AI Photo Restoration Identity Preservation: Use Three Constraint Lists

Upload a damaged family photograph to create a cleaned, softly colorized archival portrait while keeping the people, objects, and original framing recognizable. A three-list prompt gives the restoration a visible priority order: what must stay, what may be repaired, and what the edit must avoid.

What the restoration preserves

The workflow suits a portrait with identifiable faces, clothing, props, and architecture. In the test scene, a seated mother remains on the left and her daughter remains on the right, holding a cloth doll on a wooden porch. The oval brooch, chair, railing, clapboard wall, and screen door remain part of the composition.

The output uses restrained period color with an archival photographic grade. Skin texture, fabric, wood, and the photograph's subtle grain stay readable, preserving the feeling of an archival print.

The three-list prompt structure

Start with the restoration target: remove scratches, fold creases, stains, dust, fading, noise, and mild blur, then apply historically plausible color.

Next, write a Must Preserve list. Name identity, apparent age, hairstyle, expression, pose, hand placement, object positions, crop, and camera viewpoint. Concrete nouns give the edit a checklist it can revisit.

Then add Only Allowed Repairs. Limit reconstruction to small damaged areas supported by adjacent visible evidence, and specify restrained color, natural skin texture, and subtle grain.

Finish with Absolutely Forbidden. Lock the frame and forbid face redesign, beautification, modern lighting, background replacement, added objects, removed objects, HDR, plastic skin, and painterly reinterpretation.

When to use it

Family archivists can use the structure when a scan contains scratches and water marks but still has legible faces and heirloom objects. A local historical group can apply the same wording to a porch, storefront, or studio portrait while keeping the period setting visible. A creator making a before-and-after restoration post can show the damaged scan beside the cleaned result and explain each constraint in plain language.

A practical three-step flow

1. Choose a well-framed scan where the important faces and objects are visible.
2. Describe the repair target, then separate preservation, permitted repair, and forbidden changes into three lists.
3. Inspect the output against the original crop, faces, hands, props, and architecture before sharing it.

Frequently asked questions

What input works best?

A scan with visible subjects and enough adjacent detail for small repairs gives the prompt useful evidence to follow.

What should I name in the preservation list?

Name identity cues, pose, hands, clothing, distinctive props, architecture, crop, and viewpoint.