Upload one product photo to ChatGPT, type three short prompts (minimal poster, hero offer, clean flyer), then check every digit before it posts.
Swipe or tap Next · checked on 2026-10-31
Not technical? Start here. Copy this prompt into a new ChatGPT chat (the free version works) and it walks you through the whole guide, one step at a time, in plain English, and makes the images right there. In Claude it coaches you the same way while you make the images in ChatGPT.
It asks before it assumes anything and checks with you before anything that costs money, reaches a real customer or deletes something. The rest of the page is the same guide, written out, if you'd rather do it yourself.
The walkthrough prompt (copy all of it) · lines 1-4 of 46. Copy all copies the whole prompt; read it in full on the page.
Hi! Please be my patient guide for this step-by-step guide from Receipts Group: "Turn 1 photo into 3 ads that don't look AI-made"
The guide is here: https://receiptsgroup.com/guides/one-photo-three-ads (open it if you can; if you can't, everything you need is below).
About me: I run a small local business and I'm not technical. Use plain words and short sentences.This is the manual way: every step, written out. The prompt above follows it too.
Something wrong? Fix just that spot with Select, then run the check again from the top.
In focus, good light, the whole product in the shot. A window on a clean wall beats a window behind a ladder and a shop vac. ChatGPT takes PNG and JPG, and OpenAI caps an image at 20 MB, which a normal phone photo is way under.
Your own photo is the whole point: the ad shows what you actually sell, not a made-up one.
If it's a customer's home, crop out the house number, street signs, faces and license plates before you upload, and only use it if they said yes. That's their house, not your stock photo.
A short headline, one offer you really run, your company name, your phone number and your website. For the flyer, three short points. Writing them down first means you type them once, exactly right, and you've got something to check the image against.
Example (made up): headline “Storm season's coming”, offer “Free in-home estimate”, points “Impact-rated glass”, “Installed by our own crew”, “Financing available”.
Log in and open a fresh chat, so nothing from an old conversation leaks into your ads.
Tap the + in the prompt area, choose Add photos & files, and pick your photo. You can also drag it into the chat or paste it.
On the Free plan OpenAI limits you to 3 file uploads a day. Upload the photo once and make all three ads right here, one after the other, so you're not burning an upload on every ad.
Here they are exactly like in the video. They work as is. But the short version makes ChatGPT guess your headline and your offer, and a guess is how a made-up claim ends up on your ad. So under each one is the longer version I'd use: same idea, with your words in it and the glow banned. Swap everything in [BRACKETS] for your own words before you send it.
Your product on a plain background with a lot of empty space and one short line. This is the one that tests whether ChatGPT kept your product looking like your product.
Short (as in the video)
minimal posterLonger: the one I'd use
Minimal poster. Use the product in my photo exactly as it is: same shape, same color, same frame. Don't redraw it or add anything to it.
Plain, light background with a lot of empty space. Natural daylight, real shadows.
One short headline at the top: [YOUR HEADLINE]
No other text. No glow, no neon, no purple, no gradients, no sparkles, no extra objects.
Tall 4:5 shape.Your offer goes up big with room to breathe. Only put an offer on it that you actually run, word for word, because people will call and ask for it.
Short (as in the video)
hero offer minimalLonger: the one I'd use
Hero offer minimal. Same product from my photo, unchanged.
Put this offer big, as the main thing on the page: [YOUR OFFER]
Under it, smaller: [YOUR COMPANY NAME] and [YOUR PHONE NUMBER]
Copy every word and digit exactly as I typed them. Don't add any other claims, badges, stars, prices or dates.
Lots of room to breathe, two colors at most, plain background, no glow.
Tall 4:5 shape.The one I'd post. Headline, three points, and how to reach you, laid out clean. It has the most words on it, so it has the most chances for a typo. Check it hardest.
Short (as in the video)
clean flyerLonger: the one I'd use · Part 1 of 2 · Copy copies all 10 lines
Clean flyer. Same product from my photo, unchanged.
Headline: [YOUR HEADLINE]
Three short points:
- [POINT 1]
- [POINT 2]
- [POINT 3]
Bottom: [YOUR COMPANY NAME], [YOUR PHONE NUMBER], [YOUR WEBSITE]
Copy every word and digit exactly as I typed them. Nothing else on it: no extra claims, badges, stars or prices.
Simple layout, plain background, easy to read from across a room, no glow.Longer: the one I'd use · Part 2 of 2 · Copy copies all 10 lines
Tall 4:5 shape.Click the image to open it, choose Select, highlight only the part that's wrong, and type what it should say. You can also just say the fix in the chat. OpenAI warns the highlight isn't always exact and the edit can spill past it, so read the whole image again after.
Use the aspect ratio picker, or say it in the prompt: 4:5 for a tall feed post, 9:16 for a story. Then check the words again, because a reshaped image is a redrawn image.
This is the step that actually protects you. ChatGPT can put clean text on an image, but it doesn't know your phone number. It knows what you typed and what it drew, and OpenAI's own docs say the image model can still struggle with precise text placement and clarity. So read it off the image, not off your notes, because your brain fills in what you meant to type.
Copy this: the check
BEFORE IT GOES OUT (read it off the image, not off your notes)
[ ] Phone number: read it out loud, digit by digit
[ ] Then call it from another phone. Does YOUR phone ring?
[ ] Website: type it into a browser exactly as it's spelled on the image
[ ] Company name: every letter
[ ] Headline and points: no doubled words, no made-up words
[ ] The offer: it's one you actually run, word for word
[ ] No extra claims it slipped in (stars, "#1", a warranty, a price, a date)
[ ] The product looks like what you actually sell
[ ] No customer's house number, street sign, face or plate in the photo
[ ] Someone else reads it too, because you'll see what you meant to typeReading it isn't enough. Call the number printed on the image from another phone and make sure your phone rings. It takes ten seconds, and a wrong digit means you're paying for an ad that rings a stranger.
Everything ChatGPT makes is kept under Images in ChatGPT too, so if you lose the file you can go back and grab it.
Before you put any words on it, see if ChatGPT keeps your product looking right. Upload the photo and send this:
Prompt: the no-words test
Minimal poster. Use the product in my photo exactly as it is: same shape, same color, same frame. Plain light background, lots of empty space, natural daylight. No text at all. No glow, no gradients, no extra objects. Tall 4:5 shape.Good: it's clearly your product, same shape and color, on a plain background with real-looking light. Bad: the frame changed, extra panes showed up, it's glowing, or there's random text on it. Then say exactly what changed (“keep the frame exactly like my photo”) and run it again before you move on to the ads with words.
1. A story version
Make the same clean flyer in a 9:16 shape for a story. Keep every word exactly the same, keep the product unchanged, and leave empty space at the top and bottom.2. Fix one word
On the last image, change only this text: "[WHAT IT SAYS NOW]" should say "[WHAT IT SHOULD SAY]". Don't change anything else.3. Strip the AI look
Make it look less AI-made: flat natural light, no glow, no gradients, no reflections that aren't in my photo, plain background, one font. Keep the product and every word exactly the same.4. Read it back (a second pair of eyes, not the final check)
Read every word and number on the last image back to me exactly as it's spelled there, line by line. Don't correct anything, just read it.The read-back helps you catch things, but ChatGPT can misread an image too. The final check is still you reading it and calling the number.
Checked 2026-10-31 on OpenAI's pricing page and help center. OpenAI doesn't publish a fixed number of images per plan. Prices change: check before you buy.
I'd start on Free. You'll know after the first test whether it's worth paying for more, and if you hit the limit, ChatGPT tells you.
Fix just that spot: click the image, Select, highlight it, type the right text. Or use prompt 2 above. Then read the whole thing again and call the number again.
Leave the number off the image and put it in the post's text, where you type it yourself. Or add it afterward in whatever app you normally put text on photos with.
Tell it to remove it, and say “don't add any claims I didn't give you”. Never post a claim you can't back up.
Say exactly what changed (“keep the frame and the grid exactly like my photo”) and run it again. If it won't hold, use a cleaner photo with the product bigger in the shot. Don't post a product you don't sell.
Add “no glow, no neon, no gradients, plain background, natural light” to the prompt, or use prompt 3 above.
Use the aspect ratio picker, or say “4:5” or “9:16” in the prompt. Then check the words again.
OpenAI says highlights aren't always exact. Undo it, highlight a slightly bigger area, or describe the fix in the chat instead.
Make sure it's a PNG or JPG under 20 MB. On the Free plan you get 3 uploads a day, so if you're out, wait or reuse the photo already in the chat.
Wait for it to reset, or upgrade to Go or Plus. OpenAI doesn't publish a fixed number, so ChatGPT tells you when you're there.
What each page backs up is listed at the end of the full guide.
One real photo, three short prompts, and you read every letter and call the number before it goes out. That's how you get an ad that looks clean and rings your phone, not a stranger's.
Want help setting this up? Receipts Group builds these systems.
Open the full guide as a page (every file in full, plus the table of contents).