Local task records

Source copies, results, diagnosis, and QA evidence persist on the computer running the local service.

External AI processing

Before generation, the product discloses that images may be sent to the configured external service and waits for confirmation.

Review before delivery

Comparison tasks require a preferred result while QA risk and output evidence stay visible.

Source diagnosis

Map each mark before deciding what to remove

Direction and continuity

Physical scratches often cross unrelated subjects, while real edges usually follow faces, clothing, architecture, or the print border.

Check whether texture and tone continue on both sides of the mark; wide cracks may cover information that cannot be reconstructed exactly.

Marks crossing eyes, mouths, or facial outlines carry higher invention risk than scratches in low-detail backgrounds.

Restoration evidence

Key principles

Low-risk areas establish whether the cleanup preserves natural grain before any facial repair is trusted.

When a scratch crosses a face, preserve symmetry, age, expression, and surrounding evidence instead of designing a perfect feature.

Original print edges, captions, stamps, and handwritten notes should not disappear as accidental cleanup.

FAQ

Can deep cracks be repaired accurately?

Small interruptions can often be inferred from both sides, but wide missing areas require uncertainty and manual review.

Why keep some surface texture?

Natural grain and paper texture help the result remain photographic and prevent over-smoothed, synthetic surfaces.

Should every white line be removed?

No. Some lines belong to clothing, hair, borders, writing, or the photographed scene rather than physical damage.

Ready to begin

Upload the source, choose a restoration route, and inspect likeness, composition, and quality evidence before delivery.