AI Image Editing Prompts

September 15, 2026

Most AI edits go wrong for one reason: the prompt asks for a change but never says what has to stay. Tell a model to "remove the background" and it will happily redraw the hairline while it is at it. The fix is unglamorous but reliable — name the change, name what to keep, add one hard constraint. That is the whole method behind the prompts below.

Here are 50 copy-paste AI image editing prompts for retouching, background work, color grading, inpainting, restyling and upscaling, plus the formula used to write them. If you need new images rather than edits, start with our portrait prompts for men and women or the photoshoot prompts guide; if you shoot for a living, prompts for photographers is the closer fit.

50+ Image Editing Prompts

Every prompt below assumes the image is already attached and that you will swap the subject details for your own. Keep the "keep" clause. It is the part that stops the model inventing a new face, jacket or sky.

Retouching and Skin Prompts

Skin work is where editing prompts get judged. If the retouch removes texture, the photo reads as fake the second someone pauses on it — the same trap that ruins half the images in a portrait set. Compare the result against the untouched frame before you export, and browse the photography library when you need a reference for how much texture should survive.

Background Replace and Cleanup Prompts

Background swaps fail at the edges: hair, glass, thin straps. Zoom to 200 percent before you accept a result and check that the rim light still points the same way as the fill. When a client needs a whole new scene rather than a new backdrop, the cinematic fashion portrait prompt and the luxury advertising portrait prompt are better starting points than a cleanup prompt.

Color Grading and Light Prompts

Grading prompts land harder when they name a reference: a film stock, a time of day, a known palette. Vague mood words get you a filter, a reference gets you a look — the same logic behind the flat-color studies in the abstract library. For a darker register try the dark mood cinematic prompt.

Object Removal and Inpainting Prompts

Inpainting prompts live or die on what surrounds the removal. Always name the surface being rebuilt — fabric, wood, sky, tile — and the light direction in play, otherwise you get a smooth patch that looks airbrushed. Texture-led references such as this texture portrait prompt are useful for calibrating how much surface detail to put back.

Style Transfer and Relight Prompts

Style transfer keeps the composition and throws the materials away, which is exactly how a photograph turns into a minimalist watercolor illustration or a flat watercolor portrait. Watch the face: restyling prompts drift identity faster than any other edit, and heavy styles bury expressions you wanted to keep.

Upscale and Detail Enhance Prompts

Upscaling is the last step, not the first. Clean, retouch and grade before you upscale, or you bake the grain and the dust into four times the detail. Teams shipping edits at volume keep one house-style prompt and reuse it: a watercolor storyboard prompt for soft content, a 3D product photography prompt for commerce work.

The Editing Prompt Formula

Three parts, always in this order. Copy the shape and swap the words.

  1. Keep: the identity, texture, light and framing that must survive. "Keep the freckles, the expression and the window light from camera left."
  2. Change: one action with its object and an amount. "Reduce the cheek redness by about 30 percent." "Replace the cluttered background with a seamless grey backdrop."
  3. Constrain: the specific failure you are steering away from. "No plastic smoothing. No halo around the hair. No color shift in the shadows."
Keep the freckles, the under-eye creases and the soft window light. Retouch the temporary blemishes, even out the skin tone on the neck, leave every pore intact. No plastic smoothing, no eye reshaping, keep the original grain and color.

A background swap uses the same three parts: keep the rim light and the shadow direction, replace the room with a seamless grey backdrop, no halo around the hair. If a prompt has a keep clause and a constraint it will survive a rerun. If it has neither, you are rolling dice.

How to Write AI Image Editing Prompts

Six habits separate prompts that work on the first try from prompts you rerun ten times.

Keep your best AI photo editor prompt in a text file with the subject words in brackets, so the next job is a two-second edit instead of a fresh start. And check whether the task belongs in an editor at all: anything with readable text, layered design files or precise geometry usually belongs in a layout tool rather than a chat window.

FAQ

Can an AI editor keep the same face while replacing the background?

Usually, if you say so. Add "keep the subject's face, hairline and expression exactly as in the original" and avoid words like "portrait" or "model" that push the model toward generating a new person. If the face still drifts, mask it and edit only the background.

Do these editing prompts work in ChatGPT, Gemini, Midjourney and Nano Banana?

Yes. All of them take an attached image plus a text instruction. Midjourney leans harder on short phrasing and image weights, while ChatGPT and Gemini tolerate longer conversational prompts. For Midjourney, cut the prompt roughly in half.

How do I keep a retouch from looking plastic?

Never write "smooth skin" on its own. Write "even tone while keeping pores, freckles and fine lines", then add "keep the original grain". Texture words are what keep a retouch believable, and they matter more than any slider.

What is the difference between an editing prompt and a generation prompt?

A generation prompt describes a whole new image. An editing prompt describes a delta: what changes, what stays, and what the result must not do. Editing prompts are shorter and always more specific about preservation.

Can I stack several edits in one prompt?

Two or three, yes — retouch plus grade plus background. Past that, quality drops and the model starts ignoring clauses. Do the structural change first, then grade and upscale in a second pass with the cleaned image as input.

Browse AI image prompts by model