3 ChatGPT Prompts to Find Anything in a Photo
The 3 prompts I'd use to find something you've lost, read a label you can't, and name a broken part so you can buy a new one, all from one photo.
Back to all free guidesSo in the reel I showed how ChatGPT can find things in a photo that you'd walk straight past. A reel of it finding a pair of earbuds in a pile of gravel has passed 2.3 million plays. Here's how to take a photo it can work with, and the 3 prompts I'd use.
Take a photo it can search
- Get the whole area in. Step back far enough that the thing you're after is somewhere in the frame. A tight shot of the wrong corner can't help.
- Use good light. Open the curtains or turn the big light on. Shadows hide small things.
- Hold still. A sharp photo beats a close one, because ChatGPT can look closer at a sharp photo.
- Pick Thinking if you've got it. If your plan shows a Thinking option in the model picker, use it for photos. It's slower, because it works the problem through before it answers.
The 3 prompts
1. Find something you've lost
Photograph the whole mess, upload it, then send this with the bits in angle brackets filled in:
I've lost my <thing>. It's <colour>, about <size>, and shaped like <shape>. Look through this photo carefully, zoom into each part, and tell me where it is using top, middle or bottom and left, centre or right. If you're not sure, give me your two best guesses.The colour, size and shape do most of the work, so describe them the way you'd describe them to a friend helping you look.
2. Read a label you can't
For the sticker on the back of a boiler, a TV, a washing machine or anything else with print too small to read, take the photo as close as you can while it's still sharp, then send:
Read every piece of text on this label exactly as it's written, especially the model and serial numbers. Tell me if any character is hard to make out.If it says a character is hard to make out, take one more photo from a slightly different angle and send that too.
3. Name a broken part
Once you've got the model number, put the broken bit on a plain table, photograph it, then send:
This part is broken. Tell me what it's called, the size or spec I need to check, and the exact words to search to buy a replacement in the UK.Change "in the UK" to wherever you are. If you got the model number from prompt 2, paste it in after the first sentence, so the search words match your exact model.
Why it can do this
OpenAI first showed its thinking models zooming in, cropping and flipping photos while they work out an answer back in April 2025, in its Thinking with images post. That's why the prompts ask it to look through each part of the photo, rather than glance at the whole thing once.