Remove Specific Words from an Image

Define removal and protection rules

Suppose your image contains a sale label and a brand name. Add the sale wording to Words to remove and the brand to Words to keep. TextWiper has no built-in removal list, and an empty removal list does not erase anything.

Keep rules take priority when a detection matches both lists. They add protected areas around matching detected text. Protection still depends on recognition: if a brand is missed, paint its area with Protect before repairing.

Use commas or new lines to separate entries. OCR errors and phrase layout can affect matches; inspect what is actually selected.

Use keyword rules on one image

1. Choose the image. Add a JPG, PNG or still WebP. A file can be up to 10 MB, 16 megapixels and 4096 pixels per side.
2. Choose a detection language. The available OCR options are English and Chinese + English. Changing the interface language does not add a new OCR language.
3. Enter your words. Open Use your keyword library, fill the removal and keep lists, then choose the match style.
4. Scan and match. Use Scan & match my words to detect text and create suggestions. If you have already detected the image, Apply to detected text uses the current detections.
5. Check the mask. Green marks removal and blue marks protection. Brush can cover a missed letter; Erase can shrink an area; Protect can preserve nearby details.
6. Repair and compare. Choose Quick repair or Local AI repair, repair the selected area and compare the result before downloading.

Reuse the rule across a batch

The batch workspace accepts up to 10 images and 40 megapixels total. Use Suggest matching words when you want to inspect selections before repair. Use Scan & erase matches only when you are ready for automatic matching and repair across the queue.

For repeated work, save your entered rules as a template. An optional signed-in account can store keyword templates; signing in is not required to edit an image. Templates store your rules, not a cloud copy of the images.

Troubleshoot a missing or unwanted match

The word is visible but was not matched.

Check the detected text first. Small, rotated, decorative or low-contrast letters may be missed. You can select a detected region or brush over the intended letters manually.

A short rule selects too much.

Switch from Contains phrase to Whole word or phrase, use a more specific phrase and inspect the removal mask.

A word appears in both lists.

The keep rule takes priority. Remove it from the keep list only if you intend to erase that text, then apply your suggestions again.

A completed image needs different rules.

Apply new suggestions or edit the image before repairing again. Inspect the new result; previously completed results do not automatically prove that a changed rule was applied.

Save the result in the right format

The repair fills the selected pixels using surrounding information. It does not retrieve the original background hidden underneath the text. Local AI's first use downloads approximately 28 MB; batch processing remains serial.