How to Restore Photos with Torn Edges, Missing Pieces, or Physical Damage

Start here: ArtImageHub's Old Photo Restoration handles fading, yellowing, fold marks, and surface scratches automatically — $4.99 one-time. For the honest guide on what it can and cannot do with physical damage, read on. Related tools: Photo Denoiser, Photo Deblurrer, Photo Enhancer.

Every family has at least one: the photo that was torn in frustration, worn down at the edges, or punched through for filing and never quite recovered. Physical damage to photographs is one of the most emotionally charged problems in home archiving — and one of the most misunderstood in terms of what modern AI can actually do.

This is an honest guide. It explains the different types of physical damage, what AI restoration genuinely handles well, where it stops, and what you need to do when AI is not enough.

What Types of Physical Damage Do Photos Actually Suffer?

Physical damage to photographs falls into two fundamentally different categories, and this distinction drives everything that follows.

Damage that sits on top of the image:

• Fold lines and crease marks from folding or rolling the photo
• Yellowing and overall color shift from chemical aging
• Foxing spots (brown oxidation dots) and water stain rings
• Surface scratches from abrasion
• Dust and debris embedded in the surface
• Silver mirroring on black-and-white prints (a shiny metallic sheen over dark areas)
• Fading and contrast loss from UV exposure or poor storage

In all these cases, the underlying image is still there. The damage is a layer sitting on top of, or degrading, the image content — but the content itself is intact.

Damage that removes image content:

• Torn corners or sections where material is physically gone
• Punch holes through the photo
• Sections burned, water-soaked beyond legibility, or peeled away
• Writing or stamps on the front that cover part of the image

In these cases, the image content itself is absent. There are no pixels to clean — only a gap where pixels used to be.

This distinction is the single most important concept in photo restoration. Everything else follows from it.

What Does AI Restoration Genuinely Fix?

ArtImageHub's Old Photo Restoration is trained to recognize and repair the first category: damage that sits on top of intact image content.

Fold lines and crease marks appear in scans as linear artifacts — sharp brightness discontinuities, color shifts along the fold, and sometimes physical texture that casts a subtle shadow. The AI treats these as image noise and smooths them out. For a photo folded once in half, the crease line typically disappears almost entirely after processing.

Yellowing and fading are addressed by the AI's color correction layer, which learns what white balance and tonal range the photo likely had when fresh. It reverses the amber shift of aging paper and restores contrast lost to oxidation. The effect on heavily yellowed photos can be striking.

Surface scratches and foxing are treated as localized artifacts — the model identifies regions that are inconsistent with the surrounding image and reconstructs them from context. Thin surface scratches across plain backgrounds (sky, wall, a dress) are very cleanly removed. Scratches that cross fine detail areas (a face, a hand, text) recover well but not always perfectly.

Facial enhancement is handled by a GFPGAN-derived component that specifically sharpens and reconstructs facial features in portraits. This is where AI restoration tends to produce the most visually striking improvements — even significantly faded portraits often recover clear, sharp facial detail.

Where Does AI Restoration Stop?

The boundary is clear: AI restoration does not generate image content that was never there.

When a corner is torn off and the piece is gone, the AI has no information about what was in that corner — no pixels to clean, no latent signal to amplify. What it sees is absence, typically represented as a white or black void in the scan. Standard restoration models are not trained to invent plausible content for that void. They will clean up the edge of the tear and improve everything that survived, but they will not fill the missing corner.

This is not a limitation of ArtImageHub specifically — it is the boundary between two different AI tasks:

• Photo restoration: cleaning, denoising, and sharpening image content that exists but is degraded
• Inpainting: generating new image content to fill regions where nothing exists

ArtImageHub's Old Photo Restoration is a restoration tool. Inpainting is a separate capability offered by tools like Adobe Firefly, Stable Diffusion with inpainting models, or professional retouchers who paint missing content manually.

Writing on the front of the photo (handwritten captions, stamps, pen marks) is a partial case. If the writing is light and in a plain area, AI restoration sometimes smooths it away. If it is dark ink over complex image content — faces, clothing — it reads as part of the image and is not removed.

Punch holes are another case where AI fills the immediate region around the hole with some success, but a hole through a face or significant detail requires manual retouching.

What Is the Realistic Workflow for a Physically Damaged Photo?

This is the practical workflow used by restoration technicians who integrate AI tools:

Step 1: Scan properly. Scan at 600 DPI minimum (1200 DPI for small prints). Use a flatbed scanner, not a phone camera. If the photo is in pieces, scan each piece separately. If there are loose edges or curling, place a clean sheet of glass over the photo to flatten it without tape or contact adhesive.

Step 2: Align pieces digitally. If the photo was in multiple pieces, align them in any image editor (even the free GIMP) before restoration. The AI performs better on a unified, aligned image than on separate fragments.

Step 3: Run Old Photo Restoration. Upload to Old Photo Restoration. This handles all the surface damage: yellowing, fading, fold marks, scratches, foxing, facial sharpening. Download the result. Assess what remains.

Step 4: Evaluate missing content separately. Look at the restored image. The surviving portions should look significantly improved. Now assess the missing regions — the torn-off corners, the punch holes, the burned sections. These are a separate project.

Step 5: Decide on inpainting or retouching. For small missing sections in plain background areas, inpainting tools (Adobe Firefly's Generative Fill, Stable Diffusion inpaint) can produce plausible results. For missing sections that contain faces or significant detail, a professional retoucher who can manually paint the region is the most reliable path. Fiverr and specialized photo restoration services offer this starting around $30–60 per photo.

How Does Scanning Affect What AI Can Fix?

The quality of the scan directly limits the quality of AI recovery. A phone photo of a damaged print (taken at an angle, with non-uniform lighting, uneven focus) gives the AI much less to work with than a proper flatbed scan.

Specifically:

• Resolution: scanning at 600 DPI gives the AI more pixels to work with for fine detail recovery. A 3.5x5-inch print at 600 DPI produces a 2100x3000-pixel image — enough for the AI to reconstruct fine texture. A phone snapshot at typical distance gives you perhaps 1000x1400 useful pixels with uneven sharpness.
• Lighting uniformity: flatbed scanners use even backlit illumination, which means crease marks and surface variations are captured cleanly. Phone photography creates shadows at raised edges and variable brightness across the photo, which the AI may misread as image content.
• Color accuracy: flatbed scanners have consistent color response. Phone cameras introduce automatic color correction that can interfere with the AI's own color restoration work.

If you have a photo that is fragile — lifted emulsion, active mold, extremely brittle paper — consult a conservator before scanning. Pressing a fragile print to a scanner bed can cause further damage.

Is It Worth Being Honest About What AI Cannot Do?

Many AI restoration services market themselves with before/after examples that show complete photos with only surface damage — fading, yellowing, scratches. That is genuinely what AI restoration does well.

What the marketing rarely shows is a photo with a corner torn off and what the AI returns. The answer is: the surviving three-and-a-half corners look dramatically better, and the torn corner remains a gap.

This is not a failure of the technology. It is the correct behavior of a tool designed to restore what exists, not to fabricate what does not. Understanding this distinction helps you know exactly what you will get, so the result is useful rather than disappointing.

For the surface damage that AI does handle — and for most families, that is the majority of what their damaged photos need — Old Photo Restoration produces results that would have required a professional retoucher five years ago. That part of the promise is real. The inpainting limitation is simply a separate problem that requires a separate tool.