No. 05 — AI Tools · Cloud HDPro · on our GPU

Grainy old photos, smooth again.

Some of that speckle is the photograph and some of it is an error, and the two want opposite treatment. Our server rebuilds the faces, which clears the grain from exactly the region you were squinting at — and leaves the rest of the frame honestly alone.

Pro · Cloud HDDeleted on completionUp to 25 MB · 24 MP
Restore a grainy photoCloud HD · our GPU

How it works

  1. Pick the grainy photograph of a person — JPEG, PNG or WebP, up to 25 MB and 24 megapixels.
  2. Our GPU restores the faces and checks each one against your original. If the result drifts too far from the person in your photo, the run is refused rather than handing back an invented face — and you are not charged.
  3. Download the restored PNG at the same resolution, then judge at 100 per cent whether the background grain needs anything at all.

This page is the photo restorer, tuned to grain: where it comes from, and how much of it you should actually remove.

All Image Tools

AI Tools

Solutions by use case

Fix Grainy Old PhotosPro, on our server

Grain and noise look similar on screen and come from opposite places. Film grain is physical. The image on a negative is made of clumps of silver crystals, and faster films use larger crystals, which is why a photograph shot indoors in 1978 on high-speed film is visibly coarser than one shot outdoors on slow film the same afternoon. That texture was in the photograph on the day it was made; it is a property of the original, not a defect that crept in later. Digital noise is the opposite — an electrical artefact produced by a sensor straining in poor light, showing up as fine speckle and coloured blotches in the shadows, representing nothing that was ever in front of the lens. Your scanner can add its own version of the second kind. Deciding which you are looking at is the whole of the diagnosis, and it takes one look at the shadows at 100 per cent.

Restoration does something specific to grain, and only in one place. RestoreFormer++ does not filter or smooth your photograph; it rebuilds each face it finds and blends that reconstruction back into the frame. Note too that a photograph with no face in it for the model to find never runs at all — the tool says so and spends nothing. The practical effect is that grain inside a face is replaced by reconstructed skin, lashes and hair, while grain everywhere else — the coat, the wall, the sky — comes back exactly as it went in. On a moderately grainy print that is precisely what you want, since the face is the part you were straining to see. On a very grainy one it can produce a visible mismatch: a clean, detailed face sitting in a speckled scene. Look at the result at full size before you decide you like it, and if the mismatch bothers you, a light global denoise applied afterwards in your own editor evens it out better than anything applied before the run would have.

Several scanning decisions make grain worse, and one of them is counter-intuitive. Scanning at a very high resolution can make a print look coarser rather than finer, because past a certain point the scanner stops averaging crystals into pixels and starts resolving the crystals themselves, sometimes with a mottled interference pattern on top. If 1200 DPI looks worse than 600, believe the 600. Scanner sharpening amplifies grain enthusiastically and should be off — the scanning workflow page lists the rest of the settings worth disabling. And do not confuse either of these with compression blocking, which is square, regular and concentrated around edges; telling the two blurs apart covers how to spot it. Real physical damage sitting in the grain — spots, tide marks — is a job for the brush on the stain remover.

Remove less than you think, and know where the file goes while you decide. A photograph scrubbed clean of grain often reads as synthetic, with skin like moulded plastic and fine textures gone; grain is part of how a print from 1963 announces that it is a print from 1963. Do any denoising gently, last, at final output size, and always with the original open beside it. The one step that leaves your machine is this one: restoration is Cloud HD, our Pro-only, opt-in server-side feature set, running on a GPU because the model is too heavy for a browser. Your photograph is encrypted in transit, uploaded to our server for that run and deleted when the run finishes, and in all cases within 24 hours, processed in the United States on Cloudflare R2 storage and a Modal GPU, and never used to train anything. The restoration hub covers the identity check on every rebuilt face, and the on-device upscaler handles any enlargement afterwards without uploading anything.

How it works

  1. Work out which one you have: Grey, even, organic speckle across the whole frame is film grain. Coloured blotches concentrated in the shadows are digital or scanner noise.
  2. Scan at a sensible resolution: 600 DPI for a small print. Going higher can start resolving the grain crystals themselves and make the file look coarser, not better.
  3. Restore the faces: Upload here, up to 25 MB and 24 megapixels. The faces are rebuilt, which replaces the grain within them; the background keeps its own.
  4. Judge the mismatch before denoising anything: Open the result at 100 per cent. Only if the clean faces clash with a speckled background is a light, global denoise worth applying at the end.

Frequently asked questions

What is the difference between film grain and digital noise?
Film grain is physical: clumps of silver crystals in the emulsion, larger on faster films, and genuinely part of the photograph as it was taken. Digital noise is electrical, produced by a sensor working in poor light, and appears as speckle and coloured blotches that were never in the scene. One is a property of the original; the other is an error introduced by a machine.
Does the restoration remove the grain?
Inside a face, effectively yes — that region is rebuilt rather than filtered, so reconstructed detail replaces the grain that was there. Everywhere else the grain is untouched, because nothing outside the faces is altered. On a very grainy photograph that can leave a clean face against a speckled background, which is worth checking at 100 per cent before you keep the result.
Why did my scan get grainier when I scanned at a higher resolution?
Because you started resolving the grain itself. Below a certain resolution the scanner averages several crystals into one pixel and the image looks smooth; above it, individual clumps become visible and can interfere with the scanning grid to produce a coarse mottling. If a print looks worse at 1200 DPI than at 600, that is not a fault — take the 600.
Should I be removing grain at all?
Often not. Grain is evidence of the medium, and a photograph stripped of it can look strangely synthetic, with skin like moulded plastic. Heavy denoising also eats the fine texture that made the picture worth restoring. If you do denoise, do it gently, after restoration and at final output size, and compare against the original before committing.
My photo has coloured speckles rather than grey grain. What is that?
That is chroma noise, and it is a digital signature rather than a film one: either the camera was working in poor light, or your scanner is adding it. Rescanning at a lower resolution with a longer exposure setting often helps if the scanner is the culprit. If the photograph was born digital in a dark room, the speckle is baked into the file.
Is a grainy scan retained anywhere after it is processed?
No. It is uploaded to our server for that run and deleted when the run finishes, and in all cases within 24 hours. The transfer is encrypted, the work happens in the United States on Cloudflare R2 storage and a Modal GPU, and nothing you upload is ever used to train anything. This is Cloud HD, the only part of MiniPx that uploads an image at all.