Is Gemini Omni 1.2 Flash safe? Photos, faces and your data

Short answer: Gemini Omni 1.2 Flash is as safe as the route you use it through and the material you give it. Google released the video model on 5 November 2026, and its headline feature is holding onto reference photos, faces and voices. That makes it more useful for product videos, and it makes the privacy question sharper than it is for a chatbot.

This piece is based on Google’s own announcements and terms as of 6 November 2026. We have not tested the model ourselves, and this is not legal advice.

What Google shipped

According to Google, Omni 1.2 Flash makes clips from 3 to 15 seconds long, up from a 10-second cap in version 1.1. You can edit a clip over several turns of conversation, swapping the background, lighting or aspect ratio, instead of generating it again. The API price falls 30% to $0.07 per second of 720p video. Google Cloud announced general availability for businesses on 6 November.

The part that matters for safety is the one Google pushes hardest. You can combine character and product images, motion clips of up to three seconds and audio cues, and the model is meant to keep faces, clothes and packaging consistent from the first frame to the last. The Gemini Omni model page leads with the same idea. Upload a colleague’s photo and you are now handling their personal data, whatever the video is for.

Route 1: the Gemini app and Google Vids

Google says 15-second clips are rolling out in the Gemini app to Google AI Ultra subscribers, and are available in Google Vids at no extra cost with a Google account or Workspace subscription. For the app, the rules are in the Gemini Apps Privacy Hub. Three points are worth knowing before you upload anything.

• —Some of what you submit is read by human reviewers, including reviewers at Google’s service providers. Google tells you not to enter anything confidential you would not want a reviewer to see.
• —Conversations that reviewers have seen are kept for up to three years, even if you delete your activity.
• —With activity saving turned off, chats are still held in your account for 72 hours, but are not used to train models unless you send feedback.

The Privacy Hub also asks you to get permission before using other people’s faces or voices in Gemini, and to upload only material you have the rights to. For a video model that is the line that matters most. A made-up clip of a real person saying something they never said can do real harm, whoever made it and however good the intent.

If your company uses Vids through Workspace, the terms depend on its contract. That is a question for whoever runs IT or data protection, not for a blog post.

Route 2: AI Studio and the Gemini API

The Gemini API Additional Terms split free and paid use, and the split is the most important fact in this article. On unpaid services, including Google AI Studio and the free API quota, Google uses what you submit and what comes back to improve its products, and human reviewers may read it. The terms say plainly not to submit sensitive, confidential or personal information there.

On paid services, Google says it does not use your prompts, files or responses to improve its products, and processes them under its data processing terms. It still logs them for a limited period to detect breaches of its Generative AI Prohibited Use Policy, and the data may be stored briefly in any country where Google has facilities.

There is one regional exception. Users in the European Economic Area, Switzerland and the UK get the paid-service data rules on everything, free tools included. Everyone else, including users in the US, does not. So a team photo uploaded to free AI Studio from New York should be treated as training material; the same upload from Berlin should not. We covered the same terms for a text model in is Gemini 4 Argon safe.

One more detail from Google Cloud: audio references are gated behind allowlisting. Do not assume every voice feature is switched on for your account, and do not build a workflow around one until you have checked.

What the watermark does, and what it does not

Google says every output carries an imperceptible SynthID watermark. That lets a checking tool confirm later that a clip came from Google’s models. It does nothing for the person watching the video, who cannot see it. If a clip shows a real person, label it as AI-generated in the video or the caption as well.

The same pattern has now shown up across three Google launches in six weeks. Live talking faces came first, covered in is Gemini Live Avatar safe, and cloned voices in is Gemini 3.8 TTS safe. Each comes with a watermark and rules on who may use it. Neither kind of control helps the person on the receiving end of a convincing fake, which is why a face or voice on a screen is still not proof of who sent it.

Before your first clip

1. 01Decide which route you are using: the Gemini app, Vids on a work account, free AI Studio or the paid API. Who can read your uploads depends on it.
2. 02Keep company material on the paid API or a work account your data team has signed off. Outside the EEA, Switzerland and the UK, never on the free tier.
3. 03Use a real person’s photo or voice only with their consent.
4. 04Start with product shots and invented characters, not your team’s faces.
5. 05In the Gemini app, check your activity setting and leave out anything a reviewer should not see.
6. 06Label any generated clip that shows a real person, even though SynthID is already in it.

Do you need it at all?

Honestly, for one short video in a slide deck, probably not. Omni 1.2 Flash is aimed at teams producing many versions of the same product or ad clip. If you are starting out, writing a precise brief matters more than which model renders it, and the same habits from writing AI image prompts carry over to video.

A better model also does not make a clip correct. Google itself says fast sports action and complex physics still need work. A product label that blurs or a detail that changes halfway through only gets caught if someone watches the whole thing, the same way checking an AI answer when you are not the expert only works if you actually do it.

Coursium is an app that teaches you to use AI at work. Giving clear instructions and checking the result before you pass it on stays your job with every new model. If you want to practise that, have a look at Coursium or try the free practice example.