumer surfaces it is free for users 18 and older in YouTube Shorts Remix and the YouTube Create app, and it is available inside the Gemini app and Google Flow for Google AI Plus, Pro, and Ultra subscribers.

How long can Gemini Omni Flash videos be?

Clips run from 3 to 10 seconds, at 720p resolution and 24 frames per second, in 16:9 or 9:16. Google describes the 10 second limit as a current choice rather than a hard ceiling, so longer durations may come later.

Does Gemini Omni Flash generate or edit audio?

Not in the current preview. Google's documentation states that uploading audio references is unsupported in the current version of the API, and there is no voice editing. The wider Gemini Omni family is described as multimodal with audio, but that is not part of this preview, so you add your own sound.

Is Gemini Omni Flash better than Veo 3.1?

Google's video documentation makes Omni Flash the default model for video generation, citing its coherence, multi-input reasoning, and conversational editing, and reserves Veo 3.1 for scene extension, last-frame control, and legacy pipelines. So it depends on the task: Omni Flash for editing and iteration, Veo 3.1 for those specific one-shot jobs. Both cost $0.10 per second at the fast tier.

Written by

Paras Tiwari

Founder, Spectrum AI Labs

Founder of Spectrum AI Labs — testing AI tools and models, and writing up what actually ships.

More about Paras →