AI you can video call. Point your phone camera at something — a document, a screen, a sign in a language you do not read — and ask about it out loud.
AI has moved past typing questions into a box. With Gemini Live you can effectively make a video call to an AI — turn on your camera, show it something, and ask about what it is looking at.
It sounds like a novelty until the first time it saves you ten minutes.
How to use it
Open the Gemini app on your phone, go to the live chat option, and turn the camera on. From there you talk normally. Point the camera at whatever you are asking about and say the question out loud — nothing to type, nothing to upload.
The example from the video
We came across a tweet from Emmanuel Macron, the President of France. It was written in French, which made it effectively unreadable without a detour through a translation tool.
Instead we opened Gemini Live, pointed the camera at the tweet on screen, and asked — in Hindi — to translate it and explain what it was about. Gemini read it through the camera and answered in Hindi, summarising the post's content about international security concerns and France's stated position.
Foreign-language post to understood-in-your-own-language in one step, without copying text, opening a translator, or losing the context of what you were reading.
Why the camera changes things
The difference is not translation — translation tools have existed for years. It is that you no longer have to get the content into the tool.
Normally you select text, copy, switch apps, paste, read, switch back. That works for text on a screen and fails completely for anything else: a printed document, a product label, a handwritten note, an error message on someone else's machine, a sign on a wall.
Pointing a camera works for all of it. The question becomes "what is this?" rather than "how do I get this into a format the tool accepts?"
How this differs from scanning or OCR
It is easy to assume this is optical character recognition with extra steps. It is not, and the distinction matters.
OCR extracts text. You get characters, which you then have to do something with. If the document is a table, you get a jumble. If it is a diagram, you get nothing useful.
Gemini Live interprets. You can ask about layout, about what something means, about whether a label indicates what you think it does, about a part you are holding up. And because it is a conversation, you can follow up — "what about the second line?", "is that the same as…?" — without starting over. That is a different capability wearing similar clothes.
Where this is genuinely useful
- Documents in another language — contracts, invoices, specification sheets from an overseas supplier
- Product labels — ingredients, compliance markings, batch information
- Something on a screen you cannot copy — an error dialog, a locked PDF, a screenshot someone sent
- Physical things you do not recognise — a component, a connector, a machine part
- Travelling — signage, menus, forms
Where it struggles
Knowing the failure modes saves you from trusting it at the wrong moment:
- Poor lighting and awkward angles. Glare on a glossy page or a screen photographed at a slant degrades accuracy sharply.
- Handwriting. Still unreliable, especially cursive or hurried notes.
- Technical jargon and part numbers. Exactly where precision matters most, and exactly where recognition is weakest.
- Dense tables. It will summarise, but do not assume every figure transferred correctly.
For anything consequential, treat the answer as a fast first reading rather than a verified translation. A supplier contract still needs proper review by someone accountable for it.
The privacy dimension, stated plainly
You are streaming camera footage to a cloud service. That is fine for a menu and a considered decision for a client contract, an internal report or anything covered by a confidentiality agreement.
It is also worth noticing what else is in frame. A camera pointed at a document on your desk sees the rest of the desk — other papers, a screen, a whiteboard behind you. Set your own rule before it becomes a reflex.
Worth trying today
Open Gemini, go to live chat, turn on the camera, and point it at the first thing you do not fully understand. It takes thirty seconds, and it is the quickest way to grasp how far this has come.
Share this article
