imvori“Restore” can mean reducing dust, recovering faded contrast, improving a face or adding color to a black-and-white photograph. These are different operations. Imvori separates them so you can choose an appropriate treatment, compare it with the original and stop before the image looks overprocessed. None of these tools can establish what a missing detail or an unknown historical color actually was.
Open Restore old photoStart with the best copy you have
Use a scan or an evenly lit photograph of the print. Avoid glare and keep the whole photograph in frame. Clean the scanner surface before scanning rather than relying on software to remove a large mark later. Keep an untouched original. A small, heavily compressed social-media copy contains less useful information, and enlarging it does not bring back the original film or paper detail.
Repair small defects and faded contrast
Restore old photo runs on the server CPU. It detects small bright dust spots and thin scratches conservatively, fills selected defects from their surroundings and blends a contrast correction. Try Light before increasing the strength. Wide tears, dark scratches, stains and missing areas may remain. Thin bright features that belong to the photograph can sometimes be mistaken for defects, so compare faces, jewellery, hair and architectural lines before saving. The source limit is 12 megapixels.
Treat faces as a separate decision
Restore face uses a pretrained face-restoration model and can raise the overall image resolution. Reconstructed features may differ from the actual person; the output is not evidence of details absent from the original. Start with a gentle strength and compare the eyes, mouth and face shape. The resolution limit is separate from old-photo repair: the resulting upscaled image must remain within the stated output pixel limit.
Add color as an interpretation
Colorize photo predicts color channels from grayscale using a pretrained model on the server CPU. It retains the image’s luminance structure while introducing estimated colors. Clothing, skin, buildings and vegetation can receive plausible but incorrect colors. Processing strength controls the intensity. Colorization does not recover a photographer’s original color information and should be labelled as a colorized interpretation when historical accuracy matters.
Compare and keep more than one result
Apply one treatment at a time and inspect the current version before the next operation. A repaired version can be used as input to colorization, and Undo returns to its parent. Download a repaired monochrome copy as well as a colorized copy if both are useful. Server uploads and results become eligible for scheduled cleanup after 24 hours; download them before expiry. Local history does not survive a page reload.
Need help with this workflow? Report a problem without sending the private image itself.
Try the tool Explore all tools