Ask AI about images: get answers from a photo or screenshot
Both GPT models in our chat can look at a picture and answer questions about it: read a sign, explain a chart, find the error in a screenshot. A clear photo and a precise question make most of the difference.

Sometimes the quickest way to ask a question is to show it. A photo of a confusing parking sign, a screenshot of an error, a chart from a report, a page of handwritten maths: both GPT models in our free chat can look at a picture and answer questions about it. They read printed text, follow the lines of a graph, notice what is in a scene and connect it to what you ask. Results depend heavily on two things you control: how clear the picture is and how precise the question is. This guide covers both, plus the limits worth knowing before you rely on an answer.
What you can ask about a picture
Some uses that work well:
- Read and translate. Menus, signs, labels, letters in a language you do not speak.
- Explain a chart or table. "What is the main trend here, and is there anything unusual?"
- Troubleshoot a screen. An error dialog, a settings page, a spreadsheet formula that shows the wrong result.
- Check your working. A photo of a maths solution or a chemistry equation, with "find the first mistake".
- Understand a document. A bill, a timetable, an instruction leaflet, a form you need to fill in.
- Describe or compare. Two product photos, the layout of a room, the parts of a diagram.
- Identify things, carefully. A plant, a bird, a type of connector or tool. Treat this as a first guess to confirm, and never eat a plant or mushroom based on it.
How to add an image in our chat
- Press the + button at the left of the composer and pick a photo or screenshot, or drag the file onto the chat on a computer. On a phone you can choose from your gallery.
- Type your question in the same message and send.
- Ask follow-up questions as usual. The model answers about the picture in the same conversation.
A few practical details. You can attach one picture per message, in common formats such as JPG, PNG or WebP. Large photos are scaled down before they are sent, which keeps things fast but means very small print may become hard to read, so crop close to what matters. Images have their own small daily allowance alongside the free messages; the FAQ has the numbers. Both GPT-6 Luna and GPT-6.1 Terra read images, so you can use either.
Take a photo the model can read
The model can only work with what reaches it. These habits make the biggest difference:
- Fill the frame. Crop to the table, paragraph or part of the screen you are asking about.
- Use light, avoid glare. Daylight beside a window beats a flash, which bounces off glossy paper and screens.
- Shoot straight on. Hold the phone parallel to the page so lines stay straight and text is not stretched.
- Prefer screenshots to photos of screens. A real screenshot is sharper and has no reflections or moiré stripes.
- One topic per picture. Two separate problems in one photo invite a muddled answer; send them in separate messages.
- Check that you can read it. Zoom in on your own photo. If you struggle to read a number, so will the model.
Prompts that get useful answers
A picture with no question, or with "what is this?", leaves the model guessing what you care about. Say what you want to know and why.
Weak: [photo of a utility bill] Explain this. Strong: [photo of a utility bill] This is my electricity bill. Why is the total higher than last month? Point out the line items that changed, explain any charges with technical names, and tell me which numbers I should compare with my meter reading.
Weak: [screenshot of an error] Help. Strong: [screenshot of an error] I get this error when I open a spreadsheet in Excel on Windows 11. I didn't change anything since yesterday. What does the message mean, and what should I try first, in order from safest to most drastic?
Other helpful lines: "Read out all the text in this image exactly as written", "Answer only from what you can see, and say if something is unreadable", "Describe the chart's axes and units before interpreting it". That last one catches misread scales early. To try it, add a chart to this chart-reading prompt with the + button before you send.
Luna or Terra for pictures?
For reading a sign, translating a menu or describing a photo, the default GPT-6 Luna is quick and capable. For dense or technical images, press Think to use GPT-6.1 Terra. OpenAI describes Terra as combining detailed visual reasoning with clear writing, with examples such as interpreting technical drawings and matching product photos to catalogue entries. It is the better choice for busy charts, wiring diagrams, floor plans, multi-step maths in a photo and screenshots full of small details. Our Luna vs Terra comparison has more on the difference.
What it cannot do
- Create or edit pictures. Our chat answers about images; it does not draw or change them. If you ask for an image, it points you to a separate AI image tool instead.
- Read other file types. PDFs, documents and spreadsheets cannot be attached. Take a screenshot of the relevant page, or copy and paste the text.
- Identify people. Do not ask it to tell you who a stranger in a photo is. That raises serious privacy problems, and models are designed to decline it.
- Diagnose. It can describe a rash, an X-ray or a strange noise in a car's engine bay, but it is no substitute for a doctor or a mechanic who can examine the real thing.
- Be certain. Models misread digits, miss small print and occasionally describe things that are not there. Check numbers and names before you act on them.
Keep pictures private
A photo can show more than you meant to share: names on envelopes, faces in the background, open tabs in a screenshot, a house number. Crop or cover those before you attach. On gpt.me, images are sent for the answer they belong to and are not part of shared conversation links. The privacy and safety guide has a full checklist. For study uses, such as photographed homework, see our guide AI for students, and for a math problem in a photo the math solver explains each step.
FAQ
Which image formats can I upload?
Common picture formats such as JPG, PNG and WebP work. Other files, including PDFs and documents, are not supported; use a screenshot of the page instead.
Can I send several pictures at once?
One picture per message. For two images, such as before and after photos, send them in consecutive messages and then ask your question about both.
Why did it misread the text in my photo?
Usually the text was small, blurred or at an angle, or the photo was scaled down too far to keep fine print. Crop closer, retake it straight on in good light, and ask it to transcribe the text before answering.
Can it make an image for me?
No. Our chat reads images but does not create or edit them. When you ask for a picture, it suggests a separate AI image generator.
Do images cost extra?
No. A message with a picture counts like any other message, and there is also a separate small daily limit on images, so text questions keep working when it is reached.
Got a question? Ask it now
No account and no download. Open the chat, type what is on your mind, and the answer starts appearing in a second or two.
Start chatting

