Gemini Omni 1.2 Flash brings text-to-video with native audio to Vercel AI Gateway

Google's Gemini Omni 1.2 Flash, a video generation model with native audio output, is now available through Vercel's AI Gateway. According to Vercel's changelog, developers can generate and edit videos using one API key, with no separate Google account or provider-specific setup required.

What the model does

Per Vercel, Gemini Omni 1.2 Flash generates clips of three to fifteen seconds from text prompts, producing audio as a native part of the output rather than as a separate step. Beyond generation, the model can edit source clips of the same length and accepts up to six reference images alongside a request.

Generated clips can be extended in increments of three to ten seconds per call, up to a maximum of thirty seconds in total. Vercel also documents an iterative workflow: follow-up calls can point back to earlier output using a previousInteractionId, letting developers refine a clip without resubmitting everything from scratch.

Output resolutions range from 360p up to 4K, in either 16:9 landscape or 9:16 portrait.

How it fits into the AI SDK

The model is addressed as google/gemini-omni-1.2-flash. Vercel's worked example generates a three-second, 360p clip and saves it to video.mp4. Developers use the standard generateText or streamText functions in the AI SDK, with the video returned as a file. In streaming mode, Vercel says to read fullStream to handle generated files as they arrive, though the video itself arrives as a finished file rather than as incremental chunks.

Gateway features and pricing

AI Gateway sits in front of a range of model providers behind a single API, handling usage and cost tracking, retries, failover, and performance tuning that Vercel claims delivers higher uptime than calling a provider directly. Other capabilities listed in the changelog include built-in custom reporting, Zero Data Retention support, budgets scoped to individual API keys, and routing rules.

On cost, Vercel states that AI Gateway mirrors provider pricing with no markup and no platform fee on inference. That policy also covers Bring Your Own Key requests, where customers supply credentials for a provider they already pay.

Why it matters

Video generation has usually meant adopting a new API surface, new credentials, and a separate billing relationship for each provider. Routing a video model through the same gateway that already serves text models collapses that overhead: an application built on the AI SDK can add video output largely by changing a model name rather than writing a new integration.

The native audio detail matters too. Models that produce silent footage push extra work onto developers, who then have to sync separately generated soundtracks, license audio, or ship mute clips. Audio arriving as part of the model output removes that step, which is notable for short-form content use cases that line up with the 9:16 portrait support.

Finally, the extension and refinement mechanics — incremental length additions and the previousInteractionId loop — signal that Vercel expects developers to treat video generation as iterative, closer to how text is drafted and revised today. Combined with no-markup pricing, including on BYOK traffic, the gateway is positioned as infrastructure that stays out of the way while broadening what a single API key can do.