ChatGPT Photo Editing Prompts That Keep Faces and Details Intact
Short answer: To edit a photo with ChatGPT or Gemini without changing the face, upload the photo and write the prompt in two parts: what must stay exactly the same (face, expression, hair, clothes, number of people) and the one thing to change. Make one change per prompt, and compare the result with the original before using it.
AI photo editors do not edit pixels the way a phone filter does. They look at your photo and redraw it with your changes. That is why a simple background swap can quietly give you a slimmer nose, different teeth or a stranger's smile.
You will not find 50 trending prompts here. Instead you get one prompt pattern that protects identity, three common edits with the failures they usually produce and the fixes, and a short privacy check before you upload family photos. This guide covers only editing a photo you already have; to create an image from nothing, use our six-part image prompt formula.
Why does AI change your face when you edit a photo?
Because the image is regenerated, not retouched. The model treats your photo as a strong reference and paints a new image that matches your instruction, so anything you did not ask it to protect can drift.
Faces drift most because small changes in eye shape, jaw line or skin texture are easy for a model to make and hard for you to notice until someone says 'that does not look like you'. Hands, text on clothes and the number of people in group shots are the next most common casualties.
The identity-lock prompt pattern
Every reliable edit prompt has four parts. Use them in this order:
- Keep: list what must not change. 'Keep the face, expression, skin tone, hairstyle, glasses and clothing exactly as in the photo.'
- Change: one edit only. 'Replace the background with a plain light-grey studio wall.'
- Match: tell it how the new part should fit. 'Match the lighting direction and colour of the original; keep natural shadows.'
- Output: say what you want back. 'Same framing and aspect ratio. Photorealistic, no smoothing or beauty filter.'
Two habits make this pattern work. First, one change per prompt: background now, colour grading in the next message. Each extra change gives the model more room to redraw everything. Second, if the face drifts, do not keep correcting the bad output. Go back to the original photo, upload it again in a new chat, and tighten the Keep list. Correcting a drifted image tends to drift further.
Edit 1: a professional headshot from a casual photo
Say you need a LinkedIn photo and you have a decent picture from a cousin's wedding in Rajouri Garden, with a crowded background.
Weak prompt: 'Make this a professional LinkedIn photo.'
What typically goes wrong: the model decides 'professional' means a new blazer, smoother skin, whiter teeth and a slightly different face. It looks polished and not quite like you, which is worse than the original for a profile people will meet in person.
Better prompt: 'Keep my face, expression, skin texture, hairstyle and kurta exactly as they are. Change only the background to a soft, out-of-focus office setting in neutral tones. Match the existing warm lighting from the left. Keep the same crop. Photorealistic; do not retouch skin or change teeth.'
Critique: the Keep list names the exact things the weak version changed; 'do not retouch skin' blocks the beauty-filter default; keeping the crop stops it from reframing you. If the result still looks off, compare eyes and jaw side by side with the original at full size before deciding.
Edit 2: a Ghibli-style family photo that still looks like your family
Style transfers such as the Ghibli-style trend are harder, because you are asking the model to redraw everything. Identity is held by features, not pixels.
Weak prompt: 'Make this Ghibli style.'
What typically goes wrong: four people become three or five, children look like generic cartoon kids, grandparents lose their glasses, and hands come out with odd fingers.
Better prompt: 'Turn this photo into a soft hand-painted animation style. There are exactly four people: from left, a man with a beard and a maroon turban, a woman in a yellow suit with a side braid, a girl of about six with two ponytails, and an older woman with round glasses and a white dupatta. Keep each person's position, pose and clothing colours. Keep the balcony and plants behind them. Do not add or remove people or objects.'
Critique: counting people and describing each one by position gives the model fixed anchors; clothing colours and accessories are what family members recognise first. Avoid asking for the style of a named living artist or studio for commercial use; describe the look instead. For brand or client work, check the tool's usage terms before posting.
Edit 3: restoring an old family photo
Restoration is where AI is most tempting and most misleading, because the model fills damage with invented detail.
Better prompt: 'Restore this scanned 1980s photo. Remove scratches, dust and the fold line across the top. Improve sharpness slightly. Keep faces, expressions and clothing exactly as they are; where detail is missing, leave it soft rather than inventing new features. Do not colourise.'
Then do colourisation as a separate step if you want it. Keep the original scan as well, and label the restored file as AI-restored so nobody in the family later mistakes invented detail for the real photograph.
ChatGPT or Gemini for photo editing?
Both accept an uploaded photo plus a text instruction and both can do the edits above. As of November 2026, the practical differences are limits and labelling, and both change often.
| ChatGPT | Gemini | |
|---|---|---|
| Edit an uploaded photo with text | Yes | Yes |
| Free use | A few images a day; more on paid plans | Daily limits on free; more on paid plans |
| Labelling of AI output | OpenAI adds content-credential metadata | Google adds an invisible SynthID watermark |
| Good habit in both | One change per prompt; re-upload the original if faces drift | Same |
For tools beyond these two, see our list of free AI image generators.
Mistakes that ruin AI photo edits
- Stacking edits in one prompt ('change the background, fix the lighting, make me look fresher and add a smile'). Each extra request gives the model licence to redraw more of the face.
- Using vague quality words such as 'enhance', 'beautify' or 'make it better'. Models read these as permission for skin smoothing and reshaping.
- Uploading a tiny, compressed WhatsApp copy. Start from the original camera file; low-resolution faces drift more because the model has less to hold on to.
- Judging the result on a phone thumbnail. Open both images at full size side by side and check eyes, teeth, ears and hands.
- Posting an edited photo of someone else without showing them first. What looks fine to you may not look like them to them.
- Using AI edits where accuracy matters: passport, visa, exam admit card or ID photos have strict rules, and an AI-altered face can get an application rejected. Use a studio or a plain crop-and-brightness edit instead.
Should you upload family photos to an AI tool?
You can, but decide deliberately, especially with children's photos. A photo uploaded to a chat is stored in your account history and, depending on your settings, may be used to improve the service.
- Check the data or training setting in your account before uploading personal photos.
- Remove location data from the file; many phones save GPS details in photos.
- Avoid photos with ID cards, school badges, house numbers or car number plates visible.
- Ask adults in the photo before you upload their picture; for children, think twice.
- Delete the chat afterwards if you do not need it.
Our guide to whether it is safe to upload files to ChatGPT explains the settings in more detail. The same identity-lock prompting is part of the image module in our Applied AI Professional course, alongside the broader method in our prompt-writing guide.
Frequently asked questions
How do I edit a photo in ChatGPT without changing my face?
Upload the photo and list what must stay the same (face, expression, skin texture, hair, clothes) before stating the single change you want. If the face still drifts, start a new chat with the original photo rather than correcting the edited one.
How do I make a Ghibli style photo with ChatGPT?
Upload the photo and ask for a soft hand-painted animation style, then describe each person by position, clothing colour and accessories and say how many people there are. Describing the look works better than naming a studio, and avoids rights questions for commercial use.
Can Gemini edit photos like ChatGPT?
Yes. Gemini accepts an uploaded photo and a text instruction and can change backgrounds, styles and details. The same prompt pattern works in both; limits on free use differ and change often.
Is it safe to upload my photos to ChatGPT for editing?
It is reasonably safe for ordinary photos if you check your data settings, remove location data and avoid anything showing IDs or addresses. Be more careful with children's photos and with pictures of people who have not agreed.
Keep reading in AI tools
- How to Write AI Prompts That Work First Time Start here
- Free AI subscriptions in India and how to claim them
- How to make a presentation with Gamma AI, from brief to PPT
- How to Use Deep Research in ChatGPT and Gemini: Brief It, Then Audit It
- How to Use Google Gemini for Everyday Office Work
Related: Applied AI Professional · How to Write AI Image Prompts That Match the Picture in Your Head · All guides in AI tools