My friend held up a handwritten recipe card during a video call last week. “Can you read this? I can’t figure out my grandmother’s handwriting.” I took a screenshot, pasted it into ChatGPT, and typed “What does this say?” It transcribed the entire recipe in three seconds flat. She stared at me like I’d performed some kind of magic trick.

That moment is the clearest answer I can give anyone asking what can ChatGPT do with images. It reads them, describes them, edits them, and creates new ones from scratch — all inside one chat window you already know how to use. No special software, no monthly fee, and no reason to feel behind. If you’ve never touched AI and the endless tool announcements have started to blur together, this is the one feature worth ten minutes of your evening.

So: what can ChatGPT do with images, specifically? Four things. It can look at a photo or screenshot and tell you what’s in it, pull text out of handwritten notes and receipts, edit a photo you upload using plain-English instructions, and generate brand-new pictures from a written description. Each one takes under a minute to try, which is more than I can say for most tech I’ve recommended to friends.

What actually changed

ChatGPT can now generate images directly in your conversation. You describe what you want — “a watercolor sunset over a mountain lake” — and it creates it. No Midjourney subscription, no Discord bot, no separate app. Just type and get.

But generation isn’t the headline. Editing is. Upload any photo and say “remove the person in the background” or “make this look like a painting” or “add a dog sitting next to me.” The AI reads your image and modifies it based on your instructions, written the same way you’d text a friend.

For someone who’s never used AI before, this is the lowest-friction entry point. You don’t need to learn prompting or understand tokens.

What you can do with it today

Product photos for your business

If you sell anything online — Etsy, Shopify, Amazon — product photography is expensive. A professional shoot costs $200-500 per product. With ChatGPT’s image feature, you can:

  • Upload a product photo and swap the background to a lifestyle scene
  • Generate multiple variations from one photo (different angles, different settings)
  • Create seasonal versions without reshooting

I tested this with a simple product photo and asked ChatGPT to “place this on a clean marble surface with soft natural lighting.” The result was good enough that I’d have happily paid a photographer for it, and I wouldn’t have known the difference without being told.

Extracting text from photos and screenshots

This one surprised me most. Point your phone camera at a receipt, a whiteboard from a meeting, a page in an old book, or a doctor’s scribbled instructions, and ChatGPT will transcribe it. I’ve used it to copy a Wi-Fi password off a router sticker and to digitize forty pages of my dad’s handwritten fishing logs. The logs part took patience — old cursive trips it up sometimes — but it got maybe 90% right on the first pass, which beat me squinting at ink for an afternoon.

To try it: tap the paperclip icon in the chat box, choose or snap a photo, then type “transcribe everything in this image.” That’s the whole process.

Understanding what’s in a photo

Upload a picture of a mystery plant and ask what it is. Paste in a screenshot of an error message and ask what went wrong. Show it a piece of furniture and ask how to fix a wobbly leg. The analysis side of what ChatGPT can do with images gets less attention than the generation side, but it’s the part you’ll use weekly — identifying an ingredient label, checking whether a parking sign applies to your car, translating a menu in another language. I’ve done all four this month.

Making images from a description

Type “a cartoon corgi wearing a tiny raincoat” and you’ll get one. Where this gets useful is when you need something specific: a flyer for a garage sale, a placeholder graphic for a slide deck, a rough mockup of a logo idea to show a designer. Tell it the size, the style, the text you want on it, and iterate. If the first version isn’t right, say “make the text bigger” or “use a darker blue” and it adjusts the same image rather than starting over.

Why this matters right now

You might have read that the AI industry is in a pessimistic mood — MIT Technology Review recently ran a piece titled “The AI industry has taken a doomer turn,” covering how even insiders are questioning where the technology is heading. That debate is real, and it mostly concerns what happens at big labs over the next decade. It has almost nothing to do with whether the tools available today are useful to you. The image feature works now, costs nothing, and solves small daily problems. You can let the industry argue about the future while you transcribe your grandmother’s recipe card.

Limitations worth knowing before you start

I’d rather you hear this from me than be annoyed later.

Generated images sometimes get text wrong — if you ask for a flyer with words on it, check the spelling before you print anything. Faces of real people can come out slightly off, which is why I don’t recommend using it for family photos you care about. And if you upload something sensitive, remember it’s going to a company’s servers; I keep medical documents and anything with my address out of the chat entirely.

One more quirk: very long or dense images (a full spreadsheet screenshot, say) can get partially skipped in transcription. Cropping to the section you actually need fixes this most of the time.

How to try it in the next five minutes

  1. Go to chatgpt.com and sign up free — an email address is all it asks for
  2. Click the paperclip icon and upload any photo from your phone or desktop
  3. Type a plain question about it: “What’s in this picture?” works fine
  4. Then try “Now turn this into a watercolor painting” to see the editing side

That’s it. The fastest way to understand what ChatGPT can do with images is to hand it one and see what comes back. Start with something low-stakes — a plant, a receipt, a screenshot — and you’ll have your own version of the recipe card story by dinner.