

Choose a current editing application that supports the specific local repair you need, and write down which facial details must remain unchanged. A restrained request might address a temporary blemish, local tone or a distracting flyaway; compare facial proportions, expression, hairline, clothing, and background with the source, and rerun only if the result is not acceptable.
Separate a local repair from a style transformation
The former four-treatment Portrait Repair screen is not a current Imagild entry point. Do not infer dedicated local-retouch controls from the Image Skill Market. Select an editor that can perform the intended local correction and record which areas should remain untouched. The editorial visual marks the review area rather than claiming an assured before-and-after result. Keep the source photo as the reference for likeness, expression, skin texture, accessories, hands, clothing, and scene continuity.
A controlled three-step workflow
Choose only the repairs the image needs
Choose only the relevant repair in the editor you actually use; verify which controls it currently provides. A small temporary blemish, uneven under-eye tone, a few distracting flyaways, or a local color imbalance are reasonable targets. Do not use a repair request as a vague instruction to make someone younger, thinner, more symmetrical, or like a different person.
Generate one restrained result at a time
Review a proposed result against the source before making another change; check your chosen editor’s current processing and saving behavior. Compare it with the source before continuing. If it smooths too much, changes the mouth or eyes, shifts the hairline, or alters jewelry and clothing, reject it and rerun deliberately rather than accepting it as final.
Compare at face scale and full-frame scale
Zoom in to inspect pores, eyelids, lashes, teeth, hair edges, and the transition between repaired and untouched skin. Then zoom out to check expression, head shape, posture, hands, lighting, and background. Accept the version that removes the distraction with the least unrelated change, then continue with reversible color adjustments if needed.
What to inspect before accepting the edit
Texture should remain believable
Even skin tone does not mean one flat color. Keep natural tonal variation, pores, fine lines, freckles, and directional light unless the person specifically wants a different treatment. Waxy highlights, repeated texture, or blurred facial edges are reasons to reject the result and, if needed, run another attempt.
Identity must be reviewed
The prompt constrains identity changes, but a generative result is not biometric evidence and cannot promise exact biometric likeness. Compare stable anchors such as eye spacing, nose and mouth proportions, jaw outline, hairstyle, visible accessories, expression, and pose against the uploaded source.
Use the lightest acceptable intervention
A restrained repair is often more credible than a polished but unfamiliar face. If only one area is distracting, avoid selecting every treatment. Keep the original available for a side-by-side comparison, and stop when the temporary issue no longer controls attention.
Frequently asked questions
Does a restrained request guarantee unchanged facial structure?
State which traits must remain unchanged, but treat that instruction as a constraint rather than an assurance. An AI result can still introduce unintended differences. Compare the result with the source and reject any version that changes facial structure or recognizable traits.
When should I make another repair attempt?
Make another attempt only when you can identify what failed and how the next correction will address it. Do not assume that a later result is more accurate or that a different application has the same billing and saving behavior. Compare it with the source; if it fails the identity or texture check, reject it before choosing whether another run is worthwhile.
Can I edit the result after choosing it?
Use an editing application that supports the required crop, color, local corrections and export settings. This does not imply that those retired workbench controls are present in the current Imagild marketplace. Keep later edits restrained and use the original comparison frequently so a sequence of small changes does not accumulate into an unfamiliar portrait.
Worked decision record
This hypothetical external Lightroom route starts with cheek-source.jpg, an accepted expression and one temporary mark below the cheekbone. Keep an untouched copy. Create a small Brush mask around the mark, inspect the overlay, and use Subtract to remove any coverage over the mouth corner. Make a restrained local tonal adjustment only if the mark is a tone distraction; a mark requiring texture replacement belongs in an appropriate healing workflow instead.
All observations below are constructed examples, not results from an executed edit.
| Input or stage | Filled observation | Action and reason |
|---|---|---|
| Mask v1 | Overlay reaches the mouth corner | Subtract the corner before changing tone |
| Tonal draft v2 | Mark is quieter but surrounding cheek becomes a flat patch | Reduce or reset the local adjustment; do not increase global smoothing |
| Protected comparison | Mouth corner differs from the accepted source | Discard that candidate and reopen the untouched expression |
The brush controls belong to Lightroom. Match crop and face size for before/after review, then compare the complete portrait. The rollback point is the accepted source, not the latest rejected output. This plan does not report an actual processed image or suggest that a tonal mask reconstructs original skin texture.
Current sources and review boundary
- Adobe Lightroom: Apply Masking for local adjustments — External local masks, person/features and selection refinement; not an Imagild feature claim.
- Runway: Creating with Gen-4 Image References — Reference input and individually reviewed iterations in Runway; its controls and limits are not transferred to Imagild.
Sources retrieved November 13, 2026. This is documentation-based editorial guidance. The filled example is hypothetical; no portrait was generated, retouched, printed or tested for this update. No identity guarantee, production approval, ranking gain or AI-citation measurement is claimed.