Yiye Renjian

Old-photo restoration and watermark removal tutorial: turn yellowed, scratched prints clear again, step by step

old-photo restorationwatermark removalimage tutorial

A rediscovered family photo is yellowed and creased, and the scanned copy carries watermarks and serial numbers in the corners. You want it clear again, but fear the restored faces will no longer look like the people. This tutorial uses Yiye Renjian's old-photo restoration and watermark-removal tools to break "turning a yellowed, scratched photo clear" into steps you can follow, with a note on each switch for whether to turn it on.

Step 1: Diagnose the photo's problems

Zoom into the photo and look it over. Problems usually fall into four kinds: heavy noise and grain, scratches and creases, overall fading and yellowing, and watermarks with serial numbers on edges or corners. Restoration fixes the first two kinds and enhances detail along the way; whether to restore faded colors is your call, and watermarks need the dedicated watermark-removal flow. For scans with insufficient resolution, restore first and upscale after. Reversing the order enlarges the flaws together with the photo.

Step 2: Restore the old photo

  1. Upload the photo: a single photo must not exceed 8MB. Blurry, scratched, creased and faded photos are all accepted.
  2. Decide the two switches:
    • Recolor: gives black-and-white or faded photos natural colors. Leaving it off strictly preserves the original tones and period feel. For likeness-critical uses like ID or archive photos, start with it off.
    • HD upscale: enlarges the restored result to a higher resolution. Leaving it off keeps the original resolution, for cases where you only want scratches gone without changing dimensions.
  3. Click "Start restoration": the AI handles noise, scratches, damage and facial details while keeping the person's identity, likeness and age unchanged. Skin keeps its real texture instead of being redrawn into someone else.
  4. Review and download: if unsatisfied, toggle the recolor or upscale switches, or rerun with a different model and then pick.

After restoring, first confirm "it is still the same person": facial features, age, composition and light direction all unchanged means a qualified result. Looking younger or slimmer is a different request, outside this tool's scope.

Step 3: Remove the watermark

  1. Upload the watermarked image: a single image must likewise not exceed 8MB, and confirm that you hold the legal rights to the image.
  2. Pick the removal mode:
    • No brushing: automatically detects and removes every watermark in the frame, fitting scattered, numerous watermarks.
    • Click "Brush": circle the watermark spots so the AI repaints only those small patches, matching the original style and fonts while leaving everything else untouched. Use this when a watermark sits on the subject, or when automatic removal comes out poorly.
  3. Click "Start removal": the consumption for this run is shown before generating, and credits are deducted per use.
  4. Check against the original: if residue remains, widen the brushed area and run again, or retry with a different model.

Before uploading you must confirm you hold the legal rights to the image and will use it only lawfully. Removing someone else's watermark and using the result commercially is infringement and will be handled under the Acceptable Use rules, see Acceptable Use.

Four common pitfalls

  1. Turning recolor on right away: an old photo's value often lies in its period feel. Repair only the damage first and check the result, then decide on color.
  2. Brushing too small an area: circling half a character leaves a ring of residue at the edges. Leave an extra margin and circle the watermark together with its transition zone.
  3. Upscaling a tiny source directly: noise and scratches get enlarged together. The correct order is repairing scratches and damage first, then ticking "HD upscale".
  4. Uploading when rights are uncertain: if the watermark, portrait or copyright status is in any doubt, do not process it. The risk outweighs one image's gain by far.

Start now