ChatGPT can edit photos on the free plan and on every paid plan: you upload a picture, type the change, and get a new version back. The catch is how it edits. ChatGPT generates a fresh image from your photo, so a face, a hand or the writing on a sign can come out slightly different. Below are the steps on the website and in the phone app, what to type, the real limits, and the jobs where a different editor does better.
What ChatGPT can do with your photos
ChatGPT can add, remove or change parts of a photo you upload, or of an image it made earlier. The feature is called ChatGPT Images, and OpenAI's help center says it's available on all tiers, Free included, on the web plus the iOS and Android apps. There are two ways to point it at a change:
- Describe it in the chat. Type the change as a normal message. If only one part should change, name it ("the gray sky behind the house").
- Highlight it with the selection tool. Open the image in the editor, highlight the area, and describe what should go there.
It suits changes you can say in a sentence or two: a clear sky over a gray one, a stranger removed from behind you, a different shirt color, a plain studio backdrop, or a photo turned into an illustration. OpenAI's help center also lists making the background transparent among the instructions it can follow, which is handy for a product cutout or a sticker. You can also attach two images, one for the content and one for the look, and OpenAI's guide suggests referring to them by order ("Image 1", "Image 2") and saying how they relate.
OpenAI's system card for ChatGPT Images 2.5, published September 8, 2026, says the model gives more consistent results when editing and keeps more of the original's details. Every image you create is saved under Images in ChatGPT, so you can reopen an earlier version and edit from there.
Where ChatGPT photo editing falls short
ChatGPT photo editing works by generating a new picture, not by nudging the pixels you already have, and most of its limits follow from that.
- Each edit is a new image. OpenAI's image guide advises changing a single element per round, and repeating anything that must not change, so the rest of the picture doesn't drift. Check faces, hands, jewelry and logos after every edit, not only the part you asked about.
- Selections are approximate. OpenAI's help center warns that a highlight may not be exact, so the change can spill past the area you picked. Highlight a little inside the edge of what you want changed.
- No sliders. The ChatGPT Images editor has Select, Aspect ratio, Undo, Redo, Cancel and Save. For exact brightness or contrast, use the Photoshop app inside ChatGPT (covered below) or a regular editor.
- Words in the picture. Text on signs, shirts and packaging is redrawn with everything else and can come back misspelled. OpenAI's guide says to put exact wording in quotation marks and to review every word in text-heavy images.
- Exact placement. OpenAI's Image Inputs FAQ counts precise spatial localization among the things the model finds hard, so "move the vase 5 cm to the left" may not land where you meant. Describe the new spot by what's next to it instead.
- Sideways photos. A rotated or upside-down photo can be misread, so straighten it in your photo app before you upload.
- Usage limits. Free accounts get limited image generation and 3 file uploads per day. The guide to ChatGPT's image limits covers each plan.
How to edit photos with ChatGPT
To edit photos with ChatGPT, upload the photo, type the change, then use the editor for any one area that needs more work. Uploads can be PNG, JPEG or non-animated GIF, up to 20 MB per image.
On the website or desktop app
- Click the + icon in the message box and choose "Add photos & files". You can also drag the photo into the message box or paste it.
- Type the change, plus anything that should stay as it is, then send it.
- To change one area, click the result to open the editor and choose Select.
- Highlight the area, then describe the change in the editor's text box.
- Use Undo or Redo to fix the selection, or Cancel to start over.
- Choose Aspect ratio if you need a different shape. This regenerates the whole image.
- Choose Save when you're finished.
On the ChatGPT phone app
- Tap the + icon, add the photo from your library, type the change and send it.
- Tap the result to open it full-screen.
- Tap Edit to describe a change, or tap Select to highlight an area.
- Drag the slider to make the selection tool bigger or smaller. Undo and Redo fix mistakes.
- Tap Next, then type what should change inside the highlighted area.
How to edit a photo with a ChatGPT prompt
A ChatGPT edit prompt needs the one change, the parts that must not move, and the look of the result, in one to three clear sentences. OpenAI's guide asks for concrete visual words, such as where the light comes from, instead of broad reactions. Put any text that should appear in quotation marks.
These three prompts are ready to paste. Each also works in other editors that take a typed instruction.
Remove the woman in the orange raincoat standing behind us on the left, including her shadow on the sidewalk. Fill the space with the brick wall and the bike rack that continue behind her. Leave the two of us, our faces and our drinks untouched.Replace the white sky with a clear blue evening sky with a few thin clouds near the horizon. Keep the house, the roofline, the trees and the lawn exactly as they are, and match the light on the house to the new sky.Turn this photo of our dog asleep on the green couch into a soft watercolor illustration on off-white paper. Keep his brown and white markings, his collar and the way he's curled up, and keep the couch green.What to type after you highlight an area
Once you've highlighted part of the image, the text box only has to describe that area. Use literal words: "a plain white wall", "a potted fern in a gray pot", or "remove this and continue the wooden floor". If the change spills onto something nearby, add one line about it, such as "change only the highlighted mug and leave the hands holding it as they are".
To change photos with ChatGPT over several rounds, send one instruction per message and check each result before the next. If a round goes wrong, reopen an earlier image from Images and edit that one, not the damaged version. For more wording, the photo editing prompts library covers lighting fixes, outfit swaps, new backgrounds and couple portraits.
Can ChatGPT edit photos like Photoshop?
Not on its own. ChatGPT Images generates the photo again from your description, while Photoshop changes the pixels you touch and leaves the rest alone. The exception is the Adobe Photoshop app inside ChatGPT, which Adobe launched on December 10, 2025.
At launch, Adobe's instructions were to start a message with the app's name and the job, for example asking Adobe Photoshop to blur the background of your image, and ChatGPT brought up Photoshop controls in the chat. Adobe lists sliders for brightness and contrast, settings such as exposure, adjustments to one part of an image, and effects such as Glitch and Glow. Adobe also said the apps were free to ChatGPT users and ran on desktop, web and iOS, with Photoshop for Android listed as coming soon. ChatGPT now keeps add-ons like this in its Plugins or Apps directory, and OpenAI says whether you can use a given one depends on your plan, workspace and region, so check that Photoshop shows up on your account.
Use the Photoshop app for corrections where the rest of the photo must stay exactly as shot: exposure, contrast, a blurred background. Use ChatGPT Images for changes that add or replace content: a new sky, an outfit color, a person removed.
Can ChatGPT edit photos of people?
Yes, within OpenAI's safety rules, and with a real risk that a face comes back slightly different.
OpenAI's system card for ChatGPT Images 2.5 says its safeguards look at your uploaded photo and your prompt together to catch malicious edits. It also says political, sexual or otherwise sensitive images of real people break its usage policies. Expect refusals for edits that could pass as a real, misleading photo of someone. OpenAI's image guide also asks you to confirm you have permission to use a real person's likeness, so edit only photos you're allowed to change, and ask the people in them before you post the result.
Likeness is the everyday problem. Because each edit generates the picture again, a face can come back a little off: different teeth, a softer jawline, eyes a shade lighter. It's most noticeable on people you know well. List the face in the "keep" part of every prompt, use a photo where faces are large, and compare the eyes and mouth with the original before you share it. The guide to editing without changing faces has keep lists for portraits, couples and group shots.
Can ChatGPT edit photos for free?
Yes. ChatGPT Images is part of the Free plan, which OpenAI's pricing page lists with limited and slower image generation. The Free Tier FAQ says image generation, file uploads and other tools each have a usage limit separate from text chat, and ChatGPT lets you know when you reach one. OpenAI describes the Free limit in words rather than as a daily number, and it says limits can change over time. Upgrading a free account to Plus, Pro or Business resets its limits.
As of September 2026, ChatGPT Go costs $8 per month in the US and gives more image creation than Free, Plus costs $20 per month, and Pro starts at $100 per month. Plan-by-plan details, and what to do while you wait, are in the guide to ChatGPT image limits per day.
Troubleshooting ChatGPT photo edits
| Problem | Likely cause | Fix |
|---|---|---|
| The face looks like someone else | The whole image was generated again | Start from the original upload and list the face, eyes and teeth as things to keep |
| The change spread past the highlighted area | Highlights aren't always precise | Highlight slightly inside the edge and name the area in words too |
| Writing on a sign or shirt changed | Text is redrawn with everything else | Quote the exact text in the prompt, or crop that area out |
| The object landed in the wrong spot | Precise positions are a known weak spot | Describe the spot by its neighbors ("on the table, left of the lamp") |
| The edit misread the photo | The upload was rotated or upside down | Rotate it in your photo app first, then upload again |
| The upload fails | The file isn't PNG, JPEG or GIF, or it's over 20 MB | Export a JPEG copy, or set an iPhone to Most Compatible |
| A message says you've hit the image limit | The Free plan's image allowance is used up | Wait until it resets, or upgrade |
When Smart Editor is the better route
Smart Editor in Toybox AI suits a photo you plan to edit several times, because each result can be edited again or set aside for the untouched original with one switch. ChatGPT is the better pick when you want to highlight one small area or need a transparent background, since Smart Editor has no brush, mask or transparent output. Each Smart Editor edit also redraws the whole image, so the prompt rules above apply there too.
- Open Smart Editor, sign in, and upload one JPG or PNG up to 10 MB under "Your image".
- Type the change in "What should change?", including what must stay the same.
- Under "More options", pick a "Quality" (HD or 4K) and a "Shape" (Original, Square 1:1, Story 9:16, Wide 16:9 or Social 4:5), then press "Apply edit". It usually takes about 30 seconds.
- For the next change, set "Editing" to "Latest result" to keep building, or "Original photo" to try a new instruction on the untouched photo.
- Compare the face and any text with your original before you download the PNG.
If an edit fails, the previous result comes back and the credits are refunded automatically. Smart Editor needs a Pro plan or higher and currently costs 85 credits per HD edit or 110 for 4K, charged per edit rather than against a daily image allowance. At the current Pro price of $9.99 a month for 1,000 credits, that works out to 11 HD edits per billing cycle (see plans and prices).