Erase a photo

Remove text from an image

Box the words and they vanish. The picture carries on underneath.

or drop it here · paste with Ctrl+V Open the highest-quality authorised source you have. The mask and fill stay on this device.

No photo handy?

FreeNo sign-upPhotos stay on your device

Before and after: a caption band reading Your words here, then the same photo with the words gone
Choose photo

The rectangle tool and edge-march fill are selected for narrow glyphs. Box the whole word plus its outline, then inspect whether the ground continues through it.

The opening state is a starting judgement

Lettering is normally long and thin, so a rectangle describes it more cleanly than a freehand brush and an edge-march fill can advance from intact pixels on both sides. A title crossing a face or building is no longer a narrow-ground problem.

The selected mask instrument and fill method are visible and can be changed. Draw slightly beyond the visible object so shadow, antialiasing and colour spill are included, then remove any legitimate edge the mask crossed before running the fill.

A disciplined removal pass

  1. Mark the complete footprint. Include the object, its soft shadow and any reflected colour that belongs to it.
  2. Choose from the surrounding evidence. Use a narrow march for a stroke, patch search for texture and the disclosed model only when the hole needs structure.
  3. Read the divider at one-to-one. Look for repeated texture, bent straight lines and a smooth island where natural grain should continue.
  • No fill reveals a hidden original; every new pixel is inferred from visible context.
  • A unique face, sign or architectural edge behind the mask cannot be recovered exactly.
  • Several small masks are usually safer than one huge region crossing unrelated surfaces.

Questions about the repair

What pixels are changed?
Only the mask and its feathered edge; every other pixel is written out exactly as it arrived. On a picture larger than the screen each removal is computed on the full-size picture and shown reduced, so the saved file holds what you approved.
Which fill should I trust?
Edge march suits very narrow marks, patch search suits repeated natural texture, and the optional structure-aware model suits a larger hole. Each fails differently, so the before-and-after divider matters more than the label.
Does the model receive my photograph?
No. If selected, its disclosed weights are downloaded to the browser after consent. Inference then runs on this device and no image request is made.

Related removal workflows

Film date stampsAutomatic orange-imprint proposal for scanned prints.Remove a captionA caption band, cropped off or filled under.Text from a screenshotFlat UI backgrounds, filled cleanly.Remove handwritingPen ink on a page, by colour.
Photographic visual for Remove Text
Remove Text