Scan it right, erase the damage, upscale, and revive the color. A practical guide to restoring old photos with AI, including what it can't fix.
%20(2).png)
To restore an old photo with AI, you scan the print, upload it to an AI restoration tool, erase the scratches and stains, upscale the resolution, and revive the faded color. The whole process takes a few minutes and needs no design experience.
That is the short answer. The longer answer is that the quality of your result depends less on the AI and more on two things people skip: how well you digitize the print, and knowing which kinds of damage AI can actually repair. This guide covers how to restore old photos end to end, including both.
AI photo restoration handles the most common types of damage well: scratches, dust spots, stains, creases, mild tears, faded or yellowed color, soft focus, and low resolution. If you want to fix old photos that look tired rather than destroyed, AI will usually bring them back in one pass. Damaged photo restoration gets harder as the damage gets more structural, which is where the limits below matter.
There are honest limits. AI reconstructs missing areas by predicting what was probably there, so it struggles when the damage removes unique information: a torn-away face, a missing corner with a person in it, or heavy water damage across the main subject. It will fill the gap with something plausible, but plausible is not the same as accurate. For photos with severe structural damage, expect to run several passes or accept a professional restorer's hourly rate.
Knowing this before you start saves you from judging the tool against an impossible brief.
Restoration quality is capped by the scan. Garbage in, garbage out.
A flatbed scanner at 300 DPI or higher is the best option. For small prints you may want to enlarge, scan at 600 DPI. If you only have a phone, photograph the print flat, in indirect daylight, with no flash. Flash creates a glare band that reads as damage and confuses the restoration model.
Two more rules. Clean the print gently with a dry microfiber cloth before scanning, and save the scan as PNG or maximum-quality JPG. Keep this original file untouched. Every restoration attempt should start from a copy.
Once you have a clean scan, the restoration itself is short. Here is the workflow using Recraft's photo restoration tool. Other AI editors follow a similar sequence.
Open the restoration tool in Recraft Studio and upload your scan. Work from a copy of your original file, not the file itself.
Select Edit image and describe what you want fixed: remove the scratches, clean up the stains, smooth out the crease. Recraft applies the change and reconstructs the damaged areas from the surrounding image. Ask for one fix at a time rather than everything at once. Focused instructions produce cleaner results, and if a change goes wrong you only undo that one step.
Run the image upscaler to raise the resolution and sharpen soft detail. This is the step that turns a small, blurry print into something you can reprint at a larger size. One pass is usually enough. Upscaling twice tends to invent texture rather than recover it.
Adjust the color palette to correct yellowing and bring faded tones back. If you want to go further, the AI photo editor in Recraft Studio covers finer edits. Export as PNG for archiving or JPG for sharing, and you are done.
Faces. Faces are where restoration is won or lost, because viewers notice every error. If a face comes back looking slightly wrong, undo and try again with a more specific instruction aimed at just that area. Changing less at a time keeps the model anchored to the real features.
Heavy damage. For photos with large missing areas, restore in rounds: fix the worst damage first, export, and run the result through a second pass. Each round gives the model a cleaner base to work from.
Colorizing black-and-white photos. AI can add color to black-and-white images, and the results are often convincing. Be clear with yourself about what you are getting: the colors are the model's plausible guess, not a historical record. Your grandmother's dress may not have been blue. For family archives, many people keep both versions, the original black-and-white scan and the colorized copy.
Store three files: the untouched original scan, the restored version, and, if you made one, the colorized copy. Name them so the original is obvious.
Back them up in two places, such as cloud storage plus a local drive. The point of restoring old photos is that the next generation will not have to do this again from a shoebox.
Yes, within limits. AI repairs tears by reconstructing the missing area from the surrounding image, which works well for tears through backgrounds or clothing. If the tear removes unique detail such as part of a face, the reconstruction is a plausible guess and may need several attempts.
Yes. AI colorization produces natural-looking results in seconds. The colors are predictions rather than historical fact, so keep the original black-and-white scan alongside the colorized version.
You can restore old photos online free in Recraft Studio, which includes daily free generations. Heavy use, such as restoring a full family archive, fits better on a paid plan.
No, if you work correctly. Always keep the original scan as a separate file and restore a copy. The restored image exports as a new file.
Scan at 300 DPI minimum, or 600 DPI for small prints you plan to enlarge. Higher input resolution gives the AI more real detail to work with and reduces invented texture in the upscale step.
Restoring one photo usually takes under five minutes once the scan is ready. Start with the most damaged photo in the box. If the workflow handles that one, the rest of the archive is easy. Try it with Recraft's free AI photo restoration tool.
.svg)
.webp)
