Restore an Old Photo with AI
Prepare a faded or damaged photograph for a new print without losing the family original. This restoration workflow is intended for old portraits, scanned albums, and photographs with visible wear. It can improve contrast and reconstruct some missing detail, but it cannot know the exact history of a torn or overexposed area. The result should be treated as a restoration interpretation, kept beside the scan, and reviewed before anyone treats it as a faithful record.
Use this without the search next time. Prathom Workbench puts Prathom's tools in your toolbar.
Add to Chrome — freeDrop an image here, or click to browse
Up to 25 MB. Uploaded temporarily, deleted after processing.
What it does
- Restoration workflow for faded images
- Non-destructive source handling
- GPU job progress and temporary output
- Review-ready image download
How to use AI Restore Old Photo
- 1
Scan the original
Use the highest-resolution scan available and avoid photographing a glossy print with reflections.
- 2
Start restoration
Queue the image on the configured GPU host and allow the workflow to finish before judging the output.
- 3
Compare the history
Keep the scan beside the result and mark areas where the model had to infer missing detail.
How it works, and what it does not do
Three operations, in order. The scan is colorized by DDColor, then sharpened, and the sharpened version is blended back over the colorized one at partial strength so the sharpening is applied gently rather than at full force.
That is the entire graph, and the omission in it is the most important thing on this page: there is no inpainting step. Nothing here fills a hole, rebuilds a torn corner, removes a crease, or paints out a stain.
"Restore" is a broad word and it is worth narrowing before you upload a photograph you care about. What this does is color and clarity. What it does not do is repair.
Why damage repair is a genuinely different problem
Colorizing asks a model to add a layer of information over structure that is fully present. Every pixel is there; only the hue is missing.
Repairing damage asks it to invent structure that is absent. Where a photograph is torn, the file contains no record of what was behind the tear, so anything that appears there was decided by a model rather than recorded by a camera. On a family photograph that distinction matters more than usual, because the thing being invented is often part of a person.
Those two jobs need different machinery, and running one does not partly accomplish the other. A creased photograph put through this workflow comes back as a colorized, slightly crisper photograph with the same crease.
What it does handle well
General fading. An old print that has lost contrast overall — no single defect, just a uniform tiredness — is the case this was built for.
Softness from the scan itself, or from the original print being a contact copy or a reprint. The gentle sharpen addresses exactly that.
Grayscale prints that would read better in color for a modern audience, which is usually the reason someone wants an old family portrait processed at all.
What it will not fix
Tears, creases, missing corners, punch holes, and anything stuck to the surface.
Foxing, the brown speckling on old paper, which will be colorized as though it were part of the image.
Silvering, the metallic sheen on the dark areas of old prints, which the model reads as tone rather than as damage.
Severe blur or camera movement, which is not softness and is not addressed by sharpening.
In each of these the honest sequence is to repair first, by hand or with a tool built for it, and to run this afterwards.
Before you scan
Scan the print rather than photographing it, at the highest optical resolution available, with any automatic sharpening and dust removal switched off. Those features guess, and you want the guess to happen once, visibly, in a step you chose.
Clean the glass and the print surface. Dust becomes permanent detail.
Keep the original scan untouched and work from copies. The point of restoring a family photograph is to end up with more than you started with, and overwriting the only digital record of the print is the one outcome that cannot be undone.
Examples
Faded album portrait
The useful improvement is often better separation and cleaner print preparation, not a dramatic change to the people in the frame.
Scratched studio print
Inspect eyes, hair, jewelry, and clothing because restoration models may smooth meaningful small features.
Frequently asked questions
Can restoration recover a completely missing face?
It may generate a plausible-looking result, but a plausible face is not the same as a recovered face. If an important area is missing, preserve the original scan and label the output as an AI-assisted restoration. Do not present inferred identity details as historical evidence.
Should I scan before restoring?
Yes. A flat, high-resolution scan gives the workflow more information and avoids glare, perspective, and camera noise. Keep the untouched scan in a separate folder, then create restoration copies so every later edit can be compared with the source.
Is an old family photo uploaded?
The GPU version sends the image to the ComfyUI host configured for the Prathom workspace. It is not a browser-only operation. Treat the host as a processing service, use the page only with material you are allowed to process, and download the result before the temporary job expires.
Is the AI Restore Old Photo tool free, and do I need an account?
No sign-up. The restore old photo ai workflow takes an image up to 25 MB, runs it on the GPU host and hands back a file you can download directly. Nothing is charged and nothing is stamped onto the output. Because no account exists, nothing keeps your history either, so save the result within the thirty minutes the job is retained.