ChatGPT prompts for photos: a pro photo from your selfies, still you
Short answer: a good photo prompt names six things (what, action, where, light, shot, style and format), attaches 2–3 daylight selfies and says out loud what must not change: "keep my facial features and skin texture, no retouching". Below is why each part works, two real stories that show where AI photos go wrong, and copy-paste templates.
Free course on this topicAI Images and Slides: free courseStart the course →
Why does "make me beautiful" give you a stranger?
What happens. You upload a selfie, type "make me look professional", and get someone glossy, with a slightly different nose and eyes. Pretty, but not you, and your friends can tell.
Why. The model doesn't know your face, it knows millions of captioned photos. "Beautiful" and "professional" pull it toward the average of magazine covers: poreless skin, perfect symmetry. You asked for a vibe, so it drew the vibe instead of you.
Where it comes from. Photo prompts went mainstream in March 2025, when ChatGPT's new image model turned the internet into Studio Ghibli frames. Demand was so wild that Sam Altman posted "our GPUs are melting" and OpenAI introduced temporary rate limits (Fortune). Millions learned that AI can redraw a photo. Far fewer learned how to make it keep the person.
What it means for you. Swap "beautiful" for light and texture: "keep my facial features, visible pores, no retouching, soft daylight from a window". Good light does more than any retouch. Lesson: AI photoshoot from selfies
What should a photo prompt actually contain?
A strong prompt answers six questions: what's in the frame, what's happening, where, what the light is, how it's shot (close-up, from which angle) and the style plus format. A vague prompt leaves every choice to the model, so you get a new random result each time; a prompt built on the formula narrows the options so all four results are close and easy to pick from.
Why does this work? Models learned on photos with captions, so they speak photographer: "close-up", "flat lay", "soft window light", "golden hour", "blurred background". Two or three of those words beat ten adjectives, and "masterpiece, 8K" adds nothing. Set the format up front, or many services draw a square: 4:5 or 3:4 for a feed post (the Instagram grid shows 3:4 since 2025), 9:16 for Stories and Reels, 16:9 for YouTube.
One more trick: say what you want, not "no". The word "people" in "a street with no people" can nudge the model to add people; "an empty street at dawn, closed cafés" works better. Lesson: The prompt formula in 6 parts
How do you turn selfies into a professional photo without changing your face?
Upload 2–3 selfies in daylight with your whole face visible and no filters, then describe the scene. Gemini (Nano Banana) holds a face's likeness well, and ChatGPT copes too. Copy this and swap the scene:
Prompt: Make a portrait from my photos. Scene: [a modern office / a café / a studio with a grey backdrop], soft daylight from a window. Keep my facial features, the shape of my nose and eyes, and skin texture with pores, no retouching and no gloss. Clothes: [a simple sweater / a blazer]. Waist-up shot, vertical 4:5.
Then edit in the chat instead of starting over: "same, but warmer light". But always edit from the original. Each edit on top of an edit shifts the face a little, like a photocopy of a photocopy: after eight rounds it's someone else. Lesson: AI photoshoot from selfies
How do you edit a photo with one sentence?
Upload the photo and ask for one change: "remove the cable in the background". One edit at a time, because the more tasks you stack in one sentence, the more the model changes things you didn't ask for: angle, colours, the letters on a label.
Why? When editing, many models redraw the whole frame, just very closely. So name what must stay:
Prompt: Edit the attached photo. Change only one thing: [remove the cable in the background]. Keep everything else exactly as it is: [face, hands, light, angle, colours]. Don't add any new objects or text.
Then put the original and the result side by side and check faces, fingers, letters and edges. Lesson: Edit photos with AI
Can AI really make a blurry photo sharp?
It can make it look sharp, but it doesn't reveal anything. Upscale a photo 4× in width and height and you get 16 times more pixels, so 15 out of every 16 are invented: the AI draws what usually appears in similar photos.
The proof is a famous 2020 story. Researchers released PULSE, a model that turned pixelated faces into sharp ones. People fed it a pixelated photo of Barack Obama, and it produced the face of a white man (The Verge). It didn't glitch: you simply can't recover a person from a handful of pixels, so the model filled in the most typical face it had learned.
What it means for you: enhanced photos are great for avatars, posts and family archives, and useless as evidence. For an insurance claim or a dispute, attach the original. Lesson: Enhance and restore old photos
Where should you never use an AI photo?
Documents. A passport or visa photo exists so an officer can compare it with your face, so the rule is the same almost everywhere: no digital edits. The US State Department says passport photos must not be altered with software or apps (travel.state.gov). A careful AI portrait for LinkedIn or a CV is fine; for an ID, go to a photo studio.
Someone else's face. Technically you can upload anyone. But a "photoshoot" of your ex or a friend in a swimsuit is a deepfake that can hurt people and get you into legal trouble. Your own face as much as you like, someone else's only if they asked.
Your data. A selfie is biometrics. In ChatGPT you can stop your chats being used for training: Settings → Data controls → "Improve the model for everyone", switch it off. Lesson: AI photoshoot from selfies
FAQ
What is the best ChatGPT prompt for a professional photo?
Attach 2–3 daylight selfies and use: "Make a portrait from my photos. Scene: a studio with a grey backdrop, soft window light. Keep my facial features and skin texture with pores, no retouching. Clothes: a blazer. Waist-up, vertical 4:5." Then refine one detail at a time.
How do I keep my face the same when editing a photo with AI?
Say it directly ("keep my facial features, the shape of my nose and eyes"), avoid words like "beautiful", ask for one change at a time and always edit from the original, not from an already edited version.
Can I use an AI-generated photo for my passport?
No. Passport and visa photos must show how you look now with no digital edits; the US State Department rejects photos altered with software or apps. Use a photo studio for IDs and keep AI portraits for social media and CVs.
Try it now
The free lesson on the 6-part prompt formula takes about 10 minutes, and you leave with a template that works in ChatGPT and Gemini. Start the lesson
Free course on this topicAI Images and Slides: free courseStart the course →