Gemini Omni 1.2 Flash now available on AI Gateway

Google's Gemini Omni 1.2 Flash is now available on AI Gateway. Developers can generate videos from text prompts with one API key and no separate Google account. Vercel Blog's report says the model generates 3–15 second videos with native audio and edits 3–15 second source clips with up to six reference images.

This piece is for developers calling models through AI Gateway and the AI SDK. It covers the clip lengths, reference-image cap, extension limits, resolutions, and aspect ratios Gemini Omni 1.2 Flash accepts, then the generateText and streamText calls that save an MP4. It also covers Gateway pricing, which reflects provider rates with no markup and no platform fee on inference, including Bring Your Own Key (BYOK) requests.

What shipped in Gemini Omni 1.2 Flash

Gemini Omni 1.2 Flash from Google is the model now listed on AI Gateway for text-to-video. A call generates a 3–15 second video with native audio. The same model edits a 3–15 second source clip and accepts up to six reference images on that edit. Callers use one API key and do not open a separate Google account. In the AI SDK, generateText or streamText receives the generated video as a file, and the sample looks for media type video/mp4 before it writes video.mp4.

Extend a video by 3–10 seconds per call, up to 30 seconds total. A follow-up call refines a generated clip when it passes previousInteractionId. Output supports 360p, 720p, 1080p, and 4K in 16:9 or 9:16. The published example sets the model to google/gemini-omni-1.2-flash, requests duration 3s at resolution 360p with type video, and writes the returned bytes to video.mp4. providerOptions.google on that call sets responseModalities to text and video. generateText returns the MP4 on files as uint8Array, which writeFile then stores.

What improved in Gemini Omni 1.2 Flash

AI Gateway reflects provider pricing for Gemini Omni 1.2 Flash with no markup, and it does not charge a platform fee on inference, including Bring Your Own Key (BYOK) requests. The model string is google/gemini-omni-1.2-flash. A text prompt returns a 3–15 second clip with native audio. An edit takes a 3–15 second source clip and up to six reference images. Each extend call adds 3–10 seconds, and the video can reach 30 seconds total. A later call passes previousInteractionId when it refines that clip.

AI Gateway is also the unified API that tracks usage and cost for this model and configures retries, failover, and performance optimizations for higher-than-provider uptime. Custom reporting, Zero Data Retention support, budgets for API keys, and routing rules sit on that same path. The sample call still asks for a 360p video lasting 3s and saves video.mp4 from the video/mp4 file. One API key covers the call, with no separate Google account. streamText delivers that video as a completed file on fullStream.

What you gain from Gemini Omni 1.2 Flash

Developers keep one API key and skip a separate Google account when they call google/gemini-omni-1.2-flash. The video comes back as a file from generateText or from streamText, with native audio on a 3–15 second generation. Usage and cost for that call are tracked on AI Gateway beside retries, failover, and the performance optimizations aimed at higher-than-provider uptime. Budgets for API keys, routing rules, custom reporting, and Zero Data Retention support are on the same Gateway. The sample prompt is "A red ball rolls across a white table."

The bill is provider pricing with no markup and with no platform fee on inference, including BYOK requests. A caller can request 360p, 720p, 1080p, or 4K in 16:9 or 9:16. An edit can attach up to six reference images to a 3–15 second source clip. An extension adds 3–10 seconds per call until the total reaches 30 seconds, and a follow-up can send previousInteractionId to refine the clip. Other models on this Gateway are listed in the AI Gateway model catalog.

How to get Gemini Omni 1.2 Flash

Set the model to google/gemini-omni-1.2-flash. The generateText example imports writeFile from node:fs/promises and generateText from ai. The prompt is "A red ball rolls across a white table." providerOptions.google sets responseModalities to text and video, and responseFormat to a video at 360p for 3s. The call returns files. Find the entry whose mediaType is video/mp4. If it is missing, throw the error "No MP4 video returned". Otherwise write video.uint8Array to video.mp4. That is the non-streaming path, and the model is selected by the model field on this call.

```javascript
import { writeFile } from 'node:fs/promises';
import { generateText } from 'ai';

```javascript
import { writeFile } from 'node:fs/promises';
import { streamText } from 'ai';

What to watch after Gemini Omni 1.2 Flash

Stay inside the published caps. A generation is 3–15 seconds with native audio. An edit takes a 3–15 second source clip and at most six reference images. An extension adds 3–10 seconds per call and stops at 30 seconds total. Refinement of an existing generation is a follow-up call that sends previousInteractionId. Pixels are 360p, 720p, 1080p, or 4K, and the frame is 16:9 or 9:16. The sample that writes video.mp4 asks only for 360p and 3s, with responseModalities set to text and video.

On streamText, fullStream delivers the video as one completed file with mediaType video/mp4, then the example writes part.file.uint8Array. An error part throws part.error. generateText throws "No MP4 video returned" when files has no video/mp4 entry. AI Gateway still tracks usage and cost, applies budgets for API keys, and exposes routing rules, custom reporting, and Zero Data Retention support. Provider pricing is reflected with no markup and with no platform fee on inference, including BYOK. Other models are listed in the AI Gateway model catalog.