To make an AI image edit that changes only one thing, write a prompt in two parts: what to change, and a list of everything that must stay the same (face, pose, background, lighting, framing). Then make one change per request.
The copy-paste prompts below do this for outfits, backgrounds, lighting, removing objects and fixing hands. Even so, AI editors can redraw more than you asked, so compare each result with the original.
Asking an AI to “change the jacket” is like asking a painter to repaint a portrait with a new jacket. Unless you say “and do not touch the face”, they may tidy the face while they are at it. A good edit prompt is a short work order: one change, and a list of what is off limits.
- Change, then keep: “Change only X. Keep A, B, C exactly the same.”
- One change per request. Several at once is where faces and poses drift.
- Repeat the keep list every time. The tool does not remember it reliably.
- Select the area when your tool lets you. It limits what gets redrawn.
- If it drifts, go back to the original rather than editing the edit.
- Change the outfit only new clothes on the same person, in the same pose and light.
- Change the background only a new place behind the same subject, with light that matches the person.
- Change the lighting only different light or time of day, with nothing added, removed or moved.
- Remove an object take something out and fill the space with what would naturally be behind it.
- Fix hands repair one hand in the same position; select just the hand if your tool allows it.
- Change one small detail hair colour, a mug, a sign or an expression, with everything else kept the same.
Every prompt follows the same pattern. If a face or pose still changes, see why the face or pose changes anyway.
The pattern behind every prompt here
The plain point: name the one change, then list what must not change.
This is how vendors write their own examples. OpenAI’s prompting guide says to say “change only X” and list the details to keep, and its clothing example keeps the face, body, pose and identity off limits.1 Google’s developer docs for its Nano Banana image models end an edit example with “Do not change any other elements of the image.”2
One change, one keep list. Every prompt below is this template with the blanks filled in.
Edit prompts for common jobs
The plain point: copy the box, change the parts in [brackets], and attach the image.
Change the outfit only
Change the background only
The last line matters. A new background with light coming from a different side is what makes a cut-out look pasted in.
Change the lighting only
Words for light are in colour, light and mood.
Remove an object
OpenAI’s own example of this is two short sentences: “Remove the flower from man’s hand. Do not change anything else.”1
Fix hands
If your tool lets you select an area, select just the hand first. Hands are small, and a whole-image redraw can change more than the hand.
Change one small detail
Say exactly where the change is (“the left hand”, “behind the woman on the right”), and keep everything else on the list.
Why the face or pose changes anyway
The plain point: most chat image editors redraw the image, and each round can move things a little.
OpenAI says it plainly: repeated edits can still change details you meant to keep, and if part of an image must stay pixel-perfect, paste the approved edit into the original instead of relying on the prompt alone.1 Even a drawn selection is a guide, not a wall: OpenAI says its models use a mask as guidance but may not follow its exact shape.3
What to do about it:
- Re-anchor. Paste the full keep list again with every request. OpenAI’s ChatGPT guide says to repeat any requirement that must stay fixed.4
- Make smaller edits. The same guide says to adjust one element at a time so other details do not drift.4
- Select the area. Marking just the part to change is called inpainting. Google’s docs describe a spoken version: name the part in words and the rest is left untouched.2
- Go back to the original. If a face has drifted after three or four rounds, start again from the first image with a combined, careful prompt.
- Composite. For work that must be exact, take only the changed area into a photo editor and lay it over the original.
Drift is normal, not your fault. Repeat the keep list, shrink the edit, select the area, and restart from the original when needed.
How to do it in each tool
The plain point: chat tools take the whole image plus your words; selection tools redraw only what you mark.
| Tool | How to edit | Worth knowing |
|---|---|---|
| ChatGPT | How: attach the image and paste the prompt, or select one area of an image and describe the change for it.4 | Worth knowing: OpenAI’s example keep list names the person, layout, lighting, colours and crop.4 |
| Gemini (Nano Banana) | How: open the image or upload one from your device, click Chat, enter the prompt.5 | Worth knowing: Google says the model matches the original’s style, lighting and perspective when you add or remove things.2 |
| Photoshop Generative Fill | How: select the area with any selection tool, then type what should appear, or leave it blank to fill from the surroundings.6 | Worth knowing: the result sits on its own layer, and you can add a reference image of what you want, such as a jacket.6 |
| Video edits | How: same pattern, shorter. Google’s Gemini Omni docs say simple prompts work best for video editing. | Worth knowing: add “Keep everything else the same”.7 |
With Photoshop, the prompt only needs to say what should appear inside your selection, because Generative Fill fills the selection you made. Midjourney’s help pages blocked our reader on 12 Oct 2026, so its editor is not described here.
Questions people ask
How do I change the clothes in a photo without changing the face?
Use the outfit prompt above: “change only the clothing”, then list face, hair, body, pose and background as fixed. If the face still changes, select only the clothing area in a tool that allows it, such as ChatGPT’s area selection or Photoshop’s Generative Fill.
Why does AI change the whole image when I ask for one edit?
Chat image tools usually redraw the whole picture, so details can move. OpenAI says repeated edits can still change details you meant to keep, and advises restating them and checking each result.1
Should I write “do not” in an edit prompt?
For edits, yes: OpenAI’s own edit examples use it.1 That is different from a negative prompt list of unwanted things, which some generators handle badly.
Can I edit a photo of someone else?
Get their permission first. OpenAI’s guide asks you to confirm you have permission to use a real person’s likeness, and Google says editing images in the Gemini app is not available to users under 18.45
Where to go next
- “It is an old family photo.” Use old photo restoration prompts.
- “I need the same character across many images.” See identity lock prompt templates and character consistency.
- “What is inpainting, exactly?” Read inpainting and outpainting.
- “I want it sharper after editing.” See upscale prompts.
- Related: How Does Image-to-Image AI Work? Img2img, Strength and Edits Explained.
- Related: Inpainting vs Outpainting.
- 1OpenAI, Image prompting, undated, read at source 12 Oct 2026“For edits, say ‘change only X’ and list the details to preserve, such as identity, geometry, layout, lighting, or labels.” and “Do not change her face, facial features, skin tone, body shape, pose, or identity in any way.” ↑a“Remove the flower from man’s hand. Do not change anything else.” ↑b“Repeated edits can still change details you intended to preserve.” and “If a region must remain pixel-identical, composite the approved edit into the original image” ↑c“Restate those constraints and inspect each result.” ↑d“Remove the flower from man’s hand. Do not change anything else.” ↑e
- 2Google AI for Developers, Nano Banana image generation, undated, read at source 12 Oct 2026“Do not change any other elements of the image.” ↑a“Conversationally define a ‘mask’ to edit a specific part of an image while leaving the rest untouched.” ↑b“Provide an image and describe your change. The model will match the original image’s style, lighting, and perspective.” ↑c
- 3OpenAI, Image generation, undated, read at source 12 Oct 2026“The model uses the mask as guidance, but may not follow its exact shape with complete precision.” ↑
- 4OpenAI, ChatGPT Learn, Image generation, undated, read at source 12 Oct 2026“Repeat any requirement that must stay fixed.” ↑a“Adjust one element at a time so the composition and other important details do not drift.” ↑b“Attach a reference image to the composer when you want ChatGPT to edit it” and “You can also select a specific area of an image and describe the change for that area.” ↑c“Preserve the person, desk layout, lighting, colors, crop, and every other detail exactly.” ↑d“confirm that you have permission to use their likeness.” ↑e
- 5Google, Gemini Apps Help, Generate & edit images with Gemini Apps, undated, read at source 12 Oct 2026“Upload an image and ask Gemini to make edits.” and, among its features, “Making quick and specific changes to parts of an image.” ↑a“Editing images isn’t available to users under 18.” ↑b
- 6Adobe, How to use Generative Fill in Photoshop, undated, read at source 12 Oct 2026“Use the Selection Brush tool, or any other selection tool, to select an object or area in your image.” and “Or leave it blank and Photoshop will fill in the selection based on the surroundings.” ↑a“a Generative Layer will be created in your Layers panel for nondestructive editing of your original image.” and “click on Reference Image and upload an example image containing desired content, like a jean jacket.” ↑b
- 7Google AI for Developers, Generate and edit videos with Gemini Omni Flash, undated, read at source 12 Oct 2026“Simple prompts work best for video editing. Overly descriptive prompts can lead to unintended changes.” and “When editing a specific aspect of the video, include ‘Keep everything else the same’ to maintain visual consistency.” ↑
The comparison on this page (the painter and the work order) and all prompts are this site’s own, not claims from these sources. None of the prompts carries a recorded test date.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK