What makes a good ChatGPT prompt for a photo?
A good photo prompt asks for one change and lists what stays. Paste this, fill the brackets, and attach your photo:

"Change [one thing] in this photo to [what you want]. Keep [faces, pose, framing, light, any text] exactly as they are. Style: [natural, film, studio]. Output shape: [square, 4:5 or 16:9]."
Here it is filled in for a selfie: "Change my jacket in this photo to a navy blazer over a white shirt. Keep my face, hair, expression and pose exactly as they are. Style: natural window light. Output shape: 4:5."
The keep list does the most work. Without it, an image model repaints the whole frame, and faces drift first. OpenAI's help page says ChatGPT Images 2.6 does "a better job of keeping details consistent from one edit to the next," so the keep list gives it something to hold. The 16 prompts below are written for this page and follow that shape. They work in ChatGPT, and most paste into Gemini or Whizi unchanged.
Which ChatGPT prompts fix or restyle a portrait?
Four prompts cover most portrait jobs: a headshot, a light fix, a new outfit and a hairstyle try. Upload a clear photo of yourself first. Each keeps the face fixed, because the face is what you care about.
- Work headshot. "Turn this selfie into a head and shoulders headshot. Keep my face, hair and expression exactly as they are. Put me in a charcoal blazer over a white shirt, with a soft grey wall behind me and gentle light from the left. Shape: 4:5."
- Light fix. "Relight this portrait with warm sun from a window on the right, an hour before sunset. Lift the shadows under my eyes a little. Keep my skin texture natural and change nothing else."
- Outfit swap. "Change my outfit to a dark green knit sweater with a collar showing. Keep my face, neck, hands, pose and the background exactly as photographed."
- Hair and glasses. "Show me with a shorter, textured haircut and thin round glasses. Keep my face shape, skin tone, expression and the background. Change nothing below the shoulders."
Compare the result with your original before you use it as a profile photo. OpenAI's help pages promise better consistency, but they don't promise a face stays unchanged. For a dedicated page on work photos, see AI headshot generator.
Which ChatGPT prompts make product photos?
Three prompts cover the usual shop needs: a clean white shot, a lifestyle scene and a flat catalog cutout. Upload a plain photo of the product on any surface. The keep list protects the label, which is the first thing a model changes.
- White shot. "Place this product on a white acrylic table with a soft shadow underneath. Keep the label sharp and readable, and don't change the product's shape or colors. Square crop, even studio light."
- Lifestyle scene. "Set this bottle on a mossy stone beside a stream in morning mist, with a shallow depth of field. Keep the bottle and its label exactly as photographed. Shape: 4:5."
- Catalog page. "Cut this product out and put it on a flat terracotta background with one hard shadow to the right, like a printed catalog page. Keep the product itself unchanged."
Ask for one shot per message. If a label word comes back wrong, send "restore the label text from the original photo and keep everything else." OpenAI's image guide says masking is prompt based, and text inside a picture is where models slip, so check every letter on packaging.
Which ChatGPT prompts repair or restore an old photo?
Three prompts handle a scanned print: clean it, colorize it, repair the damage. Scan at a high resolution first, keep the original file, and run them in this order.
- Clean up. "Restore this scanned family photo. Remove scratches, creases and dust, and sharpen the faces a little. Keep every face, outfit and pose as in the original, and keep it black and white."
- Colorize. "Colorize this black and white photo with muted, natural colors that suit the era. Keep faces and skin tones realistic and don't add any objects."
- Repair damage. "Repair the torn corner and the water stain on the left by continuing the wallpaper and floor pattern. Don't touch the people."
ChatGPT redraws the picture, so a restored face is a new face that looks like the old one. Hold the original next to the result and check the eyes, teeth and hands. If a face drifted, say "go back to the original face and keep the rest of this restoration." Never delete the scan.
Which ChatGPT prompts change a photo's style or scene?
Three prompts change the look without changing the people: a painted style, a film look and a clean background swap. They are the safest place to start if you have never edited a photo in ChatGPT.
- Watercolor. "Turn this photo into a watercolor painting with loose brush edges and visible paper texture. Keep the composition, the poses and every face recognizable."
- Film look. "Make this photo look like 35mm film from 1994: visible grain, slightly faded colors, soft direct flash. Keep the framing and everyone's expression."
- Background swap. "Replace the background with a quiet city street at dusk, softly blurred. Keep the people, their clothes and the light on their faces exactly as they are, and match the shadows to the new scene."
For a small area, open the image, use the selection tool on the spot first. OpenAI says "Highlights are not always precise, and edits may extend beyond the area you selected," so read the whole frame, not just the part you painted.
Which ChatGPT prompts work for social posts, couples and aesthetic photos?
Three prompts turn a normal photo into a post: an aesthetic restyle, a couples card and a travel postcard. These are the shapes people search for with "of yourself" and "of couples." Use photos of people who agreed.
- Aesthetic. "Restyle this photo with a soft pastel color grade, lifted blacks and gentle bloom around the lights. Keep my face, pose and the framing, and don't add any objects or text. Shape: 4:5."
- Couples card. "Using this photo of the two of us, make a warm autumn portrait: golden leaves, low sun behind us, a little haze. Keep both faces, the height difference and our poses exactly as they are. Shape: 4:5."
- Travel postcard. "Turn this photo into a vintage travel postcard look with muted inks, a thin white border and slightly rough paper. Keep the scene and the people. Don't add any words."
Leave words out of the prompt on purpose. Text inside an image is where models slip, and you can add a caption in any phone app in ten seconds. If you want ideas for generating a picture from scratch instead of editing one, image prompt generator covers that.
What does ChatGPT give you for photos, compared with other tools?
ChatGPT is the strongest all round choice for typed edits, and Gemini is its closest free rival. The table lists what each tool says on its own page as of 15 November 2026. A cell reads "not stated" where the page gave no figure.
| Tool | Free use | Paid price | Photo editing | Best at |
|---|---|---|---|---|
| ChatGPT | "Limited image creation"; 3 file uploads a day | Go $8, Plus $20, Pro from $100 a month | Yes, on all tiers; selection tool; thinking mode on Plus and up | Typed edits and a built in editor |
| Gemini | Image generation with Nano Banana 2 is ticked for users with no Google AI plan, under "Standard limits"; no count printed | Not stated on the limits page | Yes, through Nano Banana 2 | Free edits with no printed cap |
| Adobe Firefly | "Free daily generations" | Standard $9.99 a month; Pro $19.99 a month | Generation; outputs can be used commercially, per Adobe | Commercial work and Photoshop access on Pro |
| Canva | "Limited access" to AI tools | Pro $144 a year | Image creation with Canva AI; no image count stated | Turning a photo into a finished design |
| Claude | No image generation | Not applicable | No: it reads photos but makes none | Writing the prompt for another tool |
| Whizi | 2 pictures a month and 5 photo uploads a day, no card | Pro includes 100 pictures a month | Yes, with Nano Banana: up to 4 reference photos, keeps the photo's framing | One app for Claude, GPT, Gemini and pictures |
The sources are OpenAI's pricing page and file uploads FAQ, Google's Gemini limits page, Adobe Firefly's plans and Canva's pricing. Claude's row comes from our own page on whether Claude can generate images.
Where each one loses. ChatGPT's free plan is tight: 3 uploads a day, while Google prints no cap for Gemini's free edits. Adobe Firefly beats ChatGPT when you need Adobe's statement that its own models' outputs are fine for commercial use. Canva beats all of them when the photo is going into a flyer or a post template. Whizi loses to ChatGPT on the brush: the selection tool is ChatGPT's, and Whizi's feature list names no brush, so it edits through the prompt alone. Gemini prints no cap on free edits, and Gemini AI photo prompts has 18 prompts built for it. Plan by plan limits for ChatGPT are in ChatGPT image generation limit.
ChatGPT's help page names the model as ChatGPT Images 2.6, and OpenAI's announcement says it keeps "characters, products, and brand details consistent as you edit." Its pricing page gives Plus "more complex and accurate image creation" and Pro "unlimited and faster image creation." Neither page prints a number of images per plan, so any "50 a day" you read elsewhere is a guess. Can ChatGPT edit photos goes into the edit tools.
How do you get a good result, and which mistakes ruin it?
Send one edit with a keep list, then fix things in small follow ups. Five mistakes ruin most results, and each has a plain fix.
- Asking for five changes at once. The model repaints more of the frame with each one. Fix: one change per message, then a follow up for the next.
- No keep list. Faces, hands and labels drift first. Fix: name what stays, every time.
- Trusting the brush. OpenAI says edits may spill outside the selection. Fix: read the whole frame after every edit.
- Asking for text in the picture. Letters come back wrong. Fix: leave words out and add them in a phone app.
- Restoring from a bad scan. The model invents detail where the scan is blurry. Fix: rescan at a higher resolution, and keep the original.
Complex requests can take a few minutes, so send the prompt, do something else while it renders, and come back to look at the whole frame before you decide whether the edit worked. Save the result. Then zoom in to full size and compare it with the original side by side, looking at eyes, hands, edges and any lettering, before you post it anywhere that people you know will see it.
Can Whizi run the same photo prompts?
Yes, and it makes sense if you want one app instead of a pile of generators. Whizi's Nano Banana row edits a photo you attach, takes up to 4 reference images, and keeps the photo's framing. The free plan gives 2 pictures a month and 5 photo uploads a day, with no card. Starter includes 20 pictures a month, Pro 100 and Powerhouse 500.
Pro carries Claude, GPT and Gemini in one app, and its first week is $1.99. That helps here in a plain way: ask Claude to write a keep list from a description of your photo, then run the prompt as an edit in the same chat. The free plan also gives 10 messages a day on Whizi AI, Auto and Claude Haiku. Word, Excel, PDF and slide files made in chat have no plan gate, so the shot list for a shoot can come out as a file. See how images work in Whizi.
What can ChatGPT not do with photos?
It can't edit like a pixel editor, it can't promise a face stays the same, and the result carries marks. Three limits matter before you post.
First, precision. ChatGPT regenerates the picture, so a crop, a blemish or a photo that must stay exactly as shot belongs in a pixel editor. OpenAI's image guide says masking is prompt based and the model may not follow a mask's exact shape.
Second, ownership. The U.S. Copyright Office's January 2025 report says AI outputs get protection only where a human author has determined "sufficient expressive elements," and that "the mere provision of prompts" doesn't count. A photo you took and lightly edited is on firmer ground than an image made from words alone. That is what the Office published, and it isn't legal advice.
Third, marks. OpenAI's C2PA page says images from ChatGPT "include both signals," Content Credentials and a SynthID watermark. The credentials "can be lost through everyday actions, such as taking a screenshot or converting a file." The SynthID watermark "may persist through some edits," and may not. Don't pass an edited photo off as unedited, and don't upload a stranger's face to alter it.
- One change per prompt, then a keep list
- Add the output shape, such as 4:5 or 16:9, in the prompt
- Leave words out of the picture and add them later
- Keep the original scan or photo, and compare at full size
- Upload photos of yourself or of people who agreed
Frequently asked questions
What are good ChatGPT prompts for photos?
The good ones name one change and list what must stay the same, such as the face, pose, light and framing. This page gives 16 you can paste, grouped by portraits, product shots, old photo repair, style changes and social posts. Add an aspect ratio like 4:5 so the crop fits where you post it.
Can ChatGPT edit photos for free?
Yes, with limits, as of November 2026. OpenAI says ChatGPT Images is available on all tiers, and its pricing page lists "Limited image creation" on Free. Its upload FAQ caps free users at 3 file uploads per day. Plus adds thinking mode for images, and Pro lists "Unlimited and faster image creation."
How do you ask ChatGPT to change only part of a photo?
Open the image, use the selection tool on the area, and describe the change. You can also name the part in the prompt and add a keep list for the rest. OpenAI warns that highlights are not always precise and edits may extend beyond the area you selected, so check the whole frame.
Which image model does ChatGPT use for photos?
ChatGPT Images 2.6, per OpenAI's help page in November 2026. OpenAI's announcement says it keeps characters, products and brand details consistent as you edit. The pricing page doesn't name a model for any plan, so check the help page if the name changes.