AI Photo Editing Prompt Guide 2026: How to Write Prompts That Protect Your Face

Almost every viral Instagram portrait you have scrolled past this year started life as an ordinary selfie. The saree portrait lit by diyas, the neon night-street shot, the 90s film look — none of it needed a camera, a studio or a designer. It needed a good prompt. That is the entire skill, and it is a much smaller skill than people think. This guide is the one page that explains it end to end: what a prompt actually is, the exact sentence that keeps your own face from being replaced by a stranger’s, how to fix a result that came out wrong, and which of the trending looks on this site to start with.

Example result — your photo will look different

What an AI photo editing prompt really is

A prompt is just written instructions for an image model. You upload your selfie, you describe the photo you wish you had, and the model repaints everything around your face — the clothes, the background, the light — while keeping you recognisable. It is closer to hiring a stylist and a photographer than to using a filter. A filter changes colours on top of your photo; a prompt rebuilds the photo.

That difference matters, because it explains why results vary so much. If you tell a stylist only “make it look nice”, you will get whatever they feel like. Tell them “maroon silk saree, gold jewellery, a balcony full of diyas behind me, warm evening light”, and you get what you pictured. The model works the same way. Vague prompt, random result. Specific prompt, repeatable result.

The anatomy of a prompt that works

Every prompt worth copying has six parts, always in the same order. First the face lock — the instruction that your face must not change. Then the shot type, which decides how much of you is in the picture. Then the outfit, with fabric and colour named, not just “traditional dress”. Then the background, with a note that it should be out of focus. Then the lighting, which is the part most people skip and the part that does the most work. And finally the format and clean-up line — vertical 4:5, no text, no watermark.

Here is what that looks like assembled, and you can paste this exact one as a starting point:

Using my uploaded photo, keep my face, facial features, skin tone and hairstyle exactly the same — the face kept as close to the original as the tool can manage. Create a waist-up three-quarter shot of me in a Diwali night look. Outfit: a rich maroon and gold silk saree with delicate gold jewellery. Background: a balcony full of lit clay diyas with warm golden bokeh, kept softly out of focus. Lighting: warm golden diya glow on the face with a soft rim light. Face and body turned toward the camera, the whole head visible with space above the hair, no side profile and no cropped forehead. Finish it like a professional model shoot: 85mm lens look, shallow depth of field, magazine-quality retouch, polished natural skin. Vertical 4:5 format, high resolution. No text, no watermark, no logo.

Read it once more and notice that nothing in it is clever. It is just specific. If you would rather not assemble those six parts by hand every time, our Prompt Builder writes the whole thing from dropdowns — pick the look, the outfit, the background and the lighting, and copy the finished prompt in one tap.

Why your face changes — and how to stop it

This is the single most common complaint, and it usually has nothing to do with the prompt. In order of how often they are the real cause: the uploaded selfie is too dark, too small or too blurry; the face is partly covered by hair, a hand, sunglasses or a mask; the shot type asked for is full body, so the face ends up only a few dozen pixels wide; or the prompt never contained a face-lock sentence at all.

Fix them in that order. Start with a bright, sharp, front-facing selfie taken near a window — that alone solves most cases. Keep the shot type at waist-up for anything you will post on Instagram, because a phone screen is small and a full-body result spends its detail on your shoes instead of your face. Always keep the face-lock sentence at the very start of the prompt, where the model weighs it most.

And when a result still comes back wrong, do not paste the prompt again from scratch. Reply in the same chat with one correction: “keep the face exactly like the uploaded photo, do not change my features”, or “make the saree deeper red”. The model remembers the previous image and adjusts it. Starting over throws away everything it got right.

Which free app should you use?

Two apps cover almost everything and both have free tiers that accept photo uploads: ChatGPT and Google Gemini. In practice ChatGPT tends to follow long, detailed prompts more literally, which is what you want for these six-part prompts. Gemini is often faster and sometimes handles Indian outfits and jewellery more naturally. Neither is strictly better — if a look is not coming out right in one, paste the same prompt into the other before you rewrite anything.

A word on the apps we deliberately do not recommend: face-swap tools and one-tap “AI photo” apps. They produce a face that is not yours, often watermark the output, and several of them keep rights over what you upload. The whole point of a face-lock prompt is that the person in the final image is actually you.

The five mistakes that ruin most results

Asking for a full-body shot when the photo is going to Instagram. Leaving the background undescribed, so the model invents a cluttered one. Forgetting to say the background should be out of focus, which is what separates a professional-looking portrait from a snapshot. Naming a real actor or singer instead of describing the look — say “90s Bollywood hero look”, never a real person’s name. And piling on ten adjectives about quality (“ultra HD, 8K, masterpiece, hyperrealistic”) which modern models mostly ignore; one clear sentence about lighting does far more than ten words about resolution.

Where to start, by what you want

If a festival is coming up, start there — those are the looks people actually search for and share. We have ready prompts for Navratri Garba, Durga Puja, Dussehra, and for Diwali a full set covering the diya portrait, rangoli, crackers night, the couple look and the family Lakshmi Puja portrait.

If you want something that works any day of the year, the everyday aesthetics are the safer bet: cafe aesthetic, rainy aesthetic, gym and fitness, or the retro sets — vintage 90s Bollywood and the 1980s look.

And if the photo is only half the post, the video side matters just as much. The cinematic slow-motion prompt and the photo-to-video prompt turn a finished portrait into a reel, while the velocity edit and 3-layer reel guide cover the VN app side. For the backdrop itself, there are free HD packs for Diwali diyas and golden bokeh.

Making a set instead of a single photo

One good portrait is nice; a matching set of three is what makes a profile look designed. The trick is simple and almost nobody does it: build one prompt you are happy with, then change exactly one thing for each new image — usually the lighting, sometimes the background — and leave everything else word for word identical. Same outfit description, same shot type, same finish. The results come out looking like they were shot in one session, because in the model’s understanding they nearly were.

Common questions

Is any of this actually free?

Yes. Every prompt on this site is free to copy, and both ChatGPT and Gemini have free tiers that accept photo uploads. Free tiers do limit how many images you can make in a day, which is the only real ceiling you will hit.

Can I use these photos commercially?

For your own profile, brand page or client work, generally yes — but the terms belong to whichever app generated the image, not to us, and they change. Check the current terms of the app you used before you sell anything made with it.

Should I mention that a photo is AI-edited?

It is worth doing. Instagram already labels a lot of AI-generated content automatically, and audiences respond better to creators who say it upfront than to ones who get caught. A line in the caption is enough.

Why does the same prompt give different results each time?

Image models are random by design — the same words produce a different image every run. That is a feature, not a fault: run a prompt three times and pick the best one. It is also why the one-change-at-a-time method above works so well for building a matching set.

Start here

If you read nothing else: use a bright front-facing selfie, keep the face-lock line at the start, ask for a waist-up shot, describe the background and say it should be blurred, and correct with a follow-up message instead of starting over. Everything else is styling. Build your first prompt in the Prompt Builder, or pick a ready-made look below.