Old Photo Restoration AI Prompts for Gemini, ChatGPT and Smart Editor

Each old photo restoration AI prompt below fixes one kind of damage on a photo you upload: a crease, a tear, a water stain, faded color, a soft focus, or a black and white print you want in color. They're written for Smart Editor in Toybox AI, which edits one uploaded photo from a plain instruction, and every one of them says what must stay the same. They also work as starting points in ChatGPT and Gemini, where results vary by model. Scan or photograph the print before you start, as a JPG or PNG.

How a good old photo restoration AI prompt is built

A restoration prompt has five parts, and the order matters less than having all of them.

1. What the photo is. The era and type, such as "a black and white graduation portrait from the 1960s". Google's developer guide for Gemini image generation recommends giving context and intent, and the era tells the AI what the finished print should look like.
2. The damage and where it is. "A dark scratch across the bride's veil, from her left ear to the bouquet" leaves no doubt about which mark you mean.
3. The fix. Say what should replace the damage, such as matching fabric, the same gray wall, or clear sky.
4. What must not change. Faces, expressions, pose, clothing, background and framing. Google's developer guide for Gemini image editing suggests "change only" a named element and keeping everything else exactly the same, and OpenAI's guide for its image models gives the same advice.
5. The finish. Keep the grain, keep it black and white, or name the colors.

Fill in the brackets:

This is a [era and type of photo]. [Describe the damage and where it is]. Repair it by [what should replace the damage]. Change only the damaged area. Keep every face, expression, pose, the clothing and the background exactly the same, and keep the film grain.

Damage repair prompts

Crease or fold line

For a print that was folded in half or bent in a wallet, leaving a white line.

This is a black and white family portrait from the 1950s. A white fold line runs straight down the middle, from the top edge through the father's tie to the bottom edge. Remove the fold line and fill it with the tones on either side of it. Keep all four faces, their expressions and their clothing exactly the same, and keep the film grain.

Change: the direction of the line and what it crosses.

Torn corner

For a corner that's ripped or missing, where the background needs rebuilding.

The top right corner of this old photo is torn away, leaving a white triangle. Rebuild that corner as a continuation of the wooden fence and the leafy tree behind it. Change only the missing corner. Keep the two people, their faces and everything else in the photo as it is.

Change: what's behind the missing corner, such as a curtain, sky or a brick wall.

Tear across the photo

For a print torn in two and scanned with the pieces pushed together.

This photo was torn into two pieces and scanned with the pieces side by side. Join them into one picture along the tear that runs from the left edge through the girl's shoulder to the bottom right. Line up the edges of her dress and the porch railing, and fill the gap with matching detail. Keep her face and the woman's face exactly the same.

Change: the path of the tear and the lines that need to meet across it.

Scratches and dust specks

For prints with fine white scratches and specks from dust or handling.

Remove the thin white scratches and the small white specks across this black and white photo, especially the ones in the sky and on the man's dark jacket. Fill each one with the tone around it. Do not smooth the photo or remove its grain, and keep his face, hands and hat exactly as they are.

Change: where the scratches are worst.

Water stain or tide mark

For the brown or gray rings water leaves behind on a print.

A brown water stain with a dark ring covers the lower left quarter of this photo, over the grass and the edge of the picnic blanket. Remove the stain and rebuild the grass and the checked blanket underneath with the same light and texture as the rest of the photo. Change only the stained area. Keep the three children and their faces exactly the same.

Change: the color of the stain and what it covers.

Mold spots

For small gray or dark fuzzy spots scattered across a print. Clean the print itself only with a conservator's advice.

Small gray spots of mold are scattered over the woman's white blouse and the wall behind her. Remove every spot and restore the plain white fabric and the pale wall. Keep her face, hair, necklace and pose exactly the same, and keep the soft focus of the original.

Change: the color of the spots and the surfaces they sit on.

Faded color and tone prompts

Faded color print with an orange cast

For snapshots from the 1970s and 1980s that have turned orange, red or pale.

This color snapshot from the late 1970s has faded to an orange cast. Restore natural colors: a blue sky, green lawn, white house trim and natural skin tones. Keep the soft look of a film print. Do not change the people, their faces, their clothing or the framing.

Change: the colors of the main things in your photo.

Washed-out black and white

For a black and white print that has faded to light gray with no real blacks.

This black and white photo has faded to low contrast, with gray shadows and gray highlights. Restore deep blacks in the shadows and clean whites in the shirts and the sky, with a full range of grays between. Keep it black and white, keep the grain, and keep every face exactly the same.

Change: which areas should be darkest and lightest.

Yellowed black and white print

For a print that has turned yellow or cream but should be neutral.

This black and white print has yellowed with age. Remove the yellow tint so the whites are white and the grays are neutral, and keep it a black and white photo. Do not add color. Keep the faces, the clothing, the grain and the framing exactly as they are.

Change: say "keep a light warm sepia tone" if you like the aged look.

Colorize black and white photo prompts

Portrait with known colors

For a portrait where the family remembers some of the real colors.

Colorize this black and white portrait from the 1940s. He had sandy blond hair and gray-blue eyes, and he wore a brown tweed jacket with a dark green tie. Use natural, slightly muted skin tones. Keep his face, expression, hairstyle and the studio lighting exactly the same, and keep the grain.

Change: the colors you know, and the decade.

Outdoor family scene

For a group photo outdoors where nobody remembers the colors.

Colorize this black and white photo of a family standing in front of a farmhouse in the 1950s. Use believable, muted period colors: a white wooden house, green grass, a pale blue sky, and plain cotton clothing in soft blues, browns and greens. Keep every face, pose and the composition exactly the same.

Change: the setting and the season.

Sharpness and face prompts

Soft focus portrait

For a portrait where the camera missed focus by a little.

This portrait is slightly out of focus. Make it sharper so the eyes, eyelashes and hair are clear, as if the camera had focused correctly. Keep her exact face shape, eye shape and expression. Keep the fine lines around her eyes, and do not airbrush her or take years off her age. Keep the grain.

Change: which details matter most, such as a uniform's buttons or a necklace.

Small faces in a group photo

For a wedding party or class photo where the faces are tiny and blurry.

This is a group photo of about twenty people at a wedding in the 1930s, and each face is tiny and soft. Make each face clearer while keeping each person's own features, expression and age. Don't make the faces look alike, and do not add or remove anyone. Keep the clothing, the flowers and the building behind them the same.

Change: the event, the number of people and the decade.

Light damage in one pass

For a print with only small problems, where one edit can handle all of them.

Restore this 1960s black and white photo: remove the dust specks and the small scratch on the left, bring back contrast in the shadows, and make the faces a little sharper. Change nothing else. Keep every face, expression, pose and the background exactly the same, and keep the film grain.

Change: the list of small fixes, and keep it to three or fewer.

Tips for better old photo restoration results

• One fix per edit for bigger damage. OpenAI's prompting guide suggests beginning with a simple prompt and then making one small change per follow-up. Repair the tear first, check it, then fix the fading in a second edit.
• Repeat the keep list every time. The same OpenAI guide says to restate what must stay unchanged on each round to reduce drift. Paste "keep every face, expression and pose exactly the same" into every follow-up.
• Be exact about location and material. Google's prompting tips for the Gemini app ask for direct, specific wording when you change an existing image. "The plaid skirt below her left hand" gives the AI a place and a pattern to match.
• Describe the clean result. Google's developer guide suggests describing what you want positively instead of listing what you don't. Ask for "a clean, even gray sky" instead of listing the marks to remove.
• Say what the photo is for. "A family portrait for a memorial slideshow" gives the AI the context and intent that Google's developer guide asks for.
• Compare with the original. OpenAI's image generation guide lists keeping recurring characters consistent across generations as a known limitation. Check eyes, teeth, ears and hairlines against the scan after every edit.

Fixes for common problems

Run these prompts in Smart Editor

Smart Editor runs on OpenAI's GPT Image 2.5 and edits one photo at a time: add the scan under "Your image", type or paste one prompt into "What should change?", then tap "Apply edit". For the next fix, keep "Latest result" selected, or switch to "Original photo" to start again. It needs a Pro plan and currently costs 85 credits per HD edit (110 for 4K), and every edit is a fresh redraw of the whole photo, so check the faces after each one. To colorize without changing anything else, Image Colorizer takes a short note about colors and currently costs 75 credits. For the full order of steps, read how to restore old photos with AI, the guide to faded and torn prints, and the colorizing how-to. Toybox's privacy policy says your uploads aren't used to train AI models.