Correct a photo edit in iMini

Keep every approved part of the current image unchanged.
Fix only [one visible error]. Restore [protected detail] from the original upload.
Do not redesign the face, product, label, background, camera angle, or composition.
Do not invent detail that is not visible in the source.

A background can be right while a face, label, or edge is wrong. Keep both the original upload and the current candidate available, then identify the exact defect. This guide focuses on the next instruction after a failed edit; for a first-pass command, use the task-specific command library.

What Makes a Follow-Up Command Different?

A follow-up command names a visible error in the current result and limits the repair. Identify the original as the source of truth and the candidate as the image to correct. Asking for “a better version” can restart the interpretation and disturb details that were already approved.

Separate the Defect, the Source, and the Locked Areas

Reusable formula: "Edit this image by changing [specific area] into [desired result]. Keep [protected elements] unchanged. Match [lighting, perspective, and style]. The image will be used for [final use case]. Avoid [common mistakes]."

The original formula above is useful for a first pass. For a repair, add the visible error and the reference to restore from. A preservation instruction is not a guarantee; compare the new result before approving it.

Repair a Background Edit That Changed the Subject

If the background is approved, do not ask for another new scene. The first-pass examples below show what was requested; use the follow-up afterward to recover the subject without redesigning the background.

Portrait background: "Replace only the background with a bright modern office. Keep the person, face, expression, hairstyle, clothing, pose, and camera angle unchanged. Match soft daylight from the left and keep the result realistic."

Product background: "Change the background to a clean e-commerce studio scene. Keep the product shape, label, logo, color, material, size, and proportions exactly the same. Add a realistic soft shadow and do not add extra text."

For more dedicated examples, read ChatGPT background change prompts.

The background is approved, but [hair / product edge / contact shadow] is incorrect. Correct only that area using the original reference. Keep the subject interior, camera angle, approved scene, and all accurate details unchanged. Do not invent missing detail.

Restore a Product Label, Shape, or Shadow

Compare the candidate with the original label and silhouette before judging the scene. The table and initial request below define what should have survived; a corrective request should restore one failed detail, not repeat every possible improvement.

Marketplace command: "Turn this product photo into a clean marketplace image. Keep the product dimensions, label text, logo, color, material, and packaging unchanged. Remove dust and clutter, place it on a light background, and add a realistic shadow."

The [label wording / cap / bottle silhouette] differs from the original upload. Restore only that detail from the original. Keep the approved background, product placement, scale, camera angle, and shadow unchanged. Do not guess unreadable text. If the source does not show it clearly, flag the limitation instead of inventing it.

Restore Identity After a Portrait Edit

After a portrait edit, compare identity, expression, face shape, hairline, and skin texture with the original. If they drifted, use the original as the identity reference rather than treating the changed output as the new face.

Headshot command: "Create a professional profile photo from this portrait. Keep identity, face shape, expression, hairstyle, clothing, and pose unchanged. Improve lighting, clean the background, keep natural skin texture, and avoid an airbrushed look."

Restore the person’s original facial features, expression, face shape, and natural skin texture from the original upload. Keep the approved clothing, pose, background, and framing. Do not beautify, change age, or redesign the face.

Repair Smears Left by Object Removal

Inspect the area where an object was removed. The initial request below describes a clean removal; if it leaves a smear, correct that small area rather than removing or rebuilding more objects.

Cleanup command: "Remove the bottle on the table. Reconstruct the table surface with matching wood grain, shadow, and perspective. Keep every other object in the same position."

Repair only the smeared area where [object] was removed. Continue the adjacent [wood grain / wall texture] with matching perspective and light. Keep every remaining object and all approved details unchanged.

Recover Text That an Edit Distorted

When a logo or label changes, a plausible replacement is still wrong. Use the original as the reference for the exact wording and shape; never ask the model to infer text that the source does not show.

Text-safe command: "Edit only the lighting and background. Preserve all visible text, logo shapes, UI labels, spelling, and layout exactly as they appear. Do not translate, rewrite, or stylize any text."

For added headlines, set final type in a design tool. For an existing product label, use the original product or label layer in a non-generative composite when exact preservation is required.

The [logo / visible label text] changed during the edit. Restore it from the original reference without rewriting, translating, or stylizing it. Keep the approved scene and all other accurate details. Do not invent characters that are unclear in the original.

Fix a Crop Without Redesigning the Image

If a crop cuts off the product or leaves no room for copy, first try ordinary canvas padding or a different crop. The initial layout requests below are references, not a reason to regenerate an otherwise approved product.

Blog hero command: "Edit this image into a 16:9 blog hero. Place the main subject on the right third, leave clean negative space on the left for a headline, use bright natural lighting, and avoid tiny distracting details."

Social crop command: "Create a 1:1 social media version. Keep the subject centered, simplify the background, increase contrast slightly, and leave enough margin so the image still works after feed cropping."

Keep the approved subject unchanged and fully visible. Add uncluttered background space on the [left / top] for the layout rather than stretching or cropping the subject. Continue the existing perspective and lighting. Add no text or extra objects.

Choose the Smallest Useful Follow-Up

A Practical Workflow

Compare: open the original and candidate side by side. Identify: name one visible error and mark the affected area if the chosen editor supports it. Correct: provide both reference roles and state what is locked. Review: check the repaired area and the previously approved details again.

Stop when detail keeps drifting. Return to the clean original or use a non-generative composite for an exact label or face. Repeated requests on an already damaged candidate can compound the problem.

How to Use These Commands with iMini

Open the iMini image editor, provide the reference images supported by the selected model, and clearly label which image is the original and which is the candidate. Ask for one correction. Do not assume the interface offers a mask or multiple inputs unless those controls are available.

Need a first-pass instruction instead? Use short commands by task, or follow the complete upload-to-review workflow.

FAQ

What should I say when an AI photo edit changes the face?

Identify the original as the identity reference, request restoration of the changed facial details, and lock the approved background, pose, and framing. Review the result; the prompt cannot guarantee identical facial pixels.

Can a follow-up recover unreadable label text?

Not reliably if the source is unclear. Supply a clearer original or use the original label layer. Do not accept invented text because it looks convincing.

Should I rewrite the entire prompt after a small error?

Usually start with one focused correction and repeat the important constraints. Restart from the clean original if the candidate contains several linked defects.

When should I stop iterating?

Stop when corrections repeatedly alter approved details or cannot reproduce critical product or identity information. Use manual retouching or a non-generative composite instead.

Try a focused photo correction in iMini