A browser-based AI watermark remover that uses inpainting and GAN models to erase watermarks, logos, timestamps and signatures from photos, videos and PDF documents, with a manual brush for touch-ups and batch processing for up to 50 images.
Used this tool? Rate it
Used this tool? Rate it
Dewatermark is a browser-based AI watermark remover whose single purpose is removing overlays from media you already have. You upload a photo, a video or a PDF, the system locates the watermark layer, and it rebuilds the pixels underneath so that the result reads as an unmarked original rather than a blurred patch. According to the site, the product uses inpainting and GAN to detect and erase watermarks from images without blurring the background, which is the technical distinction it draws against tools that simply smudge or clone over the offending area.
The scope is broader than the name suggests. Beyond classic corner logos, the site advertises removal of text overlays, diagonal repeating patterns, full-screen tiled marks, timestamps burned in by cameras, transparent low-opacity marks, and photographer signatures. Three file families are handled through three separate front-ends: images at the main address, video at a dedicated page, and PDF documents at another. Each has its own limits, its own pricing arithmetic, and its own quirks.
Two facts should shape your expectations before you go further. First, this is a credit-metered service rather than a flat subscription, and the arithmetic differs by media type and by model. Second, and more importantly, watermark removal sits on a legal boundary that the site itself acknowledges in writing on every tool page. Both topics are covered in detail below, and the second one deserves reading before you upload anything.
Transparency about the operator is thinner than for most tools in this category. The website offers no reachable terms of service, privacy policy, or about page — the common paths all return errors — so its data-handling promises exist only as marketing copy on the homepage rather than as binding published terms. On the Android side there is a disclosure: on Google Play the developer of record is an individual, Phan Thị Hàn An, with a personal Gmail address and a residential address in Vinh, Nghệ An province, Vietnam. Third-party aggregator pages assert a Singapore company origin, but that claim could not be verified against any first-party source and is not repeated here.
Three automatic removal models with stated boundaries: Rather than one generic button, the service exposes tiers matched to difficulty. The site describes Auto Remove 2.0 for emojis and simple overlays, Auto Remove 3.0 for most cases, and PRO Remove for multi-layered or near-invisible watermarks at opacity below 0.05. A newer Auto Remove 4.0 is presented as the most accurate option. Being told which model suits which problem is genuinely useful, because picking the heaviest model for a trivial overlay wastes credits at triple the rate.
Manual AI brush for what automation misses: The most honest feature on the page is the acknowledgement that automation fails. When artifacts remain, the manual brush applies inpainting locally to each painted area, and the site notes it is effective for object removal, signatures, partial overlays, and irregular shapes. This makes it a two-stage tool rather than a one-click gamble: run the model, then repair what it left behind.
Batch processing up to fifty images: For volume work, batch mode processes up to 50 images simultaneously with a single removal setting applied across the set, returning everything as one ZIP. The site positions this at e-commerce product photography and media libraries, which is the workload where per-image manual editing stops being viable.
PDF handling that preserves document structure: The PDF watermark remover takes a different technical route from most competitors. Rather than rasterising pages, it preserves the original PDF layout, fonts, and resolution after removal, so the output remains a real document rather than a stack of images. A first-page preview is free, letting you check font clarity and layout integrity before spending credits on the whole file.
Video processing with a preview gate: To remove logo from video footage the tool works frame by frame, and because video is billed by duration the workflow inserts a checkpoint: you request a short preview, judge the result, and only then commit credits to the full render. Uploads are capped — the video page states max length is 60m.
Broad input format support: On the image side the tool accepts JPG, PNG, WEBP, and AVIF up to 10MB or 6000×6000px, with clipboard paste via Ctrl+V. AVIF support is unusual in this category and useful for images pulled from modern web sources.
Upscaling and high-resolution output on paid tiers: Paid plans add 4X image upscale and output up to 24 megapixels, positioned for cases where the source is too small to use after cleaning.
API and MCP integration: For programmatic use, the site advertises a Watermark Remover API and an MCP connector, with credits usable across both the web app and the API. Integration enquiries go through email or Telegram rather than a self-serve developer portal.
Cleaning up your own archive: The clearest legitimate case is media you created. Old exports carrying your own studio mark, proofs you produced, or client deliverables you hold the rights to can be stripped of proof marks without re-rendering from source files you may no longer have.
Removing burned-in camera timestamps: Cameras, dashcams and older phone apps stamp date-time text directly into the frame. Because this text is not metadata but pixels, cropping is normally the only fix, and cropping loses composition. Targeted removal preserves the full frame.
Erasing accidental objects from personal photos: The manual brush is not watermark-specific. A stranger at the edge of a holiday photo, a bin in an otherwise clean streetscape, or a reflection can be painted out using the same inpainting mechanism.
Preparing product photography at volume: E-commerce sellers who receive supplier images carrying supplier marks — and who have the supplier's permission to use them unmarked — can run a catalogue through batch mode rather than editing images one at a time.
Cleaning documents you own: Draft markings on your own contracts, reports and internal documents can be removed before final distribution. Note carefully that this use case has a boundary discussed at length in the Limitations section, and that the boundary depends entirely on whether the document is yours to alter.
Recovering usable resolution from small sources: The paid upscaling path targets images that are clean but too small, though user reports on output quality are mixed and are covered below.
Model choice is where credits are won or lost. A single opaque corner logo on a photograph is a job for the standard automatic model. A semi-transparent mark tiled across the whole frame, or a mark that sits under other image content, is what the PRO tier exists for. Running PRO on everything triples your cost for no gain on easy files; running the basic model on hard files burns a credit and returns something you then have to fix by hand anyway.
Pricing is credit-metered, and the conversion rates matter more than the headline numbers. The site sets them out directly: 1 credit = 1 image (3.0/4.0 model) | 3 credit = 1 image (Pro Model) | 1 credit = 1s video. Because video bills per second, media type — not plan tier — is the dominant cost factor. A single minute of footage consumes sixty credits, roughly the equivalent of sixty standard image jobs.
The free tier grants 3 free credits per day and never expires, covering image, video and text removal plus the manual editor. In practice three credits per day means three standard images, or three seconds of video, which makes the free tier a genuine evaluation path for photo work and effectively unusable for video.
Paid access comes in two shapes with an important difference between them. The Pro Pack starts at $10 per month billed yearly, bundling 1.2K to 12K credits with 24-megapixel output, 4X upscaling, batch processing, full video processing and no ads. It advertises no auto-renewal, which sounds consumer-friendly, but the same line states that unused credits will expire when your PRO membership ends — a detail the pricing FAQ repeats. The pay-as-you-go option inverts this: prices ladder from $7 for 100 credits down to $800 for 100K credits, and the site states Pay only once. Credit available forever. If your usage is irregular, the one-off packs avoid the expiry trap even though the per-credit price at small volumes is higher. An enterprise tier is quote-based, arranged by email.
The site places rights responsibility on you, in writing, on every tool. Directly beneath the upload area on the image, video and PDF pages, the same sentence appears: Use this tool only with images you own, created yourself, or are authorized to edit. You are responsible for ensuring watermark removal complies with copyright and other applicable laws. Its FAQ goes further and takes a position on legality: removing watermarks from images you own or have permission to edit is described as permitted for personal and commercial use, while Removing watermarks from copyright-protected images without permission violates copyright law. Read that boundary before you upload, because the tool itself enforces nothing — it processes whatever it is given.
Some of the site's own demonstration use cases sit uneasily against that boundary. This is worth stating plainly, using the site's own words. On the PDF page, the marketing copy explains that Text watermarks like 'DRAFT', 'CONFIDENTIAL', 'SAMPLE', 'DO NOT COPY' are often layered in a PDF file to protect the business documents, contracts, and report drafts, and presents removing them as a supported use. On the video page, logo removal is framed as producing a full frame video without third-party branding visible for reposting and embedding, with CapCut, InShot and social platform exports named as sources. The homepage lists photographer and artist signatures among removable watermark types, and a displayed testimonial describes erasing TikTok platform watermarks to repurpose content for other channels. The site does not state whether removing a third party's confidentiality marking, platform mark or artist signature falls inside the "authorized to edit" condition it sets elsewhere — it makes no comment on that question. These are quoted here so you can weigh them yourself; this page does not offer a legal conclusion, and you should seek qualified advice for your own situation.
No published terms, privacy policy or company information. Requests for the usual terms-of-service, privacy-policy and about paths all return errors. For a tool that receives user files and makes explicit data-handling promises, the absence of binding published terms means those promises rest on homepage marketing copy alone, with no document to hold the operator to.
Operator disclosure is minimal and inconsistent with third-party accounts. The only concrete operator information found first-hand is the Google Play listing, where the developer of record is an individual, Phan Thị Hàn An, with a personal Gmail address and a residential address in Vinh, Nghệ An province, Vietnam. Aggregator sites assert a Singapore corporate origin; that assertion is unverified and is not adopted here. For business use involving confidential material, this level of disclosure is worth weighing.
Real output quality is more mixed than the marketing suggests. The site promises removal without quality loss, but Google Play reviewers report otherwise. One states the tool leaves the space where the water marks used to be looking funky; another reports that it reduces the quality and resolution alot. With 127 reviews the sample is small, so treat these as directional rather than statistical — but they align with the known weakness of inpainting on smooth gradients.
Credit expiry differs by plan and is easy to miss. Subscription credits expire when the membership ends, while pay-as-you-go credits do not. Buying a large subscription bundle for occasional use is the expensive mistake this structure invites.
Video economics are unforgiving. At one credit per second, video work escalates quickly and the free daily allowance covers three seconds. The tool is best understood as an image tool that also does video, not as a video tool.
No independent media review exists. Searches for coverage in established technology publications returned only competitor-authored comparison pages, aggregator listings and the vendor's own blog. No named-journalist review with an independent testing methodology was found, so there is no third-party benchmark of the removal quality claims.
The site's own rating is not third-party data. The homepage displays 4.9 out of 5 based on 179 ratings, but this figure is self-published and self-hosted, not drawn from an independent review platform. There is no Trustpilot profile for the domain. The verifiable independent number is the Google Play score of 4.2 across 127 reviews.
Traffic and reach figures come from estimates. A third-party directory lists roughly 1.8 million monthly visitors, which is an external estimate rather than an audited figure, and it records the tool as catalogued in January 2024.
Hard input limits apply. Images cap at 10MB or 6000×6000px, PDFs at 50MB, video at 60 minutes. Large or high-resolution source material may need downsampling first, which itself costs quality.
The site takes an explicit position on this. It states that using the tool is legal, that removing watermarks from images you own or have permission to edit is permitted for personal and commercial use, and that Removing watermarks from copyright-protected images without permission violates copyright law. It also instructs, on every tool page, that you should use the tool only with files you own, created yourself, or are authorized to edit, and that you carry responsibility for compliance. Whether a specific file falls inside those conditions depends on facts this page cannot assess, and nothing here constitutes legal advice — consult a qualified professional for your own case.
There is a free tier of 3 free credits per day, described as free forever, covering image and video watermark removal, text removal and the manual editor. Basic image use requires no login. Since one credit equals one standard image or one second of video, the free allowance is a workable trial for photos and negligible for video.
Credits are consumed per job, and the rate depends on model and media: 1 credit = 1 image (3.0/4.0 model) | 3 credit = 1 image (Pro Model) | 1 credit = 1s video. The same balance works for the web app and the API. Two purchase shapes exist: subscription packs from $10 per month billed yearly, and one-off credit packs. The difference that matters is expiry — subscription credits lapse when membership ends, while one-off packs state Pay only once. Credit available forever.
Images are accepted as JPG, PNG, WEBP, and AVIF up to 10MB or 6000×6000px. PDFs are accepted up to 50MB. Video accepts MP4 and MOV with a stated maximum length of 60 minutes. Clipboard paste works for images.
The site says no, claiming reconstruction preserves original resolution, texture and colour. Independent user reports are less uniform: reviewers on Google Play describe visible residue where the watermark sat and noticeable resolution loss. Expect good results on textured backgrounds and weaker results on smooth gradients, and always inspect output at full size rather than trusting a thumbnail.
All three, through separate tools. Video removal works frame by frame with a five-second preview before you spend credits. The PDF tool is notable for preserving the original PDF layout, fonts, and resolution after removal instead of converting pages to images, and it offers a free first-page preview.
Batch mode processes up to 50 images simultaneously with one removal setting applied to the whole set, delivered as a single ZIP. Batch processing is a paid-tier feature. Group your files by watermark type before batching, since a single setting cannot suit different mark styles.
The site states it does not store, share, or publish uploaded images, that files are deleted from servers immediately after download, and that transfers use HTTPS. The Android listing's data safety section declares no data shared with third parties and no data collected. These assurances should be read alongside the fact that no published privacy policy or terms of service could be found on the site, so there is no binding document behind them.
An Android app exists on Google Play with 4.2 stars across 127 reviews and more than 10,000 downloads, carrying ads and in-app purchases. The site also states an iOS app is available, though a direct search of Apple's public catalogue did not surface a matching listing. Desktop software for Windows and macOS is offered as well. The browser version needs no installation and works on mobile browsers.
For maximum quality and control, Photoshop's generative removal tools; for a free manual route, Photopea; for simple image-only erasing, single-purpose cleanup tools. For documents you own, editing the source or using a real PDF editor is more reliable than image-model reconstruction. And where the watermark belongs to someone else, re-exporting from your own project file or requesting a licensed copy from the rights holder addresses the actual problem rather than its appearance.
