# Gemini Omni Flash Preview

Gemini Omni Flash Preview generates short video with audio from text or reference images, then lets you refine the result through conversational edits that keep everything you did not ask to change.

- **Model ID:** `google/gemini-omni-flash-preview`
- **Type:** chat
- **Providers:** vertex, google
- **Context window:** 1,000,000
- **Maximum output tokens:** 57,920
- **Pricing:** $1.50/1M input tokens, $9/1M output tokens
- **Canonical page:** https://vercel.com/ai-gateway/models/gemini-omni-flash-preview

## Supported parameters

Detailed capability metadata has not been reported for this model.

## Example

```ts
import { streamText } from 'ai'

const result = streamText({
  model: 'google/gemini-omni-flash-preview',
  prompt: 'Why is the sky blue?'
})
```

## About

Gemini Omni Flash Preview is the first model in Google's Omni family and reached public preview on June 30, 2026. It processes text, images, and video together and returns video with synchronized audio. Clips run 3 to 10 seconds at 720p, in landscape (16:9) or portrait (9:16), with landscape as the default. Google grounds Gemini Omni Flash Preview in Gemini's world knowledge, pairing an understanding of physics with knowledge of history, science, and cultural context.

Stateful editing is what separates Gemini Omni Flash Preview from a plain text-to-video model. Each turn carries the previous clip and its references forward, so you can generate a scene, restyle it, swap a subject, insert an object, or relight it across several turns without describing the whole shot again. You can also branch from any earlier version. Keep edit prompts short: `Make the phone invisible. Keep everything else the same` produces a cleaner result than a paragraph of instructions, because long prompts invite changes you did not ask for.

Reference images bind to roles. Mark an image as the starting frame or as a style or subject reference, and Gemini Omni Flash Preview either animates it directly or borrows its look. A `task` field accepts `text_to_video`, `image_to_video`, `reference_to_video`, and `edit` when you want to state intent rather than let the model infer it. Audio generates alongside the video, and you can describe the track you want, including background music, sound effects, and events timed to specific seconds.

Every clip Gemini Omni Flash Preview produces carries SynthID watermarking, invisible to viewers but detectable for provenance checks, and C2PA content credentials. Rates for Gemini Omni Flash Preview are listed on this page as N/A, and N/A breaks out the resolution and duration tiers. Clip length is the main cost lever, since billing scales with the video you generate. Calling Gemini Omni Flash Preview through AI Gateway adds usage and cost tracking, automatic retries, and provider failover on one API surface.

## What to consider

Gemini Omni Flash Preview is a preview model, so behavior and supported parameters can change. Several controls that text models expose are unavailable: system instructions, `temperature`, `top_p`, stop sequences, and negative prompts. Put exclusions in the prompt text instead, for example `no dialogue`. Audio references are not accepted as input, video extension and frame interpolation are unsupported, voice editing is unsupported, and prompting across multiple videos degrades output. Editing video you uploaded is unavailable in the European Economic Area, Switzerland, and the United Kingdom, though editing video the model generated is supported. Validate the behavior your product depends on before it carries production traffic.

## When to use

### Best For

- **Conversational video editing** where iterative refinement across turns preserves the parts of a clip you did not mention
- **Short social and ad video** in portrait or landscape framing at clip lengths of 3 to 10 seconds
- **Product and asset animation** where reference images act as starting frames or style guides for motion
- **Style and subject swaps** that run as a follow-up turn instead of a full regeneration
- **Agentic media workflows** where generation and editing sit inside an application rather than a separate creative tool

### Consider Alternatives When

- You need higher-resolution output for a finished deliverable. `google/veo-3.1-generate-001` renders at up to 1080p
- Cost per clip caps your volume. `google/veo-3.1-lite-generate-001` targets high-volume pipelines
- You need still images. `google/gemini-3-pro-image` and `google/imagen-4.0-generate-001` return images rather than video
- You need a stable production surface. Preview status means parameters and behavior can change
- Your users edit uploaded footage in the European Economic Area, Switzerland, or the United Kingdom, where that path is unavailable

## Best for

- **Conversational Video Editing:** Iterative refinement across turns preserves the parts of a clip you did not mention
- **Short Social Video:** Clip lengths of 3 to 10 seconds in portrait or landscape suit social and ad placements
- **Product Asset Animation:** Reference images act as starting frames or style guides for generated motion
- **Style and Subject Swaps:** Restyling a scene or replacing a subject runs as a follow-up turn, not a regeneration
- **Agentic Media Workflows:** Generation and editing sit inside your application rather than a separate creative tool

## Consider alternatives

- **Higher Resolution Deliverables:** `google/veo-3.1-generate-001` renders at up to 1080p for finished output
- **High-Volume Generation:** `google/veo-3.1-lite-generate-001` targets cost-sensitive pipelines running at scale
- **Still Image Output:** `google/gemini-3-pro-image` and `google/imagen-4.0-generate-001` return images rather than video
- **Stable Production Surface:** Preview status means parameters and behavior can change during a rollout
- **Regional Editing Limits:** Uploaded-video editing is unavailable in the European Economic Area, Switzerland, and the United Kingdom

## Frequently asked questions

### How does Gemini Omni Flash Preview differ from the Veo models?

Gemini Omni Flash Preview generates video and then edits it conversationally across turns, carrying the previous clip and its references forward. Veo models generate a clip per request and render at higher resolution. Gemini Omni Flash Preview fits iterative short-form work; Veo fits finished deliverables.

### How does conversational editing work?

Each turn references the previous interaction, so the model retains the clip and its references without a re-upload. Send a short instruction such as `change the lighting to be more dramatic` and Gemini Omni Flash Preview applies it while preserving what you did not mention. Chain turns to branch between versions.

### How long are the clips and at what resolution?

3 to 10 seconds at 720p, in landscape (16:9) or portrait (9:16), with landscape as the default. See the Specs table on this page for the current list of supported options.

### Does Gemini Omni Flash Preview generate audio?

Yes. Audio generates alongside the video, and you can describe the track you want in the prompt, including background music, sound effects, and events timed to specific seconds. Audio references are not accepted as input, and voice editing is unsupported.

### What limitations should I plan around?

System instructions, `temperature`, `top_p`, stop sequences, and negative prompts are unsupported, so exclusions go in the prompt text. Video extension and frame interpolation are unsupported, and prompting across multiple videos degrades output. Editing uploaded video is unavailable in the European Economic Area, Switzerland, and the United Kingdom.

### How is Gemini Omni Flash Preview priced?

Rates for Gemini Omni Flash Preview are listed on this page as N/A, and N/A breaks out the resolution and duration tiers. Clip length is the lever you control, so shorter clips cost less. AI Gateway reflects provider pricing with no markup and charges no platform fee on inference.

### Are videos from Gemini Omni Flash Preview watermarked?

Yes. Every clip carries SynthID watermarking, which viewers cannot see but which can be detected programmatically for provenance verification, along with C2PA content credentials.

### Which prompt languages does Gemini Omni Flash Preview support?

English is fully supported. Google has not evaluated other languages, so they may work but results vary. Content safety filters apply to both prompts and generated video and depend on your region.

### How do I use Gemini Omni Flash Preview on AI Gateway?

Use the identifier `google/gemini-omni-flash-preview` with `generateVideo` from the AI SDK. AI Gateway handles provider routing and failover. Multi-turn editing depends on the stateful interaction flow, so check https://vercel.com/docs/ai-gateway/capabilities/video-generation for the parameters AI Gateway forwards.

### How does Zero Data Retention work with Gemini Omni Flash Preview through AI Gateway?

Yes, Zero Data Retention is available for this model. Zero Data Retention is offered on a per-provider basis. See https://vercel.com/docs/ai-gateway/capabilities/zdr for details.

## Links

- [Model page](https://vercel.com/ai-gateway/models/gemini-omni-flash-preview)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/omni-flash-preview)
- [Provider pricing](https://cloud.google.com/vertex-ai/generative-ai/pricing)