AI Old Photo Restoration

Guide: restoring old family photos realistically

What this restoration covers

The tool is built around AI super-resolution. It runs an ESRGAN-slim neural network in your browser that enlarges your scan to twice its width and height while predicting sharper edges and finer texture. After that, it applies a small contrast boost, a slight colour saturation increase and a gentle sharpening pass. The result is saved as a PNG at double the original dimensions, so a 1000 by 700 scan comes back at 2000 by 1400.

In practice this makes a small, soft or low-resolution photo look clearer and larger, which is ideal for reprinting or sharing on a bigger screen. The first run downloads the model, about 5 MB, which is then cached.

What it will not do

Being honest about limits saves you time. The tool does not repair physical damage on its own. Cracks, folds, torn corners, stains and missing pieces stay in the image, and sharpening can make them slightly more visible. It does not colourise black-and-white photos; the saturation step has no effect on a purely grey image. It does not rebuild faces from scratch, so a face that is just a few blurry pixels will become a larger, cleaner blur, not a detailed portrait. It also does not correct strong colour casts such as the orange or magenta tint common in faded prints from the 1980s and 1990s.

A practical restoration workflow

• Scan or photograph the print well. Lay it flat in soft daylight, hold the phone parallel to the photo, and avoid glare from glossy surfaces. A flatbed scanner at 300 to 600 dpi is even better.
• Crop and straighten with the Crop tool and Rotate tool so only the photo is left, with no album page or table edge.
• Remove dust and small scratches with the Object Remover. Brush over each spot separately; this works best on small marks over plain areas such as sky or walls.
• Run the restoration on the cleaned image.
• Compare at full size using the before-and-after view. If the result looks too sharp, the original scan may already have been good enough.

Getting the most from a scan

Since the output is double the size of your input, it helps to think about the final use. For sharing on WhatsApp or a family group, a scan around 1000 pixels on the longest side gives a comfortable 2000 pixel result. For a print, a larger scan gives more to work with, but very large images take noticeably longer and use more memory because the network processes the image in small patches. If your browser struggles, reduce the scan with the Image Resizer and try again.

PNG files can be large. To email a restored photo or upload it to a form, convert it with PNG to JPG or shrink it with Compress to Size. Always keep the original scan as well; future tools may do even more with it.

Which photos benefit the most

The biggest improvement is usually on small prints, such as passport-size photos, contact prints and the small square prints from older cameras, because the scan of a small print has few pixels and the doubling makes a real difference. Photos that were sharp when taken but have simply faded also respond well, since the network has real edges to work with.

Photos that were blurry from the start, such as a shaky snapshot from a wedding in the 1970s, improve less. The tool can make edges cleaner, but it cannot add detail that the camera never recorded. When you have several copies of the same picture, for example one in an album and one framed, scan the sharpest one rather than the best-looking one, because sharpness is much harder to recover than brightness.

Before you start

• The original print is handled carefully and stored safely after scanning.
• The scan is PNG, JPG or WEBP and cropped to the photo itself.
• Dust spots and small scratches have been brushed out first, if any.
• You expect a clearer, larger photo, not colourisation or rebuilt faces.
• You have a stable connection for the first model download.

Related guides

• How to Restore an Old Photo With AI
• How to Make a Small Image Bigger Without Losing Quality