AI that sees

How does an AI see what you see?

Short answer

It works from a picture. The camera captures an image, the image travels with your question to an AI model trained on images and text, and the model writes an answer from what is inside the frame. It reads print, labels and charts well, struggles with blur, glare and tiny text, and knows nothing outside the shot. Ace Mobile works this way: tap Add, show it the thing, and ask.

Compare screen captures and phone photos to choose how to share a chart, a software window or something in front of your phone.

The Ace BL for iPhone and iPad app also lets you take a photo or choose an image, then ask about it.

Free Play: 2,000 credits, no card. Ace by Black Label.

SEE IT DIFFERENTLY in silver serif letters over liquid chrome with cyan light

From photo to answer, in plain terms

A camera AI does not look the way you do. It receives a still image, or a stream of them in a live session, and a model that has learned from enormous numbers of pictures paired with text turns that image into an internal summary of the shapes, words and layout in it. Your question steers what it pays attention to. The answer is the model’s best reading of the frame, written out in words.

Two things follow. The photo sets the ceiling on the answer: what is blurred for you is blurred for the model. And the model knows only what is in the shot. It cannot see the next page, the back of the box or the room behind you unless you show it.

Ask Ace on a phone: the same workspace with the suggestion cards and the composer.
Ask Ace on a phone: the same workspace with the suggestion cards and the composer. Live at tryacee.com/account?space=chat.

What it reads reliably

Where it slips

Ask Ace in the Ace web workspace: chat, library, study, tasks, computer and schools in one sidebar.
Ask Ace in the Ace web workspace: chat, library, study, tasks, computer and schools in one sidebar. Live at tryacee.com/account?space=chat.

Framing a shot it can read

  1. Light the subject from the side or from a window; a flash bounces off glossy paper.
  2. Hold the phone parallel to the page so straight lines stay straight.
  3. Fill the frame with the part you are asking about, leaving a little margin around it.
  4. Keep to one subject per photo: one page, one label, one chart.
  5. Look at the photo before you ask. If you cannot read it, the AI will struggle too, so retake it.

Questions that suit a camera

How Ace works: get the answer, finish the draft, keep the notes, keep things moving.
How Ace works: get the answer, finish the draft, keep the notes, keep things moving. Live at tryacee.com/how-it-works.

Tools built for blind and low-vision users

If you are blind or have low vision, start with apps designed around that need. Seeing AI’s site describes a free app, made with and for the blind and low-vision community, that narrates the world around you, from reading text to describing photos and identifying products. Be My Eyes connects blind and low-vision users with volunteers and companies through live video and AI, and lists its calls as free and unlimited. Ace is a general assistant rather than an accessibility tool. This guide is written by the team that makes Ace.

What happens to the picture

With Ace Mobile, the picture you take or choose goes with your question to Ace’s services and the AI provider that answers; Ace’s privacy policy names OpenAI as the provider that processes the pictures you choose. Camera access is a browser permission on your phone that you can withdraw, and the policy says selecting an image sends that image when you ask, without granting continuous access to other apps’ screens.

More from Ace

The full overview of how Ace helps with your screen is on tryacee.com/ai-on-screen.

Also asked

Does it keep watching after I take the photo?

Not with Ace Mobile. It answers about the photo you took or chose; to ask about something else, take another.

Can it see in low light?

It sees only what the camera captured. A dark, grainy photo gives a weaker answer, so turn on a light or move nearer a window first.

Can it tell me what an object is?

It can often name common things, but treat that as a likely answer, not a fact. A brand or model name printed on the object is a more dependable clue than its shape.

What about questions on my computer screen?

On a computer, Ace looks at the screen itself when you hold its hotkey and ask, so there is no need to photograph your monitor.

Does my question change what the AI pays attention to in a photo?

Yes. A camera AI turns the image into an internal summary of the shapes, words and layout in it, and your question steers what it pays attention to. The answer is the model’s reading of the frame, written out in words. Questions that suit a camera include “What does this label say I should do first?” and “Explain the parts of this diagram in order.”

Can an AI see anything outside the photo I take?

No. The model knows only what is in the shot. It cannot see the next page, the back of the box or the room behind you unless you show it. The photo also sets the ceiling on the answer: what is blurred for you is blurred for the model. To ask about the next page or the back of the box, show it in another photo.

Will a camera AI count things in a photo correctly, like screws in a pile?

Not reliably. Counting many similar things, like screws in a pile or people in a crowd, is one of the places a camera AI slips. It does better with printed text in good light and with tables and charts whose labels are in the frame.

How well does AI judge size or distance from a photo?

Not well. Exact sizes and distances are one of the places a camera AI slips, because a photo has no reliable scale unless something of known size is in it. For anything where a wrong answer could hurt someone, such as electrical work, treat the answer as a guess and ask a qualified person.

Should I use an AI that sees to identify a specific person?

No. Identifying a specific person is one of the places a camera AI slips, so do not use it for that. A camera AI reads print, labels and charts well, struggles with blur, glare and tiny text, and knows nothing outside the shot.

How should I light and frame a shot so the AI can read it?

Light the subject from the side or from a window, because a flash bounces off glossy paper. Hold the phone parallel to the page so straight lines stay straight, and keep to one subject per photo: one page, one label, one chart. Look at the photo before you ask; if you cannot read it, the AI will struggle too, so retake it.

Is Ace made for blind or low-vision users?

No. Ace is a general assistant rather than an accessibility tool. If you are blind or have low vision, start with apps designed around that need. Seeing AI’s site describes a free app that narrates the world around you, and Be My Eyes connects blind and low-vision users with volunteers and companies through live video and AI.

Who processes the picture when Ace Mobile looks at something for me?

The picture you take or choose goes with your question to Ace’s services and the AI provider that answers, and Ace’s privacy policy names OpenAI as that provider for pictures you choose. Camera access is a browser permission on your phone that you can withdraw, and selecting an image does not grant continuous access to other apps’ screens.

More questions

Deal yourself in.

More about Ace

Sources

  1. Seeing AI - A Visual Assistant for the Blind, Microsoft (Seeing AI). Checked November 14, 2026.
  2. Accessibility Technology for blind & low vision people - Be My Eyes, Be My Eyes. Checked November 14, 2026.

Who makes this guide