Rebuilding a photo’s missing pieces
A missing corner rebuilds reliably from the photo around it. A missing face is a guess — here is how to tell which one you are looking at before you pay.
A corner missing from a photograph is barely a decision — nobody hesitates over rebuilding a lost patch of sky or the arm of a sofa. A missing piece of someone’s face is a different question, and it deserves a more honest answer than most restoration services give it. Some missing pieces rebuild from real evidence still sitting in the photo. Others can only be guessed at, however convincing the guess looks — and knowing which is which changes what you should expect from the result.
What rebuilds reliably: edges, background, clothing
A missing corner, a chewed edge, a bald patch of wall or sky usually restores well because those areas are made of repeating or predictable structure. A wall continues in the same texture and light; a sky gradates smoothly from one visible edge to the other; a sleeve follows the arm underneath it, a shoe follows the leg standing on it. Rebuilding that kind of loss is closer to careful extrapolation than invention — the surrounding pixels already contain most of the answer, and the missing patch is filled in to match exactly what is already there.
The same logic covers most clothing, furniture and background objects, even when a fairly large piece is gone. A missing shoulder of a jacket can be inferred from the visible shoulder, the collar, and the way fabric normally falls. None of it requires knowing anything about the specific person in the photograph — only that the physics of light, fabric and perspective stay consistent with what the rest of the image already shows.
What is a guess: a face that is not fully there
A face breaks that logic completely. Faces are not repeating patterns — they are asymmetric and individual, full of detail that has no twin anywhere else in the photograph. A mole on one cheek does not imply one on the other. A scar, a gap in a smile, the particular tilt of one eyebrow — none of it can be recovered by extending the surrounding pixels, because the undamaged half of a face does not contain the information the damaged half is missing. Filling that area in means generating a face that is statistically plausible for the pose, lighting and apparent age — not recovering the one that was actually there.
This is the boundary Epoha is built to respect rather than paper over. A restored face is checked against whatever of the original survives, and where too little of a face remains for that check to mean anything, the honest answer is that the missing part is a reconstruction, not a memory — a believable face-shaped patch, not the person. We say so plainly rather than hand back a confident stranger and let you assume it is them.
How much of the face is actually left?
- Only the background, a corner or clothing is missing, and both eyes, the nose and mouth are intact: reconstruction is reliable — built from the structure around the loss, not from a guess about the person
- Part of the face is gone, but one eye, part of the nose or mouth, and the jawline or hairline survive: the missing half can be rebuilt anchored to that person’s real proportions and tilt — still a reconstruction, but one with real evidence behind it
- Both eyes are gone, or the whole face is missing: any result is a plausible face for that pose and era, not a recovered likeness — treat it as an illustration, not a record of the person
- Not sure which case you have? Upload the scan and look at the actual result before paying — that judgment call is exactly why the preview comes first
If the missing piece still exists somewhere
Sometimes a piece is not truly missing, only detached — a corner that tore off years ago and got tucked into an envelope, a strip that came apart along an old fold. If any physical fragment of the lost area still exists, that changes everything: scan it alongside the main piece and lay them in their true positions, because reassembling a real fragment beats reconstructing anything, however good the reconstruction looks.
The distinction matters because the two paths carry different confidence. A reassembled fragment restores what the camera actually captured — the torn edge is healed, but the content is real. A reconstructed area, whether it is a lost corner of sky or half a face, is the most plausible fill restoration can produce from what remains. Both are legitimate ways to end up with a photograph worth looking at again; only one of them is a record rather than an estimate, and it is worth knowing which you have.
Missing pieces and damaged faces — questions
Can AI actually fix a missing corner or edge of an old photo?
Yes, reliably — corners, edges and background rebuild from the structure already visible around them, which is the strongest case for digital reconstruction. That is very different from rebuilding a missing face, and the result reflects the difference in confidence.
Half my grandfather’s face is torn away. Will the restored photo actually look like him?
It depends how much survives. If one eye, part of the nose or mouth, and the jawline are still there, the reconstruction anchors on his real proportions and the likeness usually holds up well. If nothing recognizable is left, no service can know what he looked like from a blank space — you would be looking at a plausible stranger, and that is what we will tell you rather than pretend otherwise.
How does the AI know what a missing face looked like?
It does not know — it infers a plausible face from the surviving half, the pose, the lighting and the era, the way a sketch artist works from a partial description. That is often convincing, but it is an informed guess, not a record, once more than about half the face is gone.
What if there is no face left at all — just a hole in the photo?
Then any restoration, ours included, can only invent a face that fits the scene, not recover the person who was there. The more honest outcome is usually restoring everything else in the photo well and leaving that area as it is, rather than papering over a total loss.
See exactly how much comes back
Upload the damaged scan and look at the restored face yourself before paying anything. If too little survived for a trustworthy result, you will see that too.
Epoha is in early access: the restoration model is not connected yet, so the tool currently shows how the flow works. The price and the pay-after-you-see-it promise apply once it opens.