The most useful photo-editing prompt is rarely the most dramatic one. It is the prompt that names a small change and makes the boundaries of that change easy to inspect.

Use four parts

A practical prompt can be built from four parts:

1. Action: what should be removed, replaced, repaired, or adjusted.
2. Target: the exact subject or region the action applies to.
3. Reconstruction: what should appear behind or around the changed region.
4. Constraints: what must remain unchanged.

For example:

Remove the two people behind the main subject. Reconstruct the hidden pavement and shopfront using the same perspective, focus, and light. Keep the main subject, crop, signs, and remaining pedestrians unchanged.

The first sentence identifies the edit. The second describes the missing visual information. The third gives you a checklist for comparing the output with the source.

Name details that models tend to regenerate

“Keep everything else unchanged” is useful, but it is often too broad. Name the details that matter for the image in front of you:

• For a portrait: identity, expression, face shape, hairline, skin texture, hands, and jewelry.
• For a product: geometry, label text, logo, material finish, color, ports, seams, and included parts.
• For architecture: straight lines, brick or tile spacing, windows, signs, perspective, and shadows.
• For a background replacement: subject edges, transparent areas, light direction, horizon, depth of field, and contact shadow.

These constraints do not guarantee preservation. They make unwanted changes easier to notice and give the next correction a clearer target.

Start short, then add constraints

Use a short prompt for an isolated, low-risk edit. If the first output changes too much, move to a detailed prompt that locks the important details. Adding every possible instruction before seeing a result can make the request harder to reason about.

A useful sequence is:

1. Make one edit.
2. Compare the output with the original at full size.
3. List the unwanted changes precisely.
4. Retry with those details added as constraints.

Describe the hidden surface

Object removal requires the model to invent what was behind the object. “Remove the bin” leaves that reconstruction open. A stronger version describes the wall, pavement seams, texture scale, weathering, and shadows that should continue through the removed area.

The same principle applies to background replacement. Specify camera height, horizon, depth of field, light direction, and the shadow connecting the subject to the new setting. Without those cues, the new background may look like a separate layer.

Check facts, not just appearance

An output can look polished while changing something factual. Before publishing, compare:

• faces and identity;
• text, labels, and logos;
• product dimensions and components;
• hands, teeth, eyes, and jewelry;
• edges around hair, glass, fur, or reflective materials;
• repeated patterns and straight lines;
• the rights and permissions attached to the source and output.

PromptEdit’s task prompt library organizes these checks around common editing jobs. The prompts are starting points; they are not deterministic presets or a substitute for reviewing the result.