Use the guide below to plan clearer edits, choose a stronger source image, and understand where generated results need a careful human review. You can return to the AI image to image generator at any time for a blank prompt.
Restore an old photo while protecting its history
Old photo restoration is not the same as making an image look new. A good repair removes damage that blocks the original photograph while preserving the people, place, composition, and era. Scratches, dust, stains, fading, creases, low contrast, and mild blur can often be reduced with an image-to-image prompt. The source remains the visual anchor, so the model has a specific photograph to repair rather than a scene to invent from scratch.
Begin with a conservative instruction. Ask to repair visible damage, recover natural tonal range, and preserve identity, expression, clothing, pose, background, and crop. Add “do not modernize or invent details” when historical accuracy matters. The goal is a believable reading of the original image, not flawless contemporary skin, modern fashion, dramatic lighting, or detail that could not plausibly be present in the photograph.
Scan or photograph the original carefully
The restoration can only work with information visible in the source. A flatbed scan is ideal because it provides even focus and avoids glare, but a phone camera can also work. Place the print on a flat neutral surface, use soft indirect light from both sides, and keep the camera parallel to the photo. Turn off flash, fill the frame, and capture the highest practical resolution without digital zoom.
If the photo is behind glass and cannot be removed safely, change the camera angle or lighting until reflections disappear. Do not risk damaging a fragile original for a better scan. Museums and professional conservators can advise on valuable, stuck, moldy, curled, or chemically unstable prints. AI edits a digital copy; it does not repair the physical object, and it should never replace careful archival handling.
Repair scratches, dust, tears, and creases
Surface damage is a natural first target because it is visually distinct from the underlying scene. Name the problems you see: fine white scratches, dust specks, a vertical fold, water stains, a torn corner, or emulsion loss. Ask for the damage to be removed while preserving nearby facial features, fabric texture, architecture, and film grain. Specific language helps the model distinguish a scratch from a real line in the picture.
Large missing areas require caution. AI may create a plausible replacement, but plausibility is not historical proof. If a corner is gone, the result should be treated as a reconstruction. Keep the untouched scan and label heavily rebuilt versions. For genealogy, archives, or research, consider saving two outputs: a conservative cleanup that removes only obvious surface damage and a separate interpretive reconstruction for display.
Recover faded contrast and period color
Black-and-white photos often lose separation between highlights, midtones, and shadows as prints age. Ask for natural contrast, balanced exposure, and preserved highlight detail instead of “maximum sharpness.” Too much contrast can erase faces and dark clothing; too much smoothing can remove film texture. A convincing restoration usually retains some grain and tonal softness because those characteristics belong to the photographic process.
Color photographs can shift toward yellow, red, or blue as dyes fade. Request believable period color and neutral skin tones without turning the image into a modern high-saturation picture. The model cannot know every original hue with certainty, especially when channels are severely damaged. Use familiar objects, uniforms, signs, or family knowledge as references when available, and treat recovered color as an informed interpretation rather than an exact measurement.
Preserve faces during AI photo restoration
Faces carry the most personal and historical value, so identity preservation should appear directly in the prompt. Ask to keep facial structure, age, expression, gaze, hairstyle, and distinctive features unchanged. Avoid broad requests such as “beautify,” “make younger,” or “perfect the face.” Those instructions can erase the very details that make the person recognizable and can produce a polished portrait that no longer matches the source.
When a face is very small or severely damaged, no model can recover information that is completely absent. It can infer a likely eye, mouth, or skin texture, but the result remains a guess. Compare the output at the same scale as the original, not only when zoomed in. If identity shifts, use a shorter, more conservative prompt and prioritize damage removal over aggressive enhancement or sharpening.
Should you colorize a black-and-white photo?
Colorization can make an old image feel immediate, but it is a creative interpretation separate from restoration. A black-and-white source does not contain definitive color information, so clothing, paint, foliage, and interior tones must be inferred. If authenticity is the priority, first create a clean monochrome restoration. You can then make a second colorized version while keeping the restored black-and-white file as the primary record.
For a colorized copy, include any facts you know: eye color, uniform branch, car paint, house exterior, season, or regional materials. Ask for restrained, historically plausible colors and preserved film texture. Avoid modern cinematic grading unless the image is intended as artwork. Label the output as AI colorized when sharing it in family collections, exhibits, publications, or social posts so viewers understand what changed.
Archive the original and restored files
Never overwrite the master scan. Keep the original capture, a conservative restored version, and any creative variants as separate files. Use descriptive names with dates when known, and record who appears in the photo, where it came from, who performed the scan, and which AI changes were requested. Metadata and written context may be more valuable to future family members than a technically perfect image with no names attached.
Image2Img lets you try restoration without an account and does not add permanent image storage. Files are processed temporarily by the generation provider, and result links may expire. Avoid uploading sensitive archival material when you do not have permission from the owner or depicted people. Download useful results promptly, store them in at least two locations, and consider archival-quality formats for important family or institutional collections.