Turn “Make It Better” Into a Specific Edit

“Clean up the background,” “make the person look more energetic,” and “make this feel more premium” do not define a single edit target. Decide whether the change is to a background object, the person’s pose, the lighting, or the composition. Handle unrelated goals separately so one edit does not redo parts you already like.

What to specifyFill in
Base imageWhich selected image version to use
Target regionThe object, its location, and nearby references
Requested changeThe visible result after removal, replacement, or adjustment
Preservation constraintsWhich identity traits, objects, spatial facts, composition, and lighting stay
Related effectsHow to reconstruct the area the object occluded; whether its shadow or reflection also needs editing
Scope of useWhether the change applies only to this image or to related images too

For a small revision, describe the scope in plain language. You do not need to create an asset ledger for every minor edit.

Protect Details Likely to Change as a Side Effect

“Keep everything else unchanged” can summarize the request at the end, but first name the important facts. When editing near a person, list facial features, hairstyle, clothing, and pose. For a background edit, specify wall structure, door and window positions, nearby objects, and light direction.

Preservation constraints must also be compatible with the requested change. If you ask to remove a box from a table, you cannot also require the pixels the box covered to remain exactly the same; that area needs to be filled so the tabletop continues naturally. What you can preserve includes the table’s structure, grain direction, perspective, and areas that were not occluded.

Likewise, changing the light direction while requiring every shadow to stay exactly as it was creates a conflict. If an edit affects lighting or spatial structure across the image, reassess its scope instead of presenting a major reconstruction as a small touch-up.

Original Example: Remove an Unwanted Folding Umbrella From a Poster

This is a fictional editing brief: on a reading-event poster, a speaker sits on a bench with a cloth bag to the right. In the lower-left corner, a folding umbrella leans beside the bench’s front leg, covers part of a slat, and casts a small shadow on the wall behind it. The person, bag, and bench are already right; the event image does not need an umbrella.

A vague prompt would be: “Remove the clutter, make the background cleaner, and keep the person unchanged.” It does not identify what counts as clutter or protect the bag and bench.

Example local-edit prompt:

Use the selected original poster as the editing base. Remove the folding umbrella in the lower-left corner beside the bench’s front-left leg, and remove the shadow it casts on the wall behind it.

Reconstruct the bench slat that was covered by the umbrella, matching the direction, width, and perspective of adjacent slats. Fill the newly exposed wall area to match the surrounding color and lighting; add no new decoration.

Preserve the speaker’s facial features, hairstyle, clothing, and seated pose. Keep the cloth bag on the right in its current position, outline, and color. Preserve the bench legs, other slats, the person’s shadow, the crop, and the direction of light. Make no other changes beyond removing the umbrella and its shadow and reconstructing the area the umbrella covered.

This treats “remove the umbrella” and “fill the background it covered” as one related edit. If the original image has no umbrella shadow, remove that instruction; inspect the actual image instead of using a template to invent details. This example demonstrates wording only; it does not claim an edit was completed.

Afterward, Check the Requested Change and Preservation Constraints Separately

First check whether the intended change happened: is the umbrella gone, is the occluded area continuous, and is there any isolated shadow left? Then inspect details you did not want to change: did the person’s face drift, was the bag erased too, is a bench leg distorted, or has the crop changed?

DifferenceNext request
The object is gone, but its shadow remainsPoint to the remaining shadow and preserve the correctly reconstructed bench
A nearby object was removed by mistakeReturn to a suitable base that still contains it and narrow the target description
Reconstructed texture or perspective does not alignIdentify the seam and texture direction; do not redesign the whole background
The person’s features also driftedChoose the base again and state the preservation constraints; if needed, use a tool that supports the required edit scope
The new version was accepted, then an extra change is requestedState the added scope first so old and new instructions do not overwrite each other

Choose an editing tool and region-selection method based on the tool you actually use. Wording cannot replace region selection or guarantee that a model will lock every unselected pixel.

Preview the Difference Before Replacing the Current Prompt

If you are revising an image prompt in a project, compare the current image and wording first, then write a short preview:

Before: ___.

After: ___.

Preserve: ___.

Related effects: ___.

Scope of use: ___.

For example, if only the poster changes, the umbrella removal applies to that image alone; characters in the related short drama still follow their approved prop state. If a later shot still shows the character carrying an umbrella, the image edit affects story continuity. Confirm which shots and settings need to change together.

If someone changes the image or prompt after your preview, reread the current version before merging this edit. Do not let an old preview overwrite accepted new content. Replace the relevant entry only after the creator confirms scope; do not smuggle unapproved composition, identity, or story changes in as “wording improvements.”

FAQ

Why can the person change even when I say “keep everything else unchanged”?

The phrase states an intention but does not identify the specific traits to protect, and it cannot guarantee that the editing tool will make a strictly local change. Name the person’s features and nearby objects, check what region-editing method is available, and compare the actual output against the preservation constraints.

Can I change several things at once?

Yes, when the areas are linked to the same change, such as an object and its shadow. Separate unrelated goals—changing a face, background, clothing, and lighting—so their scope is clear. If they are intentionally part of one overall redesign, state that the edit is a single integrated revision.

Do it with the skill

With the current image, relevant visual references, and existing image prompt ready, use /short-drama-image-prompts (or $short-drama-image-prompts in Codex) to write the change and preservation constraints for this local edit, then preview the revision. Send identity, geography, or story-fact changes back to the settings layer for approval. This step revises prompt text only; it does not generate an image.

Read the method: Method used in this article

About Drama SkillsThe short-drama-image-prompts skill on GitHub

All guides