Blog
Log in

Remove Text from Image Online (AI LaMa)

Erase captions, subtitles, meme text and banners from photos or stills. Brush over the text and the LaMa neural network fills in the background: wall, sky or artwork.

Text removal
Scenario

Cover the letters with a small margin. For memes, work line by line from top to bottom.

Drag an image here or click "Open"

Choose a text type, paint over the text, and click "Remove text". LaMa will restore the background under the letters.

Ready, open an image with text

Instructions

  • Upload a photo, still or meme (PNG, JPEG, WebP).
  • Choose a scenario: regular caption, subtitle line or colored banner.
  • Brush over the text completely. For long lines, work in segments.
  • Click "Remove text" (LaMa method by default) and download the clean PNG.
  • You can then add new text in your editor. The background is already prepared.

Повний опис

AI text erasing, not just painting over

Captions on photos, subtitles on stills, meme text and bright banners often need to be removed before localization, redesign or reusing a frame. Simply filling with a single color looks crude. Removing text from an image with LaMa restores the wall, sky, asphalt or artwork background beneath the letters.

The service is specialized for text: brush presets are tuned for lines, subtitles and banners. This is not a generic "remove something" tool. The interface and hints guide you specifically toward a clean frame ready for new text.

Note: the site also has a "Remove Text from Photo" tool with local in-browser filling. This service is server-side AI LaMa for more complex backgrounds and a more natural result.

Typical scenarios

  • Localizing memes and infographics: erase the old text, add new text in your audience's language.
  • Cleaning subtitles from stills before editing or previews.
  • Removing "Sale" banners, broadcast dates, UI overlays from screenshots.
  • Preparing photos for adding your own headline in Canva or Figma.

How to get a clean edge

Paint the mask tightly around the glyphs, but leave 1-2 pixels of margin. Letter anti-aliasing often leaves a "halo" that should also be covered. Long subtitles are better removed in two or three passes than with one huge mask across the whole face.

Zoom and the Hand tool help with small text. If the background is very textured, stick with LaMa; for a solid banner on a plain color, a quick OpenCV fill is sometimes enough.

Frequently asked questions

How is this service different from other "text removal" tools on the web?
Other tools do the filling locally, mostly in the browser. This service sends the image and mask to a server and uses the LaMa neural network for more realistic reconstruction of the background under letters, subtitles, and banners, especially on photographic textures.
Is it suitable for memes and localizing infographics?
Yes. The "Caption" scenario is convenient for classic top/bottom lines. After cleaning, add new text in Canva, Figma, or another editor. For dense banners, choose the "Banner" scenario and remove the whole block, then touch up the edges.
Subtitles cover a face. How do I avoid ruining the portrait?
Zoom in, select "Subtitles," and paint only over the letters with a thinner brush. Several precise passes along the line are better than one wide stripe across the whole face. If the subtitle sits on a complex background, split the line into segments.
Which image formats are supported?
PNG, JPEG, WebP, and other formats the browser can open are accepted as input (usually BMP and GIF too). The result is always saved as PNG to avoid re-compressing with JPEG after retouching and to preserve the edges after inpainting without extra compression artifacts.
Where is my image processed, and is it stored?
You draw the mask locally in the browser. Inpainting (the LaMa neural network or classic OpenCV) runs on the filltools.com server only to process your request. The file is not published in a gallery and is not intended to be shown to other users after the session ends. Do not upload unnecessary personal data.
How should I draw the mask for a cleaner result?
Cover the target completely, with a small margin around the edge (1 to 3 pixels), especially if there is a semi-transparent halo. Several precise passes are better than one huge mask. Always use zoom for small details; "Hand" (H) pans the frame, and the B/E keys switch between brush and eraser. Letter anti-aliasing often leaves a light halo, so cover it with the mask.
How is LaMa different from blurring or an eraser?
Blurring and pixelation only hide the area with a gray or pixelated patch. LaMa (Large Mask Inpainting) analyzes neighboring pixels and synthesizes a plausible background (sky, wall, fabric, paper) so the edge looks more natural. For small edits, the fast OpenCV Telea or Navier-Stokes method in the method list is sometimes enough.
Can I undo a mistaken step?
Yes. The "Undo" and "Redo" buttons keep a history of the mask and results (roughly up to 20 steps in the current browser session). "Reset mask" clears only the red markings on the canvas; "Reset" fully returns the tool to the state with no image loaded.

Reviews (0)

Similar tools

Other tools from this category you may need next.

{ }

Pixelation or AI Erasure of Sensitive Data

Prepare photos for publication: mask license plates, documents, screens, and signs. Choose coarse p…

0
{ }

Old Photo Stain Removal (AI LaMa)

Remove water marks, mold spots, and dark halos from paper prints. Paint over the stain and the LaMa…

0
{ }

Screenshot Sensitive Area Cleanup (AI LaMa)

Paint over emails, phone numbers, chats, or logins in UI screenshots and get natural interface fill…

0

Blog

Articles about tools, development, and practical tips.

All articles