Remove text from image: clear distracting text from your picture

Captions, subtitles, dates, stickers and price labels can cover a picture that is otherwise useful. The removal mask tells the model which area you want replaced. It estimates a continuation of the surrounding surface, such as a wall, sky, cloth or product background.

The goal is a localized edit, rather than a rewrite of the entire picture. Keep the brush away from faces, logos and nearby lettering you want to retain. Compare those details afterward; a mask does not by itself prove that all unmarked pixels were preserved.

Each run costs 6 credits. Keep the original so you can compare the reconstruction and try a more careful selection if necessary. Save the selected result within the 24-hour file-retention window.

Try an added social caption, camera date, burned-in subtitle in a still screenshot, handwritten note or product sticker. A simple or repeating background provides more context than a large block of text covering a face or a complicated pattern. For several overlays, work on a manageable area and review each paid run before moving on.

• It removes selected lettering and estimates a fill; it does not retype a word, translate a caption or insert replacement text.
• It cannot verify what was hidden beneath a large overlay. Reconstructed faces, hands and patterns may be inaccurate even when the fill looks convincing.
• Use it for marks you own or have permission to remove. It does not grant rights to remove another creator’s watermark, signature or credit.
• It is not an upscaler. This workbench accepts files within its 15 MB / 40 MP limits and does not promise automatic 4096px resizing or a particular restored output size.