To preserve a person or scene while editing an image, state the change you want, then explicitly name what must stay the same. Keep each edit focused, define its boundaries, and identify the role of every reference image. These instructions make a prompt clearer, but they cannot guarantee that an image-editing model will preserve every detail exactly.
Start with one specific change
Tell the editor what to alter: replace a jacket, remove a background object, or move a subject. Avoid combining unrelated changes in one request. A focused instruction makes it easier to see whether the intended edit worked and what needs correction.
For a local change, name the target and its surroundings. For example, ask to remove the cup from the table while keeping the person’s hand, table surface, and objects beside it unchanged.
List the details that must remain fixed
Do not assume the editor will infer what “keep the rest the same” means. Spell out the important visual invariants in concrete terms.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitches#1 Best Overall
For a person
- Identity and likeness, including facial features and skin tone
- Body shape and proportions
- Pose, expression, and gaze
- Hairstyle
OpenAI’s image-prompting guide demonstrates a clothing edit that names these attributes while asking to match the source image’s lighting and shadows. It also recommends describing framing, relative scale, gaze, and interactions with objects when relevant.
For the scene or composition
- Background, camera angle, and framing
- Lighting, shadows, and color
- Layout, labels, and surrounding objects
For a precise edit, identify which nearby details should not change. If the edit involves text or product imagery, name labels and product features that need to remain intact.
Set boundaries around the edit
Say “change only” the target element, then mention unwanted additions that matter to the task, such as extra text, accessories, logos, or watermarks. The more specific the boundary, the clearer the request—not a guarantee that the model will obey it perfectly.
For example: “Change only the coat to a dark green wool coat. Preserve the person’s identity, face, hairstyle, pose, expression, and proportions. Keep the background, framing, lighting, and shadows unchanged. Do not add accessories, text, or logos.”
Give each reference image a clear role
When using multiple images, label them by number and explain what each contributes. Say which image is the base scene, which provides a person’s identity, and which supplies clothing or style. Also state where the referenced element belongs.
For example: “Image 1 is the base scene; Image 2 is the clothing reference. Apply the clothing from Image 2 to the person in Image 1. Preserve the person’s identity, pose, framing, and background.” OpenAI’s GPT Image Generation Models Prompting Guide provides additional official prompt-structure guidance.
Use this prompt pattern
Adapt this template to the edit rather than treating it as a fixed formula:
Edit the supplied image to [one specific change]. Preserve [identity and/or subject attributes], [pose and expression], and [composition and scene details]. Change only [target element]. Keep [lighting, shadows, and color] consistent with the source. Do not add [specific unwanted elements].
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchSpecial offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.
For a local edit, include the nearby details that must stay untouched. For multiple inputs, add a short reference key before the instruction to assign each image its role.
Inspect the result and correct one thing at a time
- Check whether the requested element changed.
- Inspect the details you asked to preserve, especially likeness, pose, expression, and scene composition.
- If something drifted, pass the result into another edit and request one targeted correction.
- Repeat the critical preservation constraints in the follow-up prompt.
For instance, if a clothing change also altered the pose, ask specifically to restore the original pose while retaining the new clothing. Avoid bundling that correction with another unrelated edit.
What prompt wording can—and cannot—do
OpenAI’s guide advises: “For edits, say ‘change only X’ and list the details to preserve, such as identity, geometry, layout, lighting, or labels.” That is useful guidance for making the request explicit, not evidence that particular wording guarantees an exact result. Image-editing consistency remains an active research problem; Bai et al.’s Edicho: Consistent Image Editing in the Wild (ICCV 2025) addresses that challenge. The reviewed sources do not establish a general success rate for identity-preserving prompts.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.If you are choosing an editing tool
Do not judge a tool by a prompt template alone. Compare how it handles the parts of the task that matter to your image:
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Best Value
- Preserving identity during changes to a person
- Making a local edit without altering nearby content
- Accepting multiple reference images and following their assigned roles
- Avoiding unwanted changes
- Allowing you to correct drift across follow-up edits
OpenAI’s prompt guide identifies instruction following, identity and product preservation, text accuracy, unwanted changes, and transparency as evaluation areas. This is a checklist for comparing tools, not independent proof that one model performs best.
API-specific image input limits
If you are using the GPT Image API, its image edit API reference lists PNG, WebP, and JPEG inputs under 50 MB. These are requirements for the documented API, not universal limits for image-editing tools; check the current API reference before uploading.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
