Verified against Nano Banana / Gemini 3.1 Flash Image · 2026-07-24
Restore and colorize a damaged old photograph without inventing detail
A restoration edit that repairs physical damage and optionally colorizes an old photo, with an explicit rule against fabricating facial or clothing detail in areas too degraded to actually reconstruct — the honesty check most restoration prompts skip.
The prompt
Ready to copy — highlighted parts are example details you can swap.
You are restoring a damaged historical photograph. The goal is repair and, if requested, colorization — not creative reinterpretation. Where the original damage genuinely obscures a detail beyond confident reconstruction, say so rather than confidently inventing it. PHOTO CONDITION a 1940s family portrait, black and white, with a large diagonal crease across the lower half and heavy foxing spots in both corners. KNOWN FACTS the woman on the left wore a navy-blue wool coat, per a family letter describing this exact photo session. Use these facts to guide plausible reconstruction and colorization choices — they are the closest thing to ground truth available here, and they should override a generic guess whenever the two would conflict. DAMAGE TO REPAIR the crease cuts directly across one subject's torso, and foxing has partially obscured the face of the child on the right. Repair these specifically: remove creases, tears, and spotting; reconstruct genuinely missing physical areas of the photograph using the surrounding context and the known facts above; do not use the repair pass as an excuse to also sharpen, denoise, or stylize areas of the photo that were never actually damaged. COLORIZATION yes, colorize using period-appropriate 1940s tones — nothing that looks like a modern color grade. If colorization is requested, base skin tones, clothing colors, and background elements on the known facts where available, and on well-documented, era-appropriate defaults where they aren't — for instance, common fabric dyes and photographic conventions of the stated time period — rather than modern color trends that wouldn't have existed when the photo was taken. HONESTY ABOUT RECONSTRUCTION This is the most important rule in this brief: if a face, a piece of clothing, or any detail is damaged badly enough that no confident reconstruction is possible from what remains, do not invent a plausible-looking but fabricated detail and present it as restored fact. Instead, reconstruct it as neutrally and minimally as the surrounding context actually supports, and separately list every area where you made a genuinely uncertain call, described specifically enough that the person requesting this restoration can decide whether that guess is acceptable to them. WHAT MUST NOT CHANGE The composition, the number and identity of people or objects present, and anything about the photo that is clearly and confidently visible in the original must be preserved exactly — restoration means repairing damage to what's there, not composing a new, cleaner-looking version of the scene. OUTPUT USE printing at roughly 8x10 inches for a family reunion display board — keep this in mind for how much visible restoration "grain" or texture should remain versus how smooth the result should look. OUTPUT One restored image, plus a separate, clearly labeled list of every area where a genuinely uncertain reconstruction call was made.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Photo restoration is one of the few image-editing tasks where a fluent, confident-looking result can be actively worse than an honest, partial one, because the entire value of a restoration to a family or archive is its claim to represent what was actually there — a model that fills a badly foxed face with a plausible but fabricated expression isn't restoring the photo, it's replacing a piece of family history with a guess dressed up as fact, and nothing about how convincing the output looks tells the requester which parts of it are real. This prompt's honesty rule targets that directly by asking for a separate, explicit list of uncertain reconstruction calls rather than trusting the model to volunteer that caveat unprompted, because an image model's default behavior under a restoration request is to produce the most visually complete and confident result it can, not to flag its own uncertainty unless it's specifically asked to track and report it. Second, weighting known facts above generic guesses when the two would conflict matters because a model with no grounding information will default to the statistically most common visual pattern for a given era or context — the most common 1940s coat color, the most typical period hairstyle — which is a reasonable fallback only in the absence of anything better, and stating that real, specific facts about these actual subjects should override that generic default is what keeps the restoration anchored to this particular photo instead of drifting toward a generic period-photo template. Third, explicitly separating damage repair from unrelated enhancement — no sharpening or stylizing of areas that were never actually damaged — closes a common overreach in AI restoration tools, where a repair pass quietly smooths skin texture, boosts contrast, or otherwise modernizes the whole image's look well beyond what damage repair required, which can make an otherwise successful restoration look subtly anachronistic or lose the authentic photographic grain that gave the original image its character in the first place.
Verified against
Nano Banana / Gemini 3.1 Flash Image Gemini 3.1 Flash Image · 2026-07-24
Changelog
- 2026-07-24 — Initial publish, verified against Nano Banana (Gemini 3.1 Flash Image) on a creased and foxed 1940s family portrait.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
