Gemini Omni 1.2 Flash: longer AI video, easier to adjust

On 5 November 2026 Google released Gemini Omni 1.2 Flash, the new version of its video model. In short: clips can be longer, the model sticks more closely to your reference images, you can adjust a clip in conversation and it is 30% cheaper than version 1.1. Here's what actually changes, what it costs and what you should and shouldn't use it for as a marketer or business owner.

What is Gemini Omni?

Gemini Omni is Google DeepMind's model that creates video from text, images, short motion clips and audio. You describe what you want to see, optionally add a photo of your product or a character, and the model turns it into a clip, including motion, camera work and spoken lines.

You'll find the model in several places:

• Gemini API and Google AI Studio, for developers building video into their own app or workflow.
• Google Cloud (Gemini for business), for companies using it at scale.
• Google Vids, the video editor in Google Workspace.
• Flow, Google's video tool for creators. Want to know how it works? Read our article on Google Flow.
• Adobe Firefly, where Omni Flash was already available and version 1.2 is being added.

Important: with this version Omni is generally available for the first time. It's no longer an experiment, but a product backed by Google's business support and infrastructure.

What's new in version 1.2?

Clips up to 15 seconds

The maximum goes from 10 to 15 seconds, and you choose the length per second, from 3 to 15. It sounds like a small step, but in practice it's exactly the length of a social ad or a short product explainer. Less stitching clips together.

The model sticks to your images better

You can add photos of a product, character or style, plus short motion clips (up to three seconds) and audio snippets. According to Google, faces, clothing, text on packaging and brand colours now stay more consistent, even when the camera moves. Google's example: one model photo plus front and back shots of three outfits, turned into a 15-second virtual try-on.

For online stores this is the most interesting part. You can turn existing product photos into rotating product shots or try-on videos without a new shoot.

Adjusting in conversation

In practice this may be the biggest improvement. You create a clip and then say: different background, different light, rain instead of sun, or portrait instead of landscape. The model edits the same clip instead of starting over.

If you've worked with AI video, you know why this matters. Often 90% is right and you only want one detail changed. Regenerating used to mean guessing again, and often losing the part that was good.

More natural speech and lip-sync

Dialogue sounds smoother, lips move more convincingly and the model can keep multiple speakers apart. You can add short voice samples so each character gets its own voice.

Steadier camera work

Long, continuous camera moves, like a flight through a building or a zoom-out, work better, with fewer unwanted jumps.

What does it cost?

Google lists $0.07 per second for 720p video via the API. A 15-second clip costs about $1.05. That's 30% less than Omni 1.1.

Google itself suggests a sensible way to keep costs down:

1. Make quick drafts in 360p or 720p.
2. Fix details with a follow-up prompt instead of starting again.
3. Only upscale your best version to 1080p or 4K.

Note: that price applies to the API. If you use Omni through Google Vids, Flow or Adobe Firefly, those tools' own plans and credits apply. Always check the current pricing of the tool itself.

What does this mean for you?

A few honest observations from how we work with AI video.

Where it really helps:

• Variants of one ad. One base clip, then versions for Instagram (portrait), YouTube (landscape), another audience or another season. Conversational editing makes that much faster.
• Product videos from existing photos. Good product photos can be brought to life without a studio.
• Concepts and storyboards. Show a client or colleague what an idea will look like before starting an expensive production.
• Short explainers. 15 seconds is enough for one clear message.

Where to be careful:

• Small details are still tricky. Even Google's own comparison shows small shifts, like objects changing between the first and last frame. Check every clip before you use it.
• Real people and real places. AI makes something that looks like it, not what it is. If you show an existing location or product, check that it hasn't quietly changed. A villa with an extra balcony quickly becomes misleading in an ad.
• It easily looks fake. Too much motion and overly smooth camera work stand out. Subtle often works better: a fixed camera with just some movement in water, leaves or light.
• Recognisable as AI. All output carries an invisible SynthID watermark from Google. Good for transparency, but don't count on nobody noticing it's AI.

How do you start?

• Without technical skills: try it through Google Vids (if you use Google Workspace) or Flow. Start with one existing product photo and a short, specific prompt.
• With a developer: the Gemini API lets you build it into your own systems, for example to generate product videos for your whole range automatically.
• Be specific. Describe the camera (static, slow push-in), the light and what may and may not move. The more precise, the less you need to correct.

In short

The 1.1 price is derived from the 30% reduction Google mentions.

Gemini Omni 1.2 Flash mainly makes AI video more practical: longer, easier to adjust and cheaper. It doesn't replace a good photographer or videographer, but it does make quickly testing variants a lot easier. Want to use AI smartly in your marketing? Read more about AI consultancy or book a strategy session.

Source: Google DeepMind announcement, 5 November 2026.