There's a photo on my grandmother's windowsill — a black-and-white portrait of her parents, taken sometime in the late 1930s. The bottom left corner is gone, torn away decades ago in some forgotten incident. There's a deep scratch running diagonally across the man's face. The paper has yellowed and developed a texture like old parchment. For years, the family accepted this as simply what the photo looked like now. It didn't occur to anyone that it could look any different.

It can. That's what AI photo restoration does — and understanding how it works helps set realistic expectations for what's possible with your own damaged photos.

The Types of Damage AI Can Repair

Not all damage is the same, and AI handles different types in different ways. Here's a practical breakdown:

Surface scratches and scuffs are among the easiest problems for AI to fix. These appear as thin lines or areas where the emulsion layer has been disturbed. The AI identifies them as anomalies inconsistent with the image content and removes them, filling in the area from surrounding context. The result is typically seamless.

Color fading and cast — the yellowing, desaturation, or color shift that affects virtually all older prints — is also well within AI's capabilities. The model learns the statistical relationships between colors in well-preserved images and applies corrections to bring the damaged photo closer to its original tone. Faces that have gone orange or grey recover their natural skin tones. Blue skies that have faded to white regain depth.

Water stains and mold spots are more complex. These appear as irregular blotchy areas that obscure the image content beneath them. AI handles these by analyzing the areas surrounding the stain and reconstructing what was likely underneath. The results depend on the severity — a light stain over a plain background is nearly invisible after processing; a heavy stain over a detailed face is more challenging, though still often dramatically improved.

Tears and physical losses are the most technically demanding. When part of an image is simply missing — a torn corner, a section that was cut or burned — the AI must reconstruct content that wasn't recorded in the original. This is done through inpainting: the model generates plausible content based on what exists in the surrounding image, the patterns typical of the image type, and training on millions of examples. For background areas and simple textures, this works remarkably well. For faces, it requires more care but can produce convincing results when the missing area is limited.

The Resolution Question: Why 8MP Matters

One of the persistent challenges with old family photos is resolution. A print that was sharp and detailed when it was new may have blurred and lost definition over decades of physical degradation. Even a high-quality original from the 1950s or 1960s may have been captured on film that, when scanned, doesn't produce a large digital file by modern standards.

AI restoration addresses this through super-resolution processing: the model upscales the image while simultaneously restoring it, adding detail that's consistent with what should be there based on the image content and training data. The result from Fotki's Full Restore mode is an 8-megapixel image — that's 3264 × 2448 pixels, or roughly equivalent to what a modern mid-range smartphone captures.

To put that in practical terms: an 8MP image can be printed at 11×8 inches at standard print resolution (300 DPI) without any visible softness. It displays at full quality on any modern monitor or phone screen. It's large enough to share, frame, include in photobooks, and archive — and small enough to be manageable. For a restored family photo, 8MP is genuinely sufficient for everything most people want to do with it.

What to Expect from Severely Damaged Originals

It's important to be honest about what AI restoration can and cannot do. The AI fills in missing information based on inference — statistical patterns learned from training data. This means the results are plausible reconstructions, not recovered originals.

For a face with a scratch running through it, the result typically looks natural and seamless because the surrounding context provides enough information for the model to make accurate inferences. For a large torn section through a face, the reconstruction will be plausible but may not precisely match what was originally there — because that information simply doesn't exist in the image anymore.

This is still profoundly useful. A plausible, natural-looking reconstruction of a face is incomparably better than a tear or a white void. The restored image conveys the person, the moment, the feeling of the original — even if every pixel isn't a perfect recovery of what the camera captured in 1952.

Practical Steps for Photographing Damaged Originals

The quality of the restoration depends significantly on the quality of the input. A few practical tips for getting the best results from damaged photos:

Use diffuse, even light. Direct sunlight or a single bright lamp creates harsh shadows and glare that add to the apparent damage. Overcast daylight or a well-lit room with no direct sun is ideal. If the photo has texture from aging, angling the light slightly can sometimes make damage more visible rather than less — so experiment.

Fill the frame. The more of the sensor you use for the photo, the more detail the AI has to work with. Get close enough that the photo fills most of the screen without cutting off the edges.

Hold steady or use a surface. Motion blur from an unsteady hand adds softness that's harder to fix than chemical fading. Rest your elbow or lay the phone flat with a timer if needed.

Don't crop before restoring. Include the full photo, including damaged edges and torn areas. The AI needs the complete context to make the best restoration decisions.

Three Modes for Different Levels of Damage

Fotki offers three restoration modes designed for different situations:

Quick Save is ideal for lightly worn photos that need surface cleanup — minor scratches, dust spots, slight color correction, and exposure adjustment. It's fast and uses 1 coin.

Details goes deeper, with face enhancement and recovery of fine detail. It handles moderate tears, water damage, and works particularly well for portrait photographs. This is the most versatile mode for family photos. It uses 2 coins.

Full Restore is the maximum processing mode, delivering the 8MP output. It reconstructs heavily damaged areas, restores missing sections, and applies the full range of AI enhancement. For severely degraded originals — heavily scratched, torn, or faded to near-illegibility — this is the mode that makes the greatest difference. It uses 3 coins.

After Restoration: What to Do with the Result

Once you have a restored image, you have something that works in ways the original couldn't. You can print it — at 8MP, the resolution is sufficient for a good-quality print up to 11×8 inches. You can share it digitally, with no concerns about further physical degradation. You can include it in a family archive, a photobook, a digital frame.

Consider sending copies to family members who may not have seen the photo before, or who only know it in its damaged state. For a grandchild who has only ever seen a grandfather as an elderly man, a clear, restored photograph of him at thirty is something remarkable — the face recognizable, the youth surprising, the connection across time suddenly real and visible.

That's what restoration ultimately provides: not just a better image, but a clearer window into the past. And for family photographs, that clarity matters more than any technical specification.