How to Edit Photos With AI Prompts: A Beginner's Guide

Editing a photo with AI comes down to one skill: describing the change clearly. Upload your photo to Gemini or ChatGPT, type what you want changed and what must stay the same, and the model does the rest. No layers, no masks, no tutorials.

This guide teaches the structure that makes those descriptions work. Once you have it, you can improvise your own edits instead of hunting for prompts. If you would rather start from ready-made lines, our photo editing prompt library has dozens you can copy first and understand later.

The four parts of every good editing prompt

Every reliable editing prompt answers four questions. Miss one and the model starts guessing.

1. The change. One clear edit. "Replace the background with a beach at sunset." Not three edits chained together; one. You can always run a second prompt for the second change.

2. The keep line. What must not change. "Keep my face, hairstyle, and outfit exactly the same." This is the most skipped part and the most important one. AI editors regenerate the whole image, so anything you do not anchor is up for grabs.

3. The look. The finish you want: "soft natural light", "warm film tones", "clean magazine style". One or two phrases, not a paragraph.

4. The format. Only if it matters: portrait crop, square for Instagram, sharp focus on the eyes.

Replace the background with a quiet beach at sunset. Keep my face, hairstyle, and outfit exactly the same. Warm golden light, soft shadows. Portrait crop.

That is the whole method. Everything else in this guide is refinement.

Your first three edits

Start with these, in this order. Each one teaches you something about how the model behaves.

Retouch this photo lightly: even out skin tone, reduce shine, keep real skin texture. Keep my face exactly the same.

The light retouch shows you the model's default taste. If it over-smooths, you know to push back with "keep pores and texture real" in future prompts.

Replace the background with a softly blurred city street in the evening. Match the lighting on me to the scene. Keep my face and clothes unchanged.

The background swap teaches you about lighting. The phrase "match the lighting on me to the scene" is what separates a convincing composite from an obvious paste job.

Relight this photo with warm window light from the left, gentle shadows, natural skin tones. Change nothing else.

The relight is the most useful everyday edit. Most mediocre photos are one lighting change away from good.

Fixing the four most common failures

The face changed. Your keep line was missing, or your style words were stronger than it. Add "do not alter my facial features" and delete the most dramatic style word. Repeat until it holds.

It looks plastic. The model over-retouched. Say "light retouch only, keep real skin texture, no smoothing". Never ask for "flawless" anything.

It ignored part of the request. You asked for too much at once. Split the prompt into passes: background first, then light, then outfit. Each pass stays accurate because the model has one job.

Each retry looks worse. You are editing an edit. Go back to the original photo and start again with the improved prompt. Generations degrade like photocopies.

Gemini or ChatGPT?

Same prompts, different personalities. Gemini is usually better at holding a face steady on the first attempt and is the tool behind most of the viral portrait trends. ChatGPT is better at conversation: you can nudge the result over several messages without re-explaining. Run your first few edits in both and see which fits how you work. The prompt structure carries over unchanged, and tool-specific lists live on our Gemini and ChatGPT pages.

Where to go next

Once editing feels comfortable, the next skill is generating images from scratch, which uses a slightly different structure. Our guide to the anatomy of a perfect AI image prompt covers that. And when an edit fights you, the answer is almost always in the keep line, so start your debugging there.