Why does the rest of the image change when I only asked for one edit?

Because an edit generates a new image conditioned on the old one rather than patching a region. Everything is in play unless an instruction pins it.

Can I remove several objects at once?

You can, but each removal is another region being regenerated and the drift compounds. One per turn is more reliable.

How do I keep a region truly pixel-identical?

You cannot, by prompting alone. OpenAI's guidance is to composite the approved edit back into the original when a region must not move at all.

Should I say what fills the gap?

Only when it is ambiguous. Left unstated the model infers from the surroundings, which is usually right for a simple background and unreliable in front of something complex.

GPT Image 2.5 prompt — remove an object without touching the rest

Nine words. Name the object, name its owner, forbid everything else.

The same photograph with the flower removed from his hand and nothing else changed

The original photograph of a man in a blue hat holding a flower

InputResult

Settings the Remove one thing and leave everything else alone prompt ran at
Mode	Model	Quality	Size	Ratio	Credits
Image to Image	Think	Standard	1K	2:3	10 cr

Published by OpenAI as an example for this model — not generated on this site. Model: gpt-image-2.5-sunburst. Source: OpenAI The published example ran at 1024x1536, a legal size we do not currently offer; the frame here is the same shape at our nearest size.

"Remove the flower from man's hand. Do not change anything else." The shortest edit in the library, and it is short because a removal has exactly two failure modes: removing the wrong thing, and quietly regenerating the person while you were not looking. One clause closes each.

In full

The prompt, nothing cut

Remove the flower from man's hand. Do not change anything else.

63 characters

Run it here

This exact prompt, at these exact settings

The mode and all five settings are locked to this template, so there is nothing to line up by hand. Press Run this prompt above and the panel opens filled in — prompt, mode, quality, size and frame — and your first image needs no account. If your plan sits below what this ran at, the line under the panel says so.

This example was published at Think · Standard · 1K. A free run is Fast · Standard · 1K, so expect less fine detail and softer small text. The composition is the same; the finish is not.

3 levers

What is safe to change

Most libraries publish this list and stop, which is why so many copied prompts come back worse than the original: the swappable nouns are the safe half. The three sections under it are the other half — the clauses that are doing the work, the ones that break it if you touch them, and why it is written in this order.

The possessive
"from man's hand" is not decoration — it disambiguates. If there were flowers elsewhere in the frame, this clause is the only thing telling the model which one you meant.
The prohibition
"Do not change anything else" is half the prompt by word count, and it is why the man's face survives. An edit regenerates the frame; anything unpinned is negotiable.
What replaces it
Unstated here, and the model infers from the surroundings. Say it when the answer is not obvious — "leave his hand open and empty", "replace it with the wall behind".

Three ways to break it

Each one is a real failure with a reason attached. A rule without a reason is not usable.

01Naming the object without locating it
"Remove the flower" in a frame with more than one flower is a coin toss, and the model will not ask. Locate it — whose hand, which side, what colour.
02Removing several things in one pass
Each removal is a region the model regenerates. Three at once is three chances for the surroundings to shift, and the drift compounds. One per turn.
03Expecting the untouched region to be pixel-identical
It will not be. The whole frame is regenerated, so faces and textures are resampled even when preserved. OpenAI's own guidance says that if a region must remain pixel-identical, composite the approved edit back into the original.

Why it is written this way

Inside the Remove one thing and leave everything else alone prompt

The shortest prompt here, and worth studying precisely because there is nowhere for anything unnecessary to hide. Nine words, and both halves are load-bearing.

The first half locates. "The flower from man's hand" rather than "the flower" — a possessive that costs three words and removes an ambiguity the model would otherwise resolve silently. This generalises: every removal prompt should name the object *and* something about where it is, because a model that picks the wrong one does not tell you it was choosing.

The second half is where the interesting behaviour is. "Do not change anything else" is fifty per cent of this prompt, which seems disproportionate until you understand what an edit actually is. The model is not erasing a region and inpainting it. It is generating a new image conditioned on the old one. That means every pixel is in play on every edit, and the parts that survive unchanged survive because nothing pushed them to change — not because they were protected.

That reframing explains a lot of confused reports about image editing. People ask for a small change and find that the subject's face is subtly different, and conclude the model is unreliable. It is behaving exactly as designed; there was simply no instruction saying the face mattered. The prohibition supplies one.

It also explains the limit, which OpenAI states plainly and which is worth repeating because it saves people a lot of retrying: repeated edits can still change details you intended to preserve, and if a region has to be pixel-identical, the answer is not a better prompt — it is compositing the approved edit back into the original in an editor. Prompting reduces drift; it does not eliminate it. For a face in a paid campaign, that distinction is the whole job.

4 questions

Remove one thing and leave everything else alone — common questions

Why does the rest of the image change when I only asked for one edit?: Because an edit generates a new image conditioned on the old one rather than patching a region. Everything is in play unless an instruction pins it.
Can I remove several objects at once?: You can, but each removal is another region being regenerated and the drift compounds. One per turn is more reliable.
How do I keep a region truly pixel-identical?: You cannot, by prompting alone. OpenAI's guidance is to composite the approved edit back into the original when a region must not move at all.
Should I say what fills the gap?: Only when it is ambiguous. Left unstated the model infers from the surroundings, which is usually right for a simple background and unreliable in front of something complex.