How to Restore Old Baby Photos — First Year Photography and Early Childhood Portraits

Baby photographs are among the most treasured family images in existence, and also among the most challenging to restore well. The subjects are photographed during the period when faces change most rapidly — a month in infant development makes more visible difference than a decade in adult life — which means the AI models handling facial restoration need to work with facial geometries that look dramatically different from adult faces.

At the same time, baby photographs are often taken in informal conditions with consumer equipment by people whose priority was capturing the moment rather than photographic technique. The results are frequently overexposed by flash, slightly blurry from the rapid movements of the subject, small in format, and stored without archival care because no one expected these casual snapshots to become treasures.

This guide covers what you need to know to restore baby photographs effectively, from hospital portraits to first birthday images to informal home snapshots of the first year.

What Makes Baby Photographs Challenging for AI Facial Restoration?

AI face restoration models like GFPGAN and CodeFormer were trained primarily on adult face datasets, because adult faces are far more numerously represented in available training data. When these models encounter infant faces — which have fundamentally different geometric proportions than adult faces — their performance changes in ways that matter for restoration quality.

Infant faces have proportions that are characteristic of early development: eyes that are relatively much larger compared to facial width than adult eyes, flatter nasal bridges, softer jaw angles, more uniform and smoother skin texture, and a much shorter face height relative to facial width. These proportions gradually shift toward adult ratios through childhood and adolescence.

The practical implication is that AI models may sometimes apply adult facial geometry priors to infant faces if the face is very small in the frame or if the image quality is very low. CodeFormer, with its fidelity parameter, gives you some control over this: higher fidelity settings preserve the original facial structure (including infant proportions) more faithfully, while lower fidelity settings allow the model to apply stronger geometric priors (which lean toward adult proportions). For infant facial restoration, start with high fidelity settings and reduce only if the facial reconstruction is clearly wrong.

NAFNet's noise reduction is particularly valuable for baby photographs because infant skin texture is smooth and fine, and noise in this area can be mistaken by upscaling models for genuine skin texture. Applying NAFNet before Real-ESRGAN reduces this confusion.

How Do Hospital Baby Photos Typically Degrade?

Hospital baby photographs have a distinctive provenance that affects their condition. From the 1950s through the 1980s, hospitals typically contracted with commercial portrait photographers who photographed newborns in the nursery within the first day or two of life. These photographs were produced and delivered in standardized formats — often a cardboard folder containing a 5x7 print, a strip of wallet photos, and sometimes a small ornamental format like an oval.

The production quality of hospital baby photographs was variable. Commercial photographers working on hospital contracts were producing high volumes of work quickly, and the results ranged from beautifully lit and composed images to technically problematic photographs with harsh lighting, closed or partially closed eyes, and unflattering positions.

The paper and chemical quality of these photographs also varied. Some hospital portrait packages used high-quality materials that have survived five decades in excellent condition; others used cheaper materials that yellowed and faded within a decade. The distinctive warm yellow-orange cast of a severely faded chromogenic hospital baby print is one of the most common starting points for family baby photo restoration.

DDColor addresses this color shift effectively because the warm fading of these prints follows a predictable pattern. Real-ESRGAN handles resolution enhancement for the 5x7 print size, and CodeFormer addresses the specific challenge of restoring a face that is both very small in the frame and has infant proportions.

What Are the Most Valuable Baby Photographs to Prioritize for Restoration?

Not all baby photographs are equally irreplaceable, and if you are working with a large first-year archive, prioritizing the most significant subjects makes sense.

First day of life photographs — particularly hospital nursery portraits and photographs taken immediately after birth — document a moment that is genuinely unique in a human life and will never be replicated. These deserve the most careful scanning and the most conservative restoration approach.

First milestone photographs: first smile captured on camera, first time sitting unsupported, first birthday, first steps. These document developmental moments that parents remember but that photographs make specific.

Photographs showing the baby with relatives who have since passed — grandparents, great-grandparents, older siblings who died young. These photographs serve double duty as documentation of both the baby's early life and the relative's presence in it.

Photographs that are the only visual record of specific contexts: the home where the baby was raised, the siblings' reactions, the first Christmas or holiday celebration. Context documentation has value that accumulates over time.

How Do You Scan Baby Photos That Have Been Mounted in Old Baby Albums?

Baby albums from the 1950s through the 1980s typically mounted photographs using adhesive corners or acidic photo mount adhesive. Many baby photographs have been in these albums for decades and removing them risks physical damage to both the photograph and any handwritten notes or decorations on the album pages.

For photographs that can be safely removed (those in corner mounts or that are clearly loose), remove them carefully, scan them individually at the appropriate resolution, and replace them after scanning.

For photographs that are adhered to album pages and cannot be removed without risk, scan the entire album page as a single scan and crop individual photographs from the full-page scan afterward. Use your scanning software's maximum available resolution for album page scans, and ensure the scanner bed is clean and the page is pressed flat against the glass without excessive pressure.

ArtImageHub ($4.99 one-time lifetime access) handles the cropped individual photograph scans effectively. The AI models — Real-ESRGAN, GFPGAN, NAFNet — work on the image data regardless of whether the photograph was scanned individually or cropped from a larger album page scan.

Why Do Parents Often Wait Too Long to Restore Baby Photos?

There is a paradox in baby photograph preservation: the photographs that document the moments parents were most emotionally present for are often the least carefully preserved afterward. The intensity of new parenthood leaves little bandwidth for archival thinking, and the small, casual snapshots from this period are more likely to end up in a box than in an archival album.

The result is that baby photographs, which should be among the most carefully preserved family documents, frequently arrive at the restoration stage in conditions that reflect decades of casual storage rather than intentional care. This is not negligence — it is the natural result of life's priorities.

The right time to restore baby photographs is whenever you have them and the motivation to do it. The photographs that have survived for thirty or forty years in a shoebox will continue to survive through the process of careful scanning and AI restoration. What you gain is a digital record that will not continue to degrade and can be shared with the now-grown child whose early life it documents.

Frequently Asked Questions

How does GFPGAN handle the small, distinctive faces of infants in photographs?

GFPGAN approaches infant faces differently from adult faces, though the distinction is sometimes imperfect. The model uses a generative face prior — a learned statistical model of face structure — to guide reconstruction of damaged facial areas. For faces that clearly have infant proportions, the model generally recognizes and respects the distinctive characteristics: the larger eye-to-face ratio, the smoother skin texture, the softer facial contours. However, when the infant face is very small in the frame — occupying fewer than 100 pixels in any dimension — the model has insufficient data to confidently identify the face as infant rather than adult, and its reconstructions may lean toward adult proportions. For very small infant faces, use CodeFormer with a high fidelity setting (0.7 or above if the tool exposes this parameter) rather than letting the model apply its strongest face prior. The result will be less aggressively enhanced but more faithful to the original infant face proportions.

Can I restore a baby photo that is very blurry because the baby was moving?

Motion blur is common in baby photography because infants and young children move constantly and unpredictably, and mid-century photographers without fast shutter speeds could not always freeze this movement. The degree of blur matters for what restoration can achieve. Slight blur — a few pixels of directional smearing — responds well to Real-ESRGAN's sharpening component, which identifies directional blur as a degradation type distinct from noise and recovers much of the underlying detail. Moderate blur — the kind where features are clearly smeared over a visible distance — is more challenging; restoration produces a softer result than a well-focused original, but still significantly better than the blurry original. Severe blur — where the subject is rendered as a completely indistinct shape — cannot be meaningfully reversed by AI restoration, because the original information about feature positions was genuinely not captured by the camera. In this last case, restoration can improve tone and color but cannot restore sharpness that was never present.

Should I restore both the black-and-white and color versions when both exist for hospital baby photos?

Yes, restore both if you have both. Hospital portrait packages from the 1960s through the 1980s sometimes included both black-and-white and color versions of the same image, taken at the same session or at closely adjacent times. Each version contains different information and has different archival value. The black-and-white version, if printed on fiber-based paper, may have better long-term stability and more faithful tonal reproduction of the lighting. The color version documents the actual colors of clothing, blankets, and skin tone that the black-and-white version cannot show. Restoring both preserves both types of information. For restoration purposes, the black-and-white version benefits from NAFNet and Real-ESRGAN applied to the grayscale tonal values, while the color version additionally benefits from DDColor addressing any color shift that has occurred over decades. Keep both restored versions in your archive labeled clearly to avoid confusion.

How do I recover a first-birthday photograph that was taken with a flash-heavy Instamatic camera?

Instamatic flash photographs from the 1960s and 1970s suffer from characteristic problems: harsh frontal illumination that eliminates shadow modeling, frequently red-eyed subjects, overexposed highlights in foreground areas, and severe underexposure in the room behind the foreground flash zone. These problems compound the typical age-related degradation of chromogenic prints from this era. Real-ESRGAN handles the resolution enhancement and general tonal recovery. Red-eye removal, while not a primary feature of restoration models, is addressed as a color artifact by DDColor and by CodeFormer's face restoration, which identifies the artifactual coloration of flash-induced red-eye and corrects it. The harsh flash lighting that creates a bright foreground and dark background is the most difficult characteristic to address, because the dark background areas simply did not receive enough exposure to contain recoverable detail. Restoration can improve tonal consistency and reduce the jarring quality of this lighting pattern but cannot add detail that was not captured.

What is the best way to share restored baby photos with the now-adult child depicted?

Adults receiving restored photographs of themselves as babies often report that the experience is surprisingly emotional, because the photographs document a period of their own life that they cannot remember but that shaped who they became. Presentation matters. Rather than simply sending a folder of restored image files, consider a presentation that provides context: a simple document or caption file that identifies when each photograph was taken, where, and who else is present. A printed album or photo book provides a physical artifact that has different meaning than a digital file. A shared cloud folder or family photo sharing service allows extended family — grandparents, aunts and uncles, cousins — to access the collection and add their own annotations and memories. The restored photographs are most valuable when they become social objects that generate conversation and shared memory, not when they sit archived on a hard drive. The restoration is the beginning of a conversation, not the end of a project.