Gemini Omni won't take photos of minors in the EEA, UK or Switzerland

Google's Gemini Omni guide says uploading or editing images that contain minors is not available for the European Economic Area, Switzerland and the United Kingdom. The same list blocks editing or extending uploaded videos in those regions, while editing a video the model itself generated is supported everywhere.

That is a rule on Google's API, quoted from the Generate and edit videos with Gemini Omni Flash guide, read on 2026-10-02. Sume's Video Router docs describe the same model's inputs and do not mention a regional rule, which is a gap in the docs and not an answer.

What exactly does Google restrict?

The guide has a regional restrictions section with two bullets. One covers uploading and editing images with minors, the other covers editing and extending uploaded videos. Both apply to the EEA, Switzerland and the UK. A note says editing model-generated video is supported in all regions. It also says content filters are applied regionally.

The guide's input rules sit next to this: an input video is at most 10 seconds unless you are extending through a multi-turn chain, and a video reference is at most three clips of three seconds each. The model page lists the stable id gemini-omni-1.1-flash and a preview id, text, image and video input up to 10 seconds, and output of 3 to 10 seconds at 24 FPS.

Does that apply to the model on Sume?

Sume's router docs say gemini-omni-flash-1.1 takes image_url, end_image_url, reference_image_urls and reference_video_urls, plus a video_url edit source. They do not describe any regional rule for photos of children, and they do not describe a content review of what you upload either. I will not guess in either direction.

What the docs do say is who is responsible for the media: URLs must be fetchable public HTTPS addresses, and some Sume routes want files imported to your workspace first. Whether you may upload a picture of a person, and in particular a minor, depends on consent, your local law and the provider's terms behind the model. Ask your counsel, not a blog.

What should I do before building on it?

The safest design for a mixed-region product is to generate from text or from synthetic characters, and to avoid uploading photos of people you cannot document. That also sidesteps Google's two restricted input types. The same advice holds if your content, rather than your users, touches a minor: put the rule in your own checks.

• Decide whether your own users are in a blocked region. If they are, a Google-backed path will refuse them.
• Keep proof of consent for any real person in a reference photo, and never upload a minor's image without a guardian's written permission.
• Read capabilities from GET /v1/video-router/models instead of assuming an envelope.
• Store the job id and the exact request for every paid call.

curl https://api.sume.com/v1/video-router/models \
  -H "Authorization: Bearer $SUME_API_KEY"

Why would a regional rule exist at all?

Google does not explain its reasons in the guide, so I will not either. What the guide does show is a pattern: the restricted inputs are the ones that put a real person's image or footage into the model. Model-generated video, which involves no upload of a real person, is supported everywhere.

That pattern is useful for planning even without the reasons. If your product depends on uploading photos of customers, expect friction in regulated regions with any provider, and design a fallback, such as a synthetic character or a text-only route.