Source: https://restorebunny.com/guides/restore-old-photos-without-changing-faces/ # How to Restore Old Photos Without Changing Faces A face-first workflow for repairing damage while checking identity, avoiding waxy skin, and separating recovered detail from AI reconstruction. Restoration techniques9 min readUpdated 11 October 2026 ## Try it on your own photo Repair a copy of your photo, then compare faces and details with the original. [Restore my photo]()1 credit per successful photo · Account required[See credit packs]() Quick answer ## The short version Use the highest-quality scan, make one moderate restoration pass, and compare eyes, mouth, face shape, hairline, and distinctive marks against the untouched scan at the same zoom. Any new detail in a severely blurred or missing area is an inference, so choose the least aggressive result that solves the visible damage. ## Why faces change during AI restoration A restoration model must decide which pixels are damage and which belong to the person. When a face is small, blurred, scratched, or partly missing, the source may not contain enough evidence for an exact answer. The model can produce a plausible eye or mouth that looks convincing without being historically correct. Strong face enhancement can also replace natural film grain, pores, wrinkles, and soft focus with a modern portrait look. The safest goal is not maximum sharpness. It is the smallest change that makes the original easier to see. ## Identity details to protect - Overall face shape, jaw, chin, and distance between features - Eye spacing, eyelids, eyebrows, expression, and direction of gaze - Nose width, mouth shape, teeth, smile lines, and facial asymmetry - Hairline, ears, glasses, jewellery, moles, scars, and other distinctive marks - Age cues such as wrinkles and skin texture rather than a beauty-retouched face ## A face-first restoration workflow 1. **Start from the best source.** Scan the print at 600 DPI in colour when possible. A compressed message attachment gives the model less real information to preserve. 2. **Keep an untouched master.** Duplicate the scan before any crop, contrast change, restoration, or colourization. 3. **Run one moderate repair.** Fix scratches, fading, and noise before adding stronger upscaling. Repeated enhancement compounds invented texture. 4. **Compare at matching zoom.** Place the original and result side by side at the same size. A sharper image can feel more accurate simply because it is easier to see. 5. **Choose the conservative version.** If two outputs both clean the damage, keep the one that changes fewer identity details. ## What natural restoration should look like A faithful result can still contain film grain, soft edges, wrinkles, uneven lighting, and the visual character of the original process. Those are not automatically defects. Remove the scratch across an eye; do not turn the entire face into a studio headshot. RestoreBunny is designed for natural-looking repair, but every generative tool has limits. Use the before-and-after preview, and read our [editorial policy on inferred detail]() before using a restoration as part of a family archive. ## Warning signs that a result went too far - The person looks younger or has a different expression - Eyes, teeth, glasses, or jewellery appear where the source is unreadable - Skin has no texture while clothing and the background remain grainy - Hair, ears, fingers, text, or uniforms have changed shape - The output looks like a modern camera portrait rather than a repaired print ## Example: a repair passes cleanup but fails identity review A scratch disappears from a grandfather’s cheek, yet the output narrows his jaw and gives him a different smile. Record those as two separate findings: scratch improved, identity check failed. Do not accept the portrait simply because it looks sharper. Compare the eye spacing, mouth shape and recognisable marks with the source and, where available, another photograph from the same period. A reference helps you judge a reconstruction; it does not prove every new pixel was recovered. If you cannot isolate the useful repair, keep a lighter edit or seek local manual work. ## FAQs Can AI restore an old photo without changing the person? An output may retain recognisable identity when source detail survives, but no generative restoration guarantees exact facial detail. Compare the source at matching size and reject changes to expression, geometry or distinctive marks. Why do restored faces sometimes look waxy? Strong denoising and face enhancement can remove pores, wrinkles, and film grain. Start with moderate restoration and avoid repeating enhancement on an already enhanced output. Which facial details should I compare? Check face shape, eye spacing, eyelids, mouth, teeth, hairline, ears, moles, glasses, jewellery, and expression at matching zoom levels. ## Sources Preservation and technical guidance reviewed for this article. - arXiv: [Old Photo Restoration via Deep Latent Space Translation]() - U.S. National Archives: [Digitizing Family Papers and Photographs]() ## Related - [Calculate print dimensions after identity review]() - [Review every face in a group photo]() - [Restoration quality checklist]() - [AI Photo Restoration]() - [AI Restoration vs a Professional]() - [How to Scan Old Photos]() - [Restore Blurry Old Photos]() ## Ready to restore your photos? Upload an image, choose your settings, and preview results with a before/after slider. [Restore my photo]()[View Pricing]() RestoreBunny Team Photo restoration editors Guides follow our [editorial policy]() for sourcing, product claims, AI limitations, updates, and corrections.