Free · no registration · unlimited uses
Image Watermark Remover
Paint over the mark. Local AI fills the background; compare before you download.
Drop or select multiple images, or paste a screenshot.
JPG / PNG / WebP / AVIF / HEIC / BMP · up to 20 MiB, 8 megapixels and 8192 px per side

A clean image starts here
Brush or draw a rectangle over a watermark, old text or date, then choose AI repair. Try our own product illustration first if you like.
The first AI repair downloads about 52 MB of model and runtime files; the browser can reuse its cache afterwards. Images stay on your device. Smart selection also downloads a detector of about 4.8 MB.
Up to 50 images and 120 MiB per batch. Reuse the current selection or repair detected candidates; check every result. Switching images keeps results but clears that image’s undo history.
Smart selection finds candidate text, light overlays and edge badges. It may also select real lettering or patterns. Preview and adjust before repair.
For old text, dates and small watermarks in images you own or may edit. AI fills plausible background, not necessarily the real hidden content. Check product lettering, people and complex textures carefully.
How to use it
- Choose an image or open the example. The model is not downloaded just by opening the page.
- Brush or frame the mark, including a little around its edge. Erase mistaken selections and split large areas into smaller repairs.
- Choose AI repair and inspect with the comparison slider. Undo if needed, then download the full-size PNG.
Frequently asked questions
Are images uploaded to a server or Gemini?
Repair never uploads your image or calls Gemini. Only when you continue editing or upscaling is a result temporarily stored in this browser; the receiving tool deletes it after reading. Expired transfers are cleared on the next transfer. Download results before leaving. Site ads do not process images.
Can it restore the image perfectly without loss?
No guarantee. AI estimates the hidden area. Only selected pixels receive the repair; unselected pixels and original dimensions are preserved in the PNG. Original alpha is kept, not reconstructed. Patterns, faces and product details may be wrong. JPEG/WebP: lossy re-encoding; transparency is flattened onto white for JPEG.
Does it support batches or automatic watermark detection?
Up to 50 images and 120 MiB per batch. Reuse the current selection or repair detected candidates; check every result. Switching images keeps results but clears that image’s undo history. Smart selection finds candidate text, light overlays and edge badges. It may also select real lettering or patterns. Preview and adjust before repair.
Can I use it on a phone, and why is the first run slower?
Touch painting is supported; switch to Pan view after zooming. The first run downloads about 52 MB and initializes the local model. Speed depends on network and device. Use small areas and cancel at any time; there is no cloud-upload fallback.
Is it free without signing up? Is full-size export paid?
Repair, batch processing, ZIP and original-size export are free, with no account, credits or added watermark. Each input is limited to 20 MiB, 8 megapixels and 8192 pixels per side for browser processing; payment does not unlock larger files.
Can it remove repeated, multiple or translucent watermarks?
Brush or select several areas, or start with smart selection and correct it. Distant selections are processed in context tiles. If the mask covers more than 35% of the image, work in smaller passes. Dense repeats and large occlusions are harder and may remain visible.
Can it also remove text, date stamps, logos or small objects?
Manually mask these areas and use the same repair action to generate a background. Smart selection suggests text, light overlays and edge badges; it is not a universal object detector. Keep the mask close to the target and inspect the result at a larger zoom.
Can I enhance, upscale or edit the repaired image?
Yes. Continue to the site’s enhancer/upscaler or editor with the current result, then choose the next action. No image upload is required. The upscaler has its own device and output-size limits; enlargement cannot recover the true hidden detail.