Restoring the Past (With AI)

Restoring the Past

How can AI Give Old Photographs and Public-Domain Art New Life?

Let me share one of my workflows from the past month with Kre8or.

Old photographs have a particular kind of silence.

They arrive carrying scratches, fading, silvering, torn corners, stains, missing details, and the unmistakable softness of time. Yet behind every imperfection there may still be a face, a landscape, a gesture, a moment of devotion, a family story, or an artist’s vision waiting to be seen more clearly.

Today, artificial intelligence gives us extraordinary tools for restoring such images.

But restoration is not simply a matter of making an old image look new. Done thoughtfully, it becomes something much more interesting: a conversation between past and present, between preservation and interpretation, and between the original maker and the person carefully tending the image centuries (or generations) later.

This raises an important question.

When we use AI to restore an old photograph, print, engraving, painting, postcard, or devotional image, how far should we go?

Restoration Is Not the Same as Reinvention

There is an important distinction between restoring an old image and reimagining it.

Restoration attempts to recover what is already there.
Reinterpretation introduces something new.

The difference may seem subtle at first, but it becomes increasingly important as AI tools grow more powerful.

Removing dust from an old photograph is restoration.
Repairing a torn corner may also be restoration, provided the reconstruction is cautious and based on surrounding visual information.

Changing a person’s clothing, replacing a background, altering facial features, or introducing dramatic lighting moves the work into adaptation or reinterpretation.

Neither approach is inherently wrong.
The important thing is to understand which one you are doing.

Why Old Images Are So Vulnerable

Photographs and works on paper rarely age gracefully without help.

Over decades or centuries, images may suffer from:

• fading
• scratches
• dust
• foxing
• staining
• mould damage
• folded or torn paper
• chemical discolouration
• loss of contrast
• cracks
• missing corners
• ink deterioration
• poor-quality scanning
• compression artifacts from early digital copies

Sometimes the original object remains beautiful despite these imperfections.

At other times, damage begins to obscure the very thing the image was meant to preserve.

A face disappears into shadow.
Fine engraving lines merge into paper grain.
A handwritten inscription becomes barely legible.
A religious print loses the delicacy of its expression.

This is where digital restoration can be remarkably useful.

What AI Has Changed

Traditional photo restoration has existed for many years.

Skilled restorers have long used tools such as Photoshop to clone damaged areas, rebuild missing sections, reduce scratches, restore tonal balance, and improve contrast.

AI changes the process because it can interpret visual patterns rather than merely manipulate pixels.

Modern tools can often identify faces, architectural forms, fabrics, foliage, handwriting, engraving lines, and repeated textures. They can estimate what may have existed in a damaged area and reconstruct it surprisingly convincingly.

They can also upscale small images, reduce noise, sharpen blurred details, colourise monochrome photographs, and rebuild damaged sections.

This power is wonderful.

It is also precisely why restraint matters.
AI does not remember what was actually present.
It predicts what is likely to have been present.

That is an important difference.

The First Principle? Preserve Before You Improve

Before applying any enhancement, preserve the best available copy of the original.
Never make your only file the edited version.

A simple archival workflow can prevent many regrets later.

1. Keep the original scan untouched.
2. Create a working copy.
3. Perform restoration on that duplicate.
4. Save major stages separately.
5. Export a final version for print or web use.

If possible, record where the image came from, when it was scanned, and anything known about its history.
For public-domain artworks, keep the artist, title, date, collection, source page, and rights information with the file.
Digital restoration is much easier when provenance is not separated from the image.

Four Levels of AI Restoration

It can be helpful to think of image restoration as four different levels.
Each level gives the AI slightly more creative freedom.

Level One = Faithful Conservation

This is the most restrained approach.
The intention is not to make the image look modern.
The intention is to make the existing image easier to see.

Typical changes might include:

• removing dust
• reducing scratches
• correcting obvious scanning defects
• repairing tiny tears
• balancing faded contrast
• gently reducing staining
• straightening the image
• improving resolution
• removing distracting border damage

At this level, the composition should remain untouched.
Faces should remain faces as they actually appear.
Clothing should not change.
Background details should not be invented.
The character of the paper, print, photograph, engraving, or painting should remain visible.

For many archival and historical projects, this is the ideal level.

Level Two = Enhanced Restoration

The second level allows slightly more intervention.

The image is still treated as a historical object, but adjustments may be made to improve its usefulness for modern reproduction. This might include richer tonal separation, careful sharpening, subtle reconstruction of damaged textures, improved edge definition, or modest correction of uneven exposure.

A nineteenth-century photograph may become easier to read.
An eighteenth-century engraving may reproduce more beautifully in a book.
A faded postcard may regain enough contrast to reveal architectural detail.

The goal remains fidelity.

You are not creating a new scene.
You are helping the old scene speak more clearly.

Level Three = Respectful Reinterpretation

Here the boundaries begin to shift.

Perhaps you have a beautiful public-domain engraving and wish to introduce restrained colour.
Perhaps you want a collection of old devotional prints from different centuries to share a harmonious palette.
Perhaps an image needs additional background space for a book layout.
Perhaps you want to transform an old monochrome print into something resembling tempera, fresco, illuminated manuscript, or classical painting.

At this point, the result should no longer be presented as an untouched historical original.

It has become an adaptation.

There is nothing wrong with that.
Indeed, some beautiful new work can emerge from precisely this kind of respectful dialogue with historical art.

Transparency simply becomes more important.

Level Four = Creative Reimagining

At the far end of the spectrum, the old image becomes a source of inspiration rather than an object being restored.

The composition may change.
Colours may be completely reconsidered.
The environment may expand.
Figures may be re-rendered.
The visual medium may transform.

A woodcut might become a luminous painting.
A faded photograph might inspire a cinematic portrait.
A Renaissance devotional image might become the basis for a contemporary sacred illustration.

This is no longer conservation.

It is new creative work based upon an older source.

That distinction is worth celebrating rather than hiding.

The Danger of Making Everything Too Perfect

One of the most common mistakes in AI restoration is over-restoration.

The software becomes so enthusiastic that every wrinkle disappears, every surface becomes smooth, every shadow becomes cinematic, and every face begins to resemble a modern digital portrait.

Something important is lost.

Age carries information.
So does photographic grain.
So does the imperfect pressure of an old printing plate.
So do faded pigments and handmade paper.

The goal should not always be perfection.

Sometimes the most beautiful restoration leaves a little history visible.

An old photograph need not look as though it was taken yesterday.
An engraving need not resemble a modern illustration.
A century-old portrait may still deserve the gentle softness of its era.

Faces Require Special Care

Human faces are where AI restoration can be most impressive... and most misleading (Fakes).

When a face is blurry or damaged, an AI system may generate eyes, eyelashes, teeth, skin texture, wrinkles, or hair that appear entirely convincing.

But convincing is not the same as historically accurate.

If the original photograph contains only a few pixels of information, the AI cannot truly know the exact shape of an eye or the precise expression of a mouth.
It creates a plausible solution.

For family history, archives, genealogy, journalism, museums, or documentary work, this distinction matters enormously.

A good rule is simple:

The less information present in the original, the more cautious we should be about claiming that an AI-generated reconstruction represents reality.

Colourising Black-and-White Photographs

AI colourisation can be enchanting.
It can make old streets feel immediate, bring clothing into focus, and help contemporary viewers emotionally connect with historical photographs.

But colourisation is inevitably interpretive unless the original colours are independently known.
The AI may choose a brown coat where the coat was actually blue.

A wall may become cream although it was green.
A military uniform may be rendered incorrectly.
Skin tones may also be inferred rather than known.

Therefore, colourised historical photographs are best described as colourised versions rather than restored colour.
This is particularly important for documentary, academic, and historical uses.

Upscaling? Useful, but Not Magical

AI upscaling is one of the most useful restoration tools available.
A small digital file can often be enlarged enough for printing.

Edges become cleaner.
Noise can be reduced.
Fine patterns may appear more coherent.

But upscaling cannot truly retrieve information that was never captured.

When an AI creates fine eyelashes, embroidery, masonry, hair strands, or text from a tiny source, those details may be generated rather than recovered.

For decorative use, that may be perfectly acceptable.
For scholarly use, it should be treated cautiously.

Restoring Old Prints and Engravings

Historic prints deserve a slightly different approach from photographs.
An engraving, etching, lithograph, or woodcut contains intentional linework.
Those marks are part of the artist’s language.

Aggressive denoising can destroy them.
Excessive sharpening can make them brittle!
AI systems may also mistake cross-hatching for dirt or replace delicate engraved faces with smoother modern-looking features.

A good restoration prompt should therefore explicitly instruct the tool to preserve:

• original linework
• cross-hatching
• engraved textures
• paper character
• facial expressions
• composition
• period style

The machine should understand that these apparent imperfections may actually be art.

Sacred and Devotional Images Require Another Kind of Sensitivity

When working with sacred art, restoration becomes more than a technical matter.
Gesture, expression, symbolism, light, and composition often carry theological meaning.

A hand may be positioned deliberately.
A downward gaze may communicate humility.
The placement of figures may follow centuries of visual tradition.
A halo, garment, object, colour, or architectural element may possess symbolic significance.

An AI system does not necessarily understand any of this.
For that reason, sacred images often benefit from conservative intervention.

Clean the image.
Restore damaged detail.
Improve reproduction quality.

But think carefully before changing expressions, gestures, posture, symbolic elements, or relationships between figures.

In such work, AI is often at its best when it acts less like a new artist and more like a careful illuminator.

Working With Public-Domain Art

Historic artworks whose copyright has expired can often provide a wonderful foundation for restoration projects. Old paintings, engravings, drawings, prints, devotional works, maps, manuscripts, and illustrations exist in enormous public collections.

Many museums and cultural institutions also provide high-resolution images marked as public domain or released under open licences such as CC0. However, it is important to distinguish between the underlying artwork and the particular digital image you have found.

A Renaissance painting may unquestionably be in the public domain.
That does not automatically mean that every photograph, scan, book reproduction, or commercial image of that painting can be used without checking.

Whenever possible, begin with a source that clearly identifies its reuse status.

Museum open-access collections and reputable public-domain repositories are usually far preferable to copying an image from a random website or social-media post.

Keep the Source Information

Whenever you use historic material, maintain a simple source record.

For each image, note:

• artist or photographer, if known
• title or description
• approximate date
• museum, archive, or collection
• source page
• public-domain or licence status
• date you accessed the file
• description of your restoration
• AI tool or editing method used, when appropriate

This small habit can save enormous confusion later. It is particularly valuable if an image eventually appears in a book, exhibition, course, website, documentary, or commercial project.

How to Credit a Restored Image

Credit language should match what you actually did.

If you performed only gentle digital cleaning, you might write:
Digitally restored from the public-domain original.

For a historical artwork:
Albrecht Dürer, Christ Carrying the Cross, 1509. Public-domain original. Digitally restored for reproduction.

If substantial enhancement was involved:
After Albrecht Dürer, Christ Carrying the Cross, 1509. Public-domain original; digitally enhanced.

If you introduced colour or considerable stylistic changes:
After Albrecht Dürer. Digitally reimagined from a public-domain original.

If the historical image merely served as inspiration:
Inspired by the public-domain work of Albrecht Dürer. Contemporary AI-assisted interpretation.

None of these formulations is universal legal language.
They are simply clear ways of telling the reader what they are looking at.

Clarity builds trust.

A Useful AI Prompt for Conservative Restoration

A well-written prompt can make an enormous difference.

For faithful restoration, something like the following works well:

“Faithfully restore this historical image. Remove dust, stains, scratches, fading, creases, and minor physical damage while preserving the original composition, facial expressions, anatomy, clothing, period details, textures, and artistic style. Improve clarity and tonal balance without inventing new elements. Retain the character of the original medium and avoid modernising the image.”

The most important words are often not those describing what you want AI to do.
They are the words describing what it must not change.

A Prompt for Old Engravings and Prints

For engravings, etchings, woodcuts, and lithographs, the instructions should be more specific:

“Restore this historical public-domain print while preserving all original engraved linework, cross-hatching, contours, facial expressions, proportions, and paper texture. Remove stains, tears, dust, fading, and scanning defects conservatively. Improve print clarity and resolution without smoothing the engraving or replacing historic detail with modern digital painting.”

This helps prevent the AI from treating artistic marks as visual noise.

A Prompt for Gentle Colourisation

For colourisation:

“Colourise this historical image with restrained, natural, historically sympathetic tones. Preserve all original linework, composition, expressions, and textures. Keep the palette subtle and believable. Do not modernise clothing, architecture, hairstyles, objects, or facial features. Treat uncertain colours conservatively.”

The phrase “historically sympathetic” is useful because it encourages moderation without pretending that the AI knows the original colours.

A Prompt for Fine-Art Reinterpretation

When the aim is deliberately creative, say so openly:

“Using this public-domain historical artwork as the primary reference, create a respectful contemporary reinterpretation. Preserve the central composition, gesture, emotional meaning, and visual dignity of the original while introducing a harmonious new palette, refined lighting, and a painterly finish. The result should feel inspired by the historical work rather than presented as an untouched original.”

This gives the machine permission to create while maintaining respect for the source.

What AI Should Not Be Asked to Decide for You

AI can make images beautiful.

It cannot decide whether an alteration is historically justifiedIt cannot determine what matters most in a family photograph.
It cannot know whether a religious symbol should be changed.
It cannot tell you whether a generated facial expression remains faithful to the person who once stood before the camera.

Those decisions belong to the human curator.

The best restoration projects therefore combine artificial intelligence with human judgment.

Technology handles repetition, reconstruction, noise reduction, and visual refinement.
The human decides what deserves to remain.

A Thoughtful Restoration Workflow

A careful restoration process need not be complicated.

1. Find the highest-quality original available.
2. Confirm the source and reuse status where relevant.
3. Save an untouched master file.
4. Correct cropping, rotation, and exposure first.
5. Remove obvious dust and physical damage.
6. Use AI cautiously on difficult damaged areas.
7. Compare frequently with the original.
8. Preserve historically important texture.
9. Upscale only after the image is clean.
10. Save both the restored image and the original.
11. Record what was changed.
12. Label creative adaptations appropriately.

The most important step is number seven.

Keep looking back at the original.
Without that reference, it is remarkably easy to drift away from restoration and into invention.

Before and After Images Can Be Powerful

When sharing restoration work publicly, showing the original beside the restored version can be fascinating.

It allows the audience to appreciate both the history of the object and the care involved in bringing it forward.
It also encourages transparency.

A restored photograph becomes more meaningful when viewers can see the damage that once obscured it.
A cleaned engraving becomes more impressive when its original paper stains and fading remain visible in the comparison.

Restoration then becomes part of the story rather than an invisible trick.

When Imperfection Should Remain

There are moments when the right choice is to leave something alone.

A handwritten crease may have been present when a photograph was carried in someone’s wallet for thirty years.
A worn edge may tell us that an image was repeatedly handled.
A faded inscription may be part of the object’s history
A stain on a family photograph might even be connected to the circumstances in which it was preserved.

Not every mark is damage.
Some marks are biography.

The thoughtful restorer learns to distinguish between the two.
AI as a Conservator, Not a Conqueror

It is tempting to think that the newest technology should produce the most dramatic transformation
Yet the most sophisticated restoration may be the one nobody immediately notices.

The viewer simply sees the face more clearly
The print breathes again.
The details return.
The image feels present without losing the soft gravity of its age.

That is a beautiful role for artificial intelligence.

Not to erase the past.
Not to improve upon the people who came before us.
But to clear away some of the dust between their world and ours.

A New Kind of Creative Stewardship

We live at an unusual moment.

Never before have ordinary people possessed such powerful tools for repairing family archives, preserving fragile photographs, studying historic art, revitalising public-domain illustrations, and preparing forgotten images for new audiences.

With that opportunity comes responsibility.

We should know when we are restoring
We should know when we are interpreting.
And we should be willing to tell others which is which.

The most meaningful use of AI in historical imagery may therefore have less to do with technological perfection and more to do with stewardship.

We inherit an image.
We care for it.
We preserve what remains.
We make thoughtful choices about what has been lost.

And then, perhaps, we pass it onward a little clearer than we found it.

Conclusion

AI can give damaged photographs, forgotten prints, public-domain artworks, and fragile visual memories a remarkable second life.

The best results come from restraint, transparency, careful sourcing, and respect for the original. Restore when restoration is enough, reinterpret when creativity calls for it, and never confuse the two.

If you have an old photograph, engraving, devotional print, postcard, or public-domain artwork waiting in a drawer or digital archive, begin with one image. Preserve the original, make a working copy, and see what a gentle restoration can reveal.

Sometimes the past does not need to be reinvented.
It only needs a little light.

✨ Fleeky

Thank you for reading, likes, shares and comments.

Where does my featured image fit?
PS
Yes, much longer as usual.