MarkGone, published at removewatermarks.com, is a browser-based suite that erases watermarks, logos, text and subtitles from images and video using AI background reconstruction, with a small daily free allowance and credit-based paid tiers.
Used this tool? Rate it
Used this tool? Rate it
MarkGone is a browser-based toolkit that removes watermarks, logos, on-screen text and subtitles from images and videos. You upload a file, the AI identifies the unwanted element, and it reconstructs what should sit behind it. There is nothing to install and no manual masking to draw.
The naming needs to be sorted out first, because the site presents itself under two names. The web address is removewatermarks.com and much of the interface refers to the service as "RemoveWatermarks", but the operating brand is MarkGone. The homepage title reads "Remove Watermarks from Photos & Videos Online Free — MarkGone", the footer carries "Copyright © 2026 MarkGone.", the support address is "support@markgone.ai" on its own dedicated domain, and the Terms of Use state the relationship plainly: RemoveWatermarks is "a service provided by MarkGone". In short, removewatermarks.com is a keyword-shaped product domain; MarkGone is the brand behind it.
This distinction matters more than usual here, because watermark removal is one of the most crowded and most imitated categories on the web. Products with near-identical names — watermarkremover.io, AnyErase, Dewatermark.ai, Unwatermark AI and dozens more — compete for the same searches. The only reliable way to know which product you are reading about is the domain.
Watermark removal carries a legal dimension that most software categories do not, and the honest thing to do is put it near the top rather than bury it.
MarkGone's own Terms of Use draw the line explicitly: "You may only use MarkGone on images that you own or have the legal rights to use. This tool is strictly prohibited from being used to remove watermarks from images protected by third-party copyrights without authorization. By using this service, you confirm that you hold the necessary rights and agree to take full responsibility for how the tool is used."
Read that carefully, because it defines the intended scope of the product and it places responsibility on you. The legitimate use case is your own material: your photographs carrying your own studio mark, a client video where you hold the rights, an old family photo with a scanning artefact, a licensed stock image where your licence permits it. Stripping a photographer's watermark from an image you found online is exactly what the terms prohibit, and the tool working technically does not make it permitted.
The flagship function. The official FAQ describes the workflow with no ceremony: upload an image, click Remove, and the AI cleans the watermark while preserving the original quality — "No Photoshop and no manual masking required".
That last clause is the actual product claim. Removing a watermark in a conventional editor means selecting it precisely, then using clone-stamp or content-aware fill and correcting the result by hand. The pitch here is that detection and reconstruction both happen automatically.
Video is a parallel product line rather than a bolted-on extra, which is unusual in this category and is the more technically demanding half. The official description is that the tool erases subtitles, captions or on-screen text from an uploaded video, after which "the AI reconstructs the background intelligently".
Video is harder than still images for a reason worth understanding: a watermark sits over moving content, so the fill has to be consistent across frames. Fill each frame independently and you get flicker — a shimmering patch where the removal happened. Whether any given tool handles this well is something you should judge on your own footage.
Rather than one generic tool, the site ships targeted variants: a TikTok watermark remover, a Sora watermark remover, a video subtitles remover, a logo remover, an AI text remover and a photo stamp remover. The Sora tool claims to handle both Sora and Sora 2 output, with the system that "automatically detects watermark patterns and cleans them frame by frame".
Targeting is a sensible engineering choice — a TikTok watermark has a known shape and moves in a known pattern, so a specialised detector can outperform a general one. But see the limitations section on the Sora case specifically, because that one raises a question the site's own terms do not address.
For video, results can be checked before committing to the full file. The site states you can "preview the generated results before processing the full video".
On a credit meter this is a genuinely valuable feature rather than a convenience. Video is billed per second, so discovering that a result is unusable after paying for a ten-minute clip is an expensive way to learn.
Beyond removal, the same account covers AI Photo Enhancer, AI Video Enhancer, Unblur Image and Old Photo Restoration. These are listed in the site's own navigation as first-class tools and have their own published credit costs, including resolution tiers for video upscaling.
The pairing is logical. Removal and restoration are the same class of problem — inferring plausible pixels that are missing or obscured — and old photo restoration in particular is a use case with none of the copyright complications, since the material is usually your own.
The supported formats are published plainly: "MP4, MOV, M4V, JPG, JPEG, PNG, BMP, WEBP, TIFF, and HEIC". HEIC matters for iPhone users, and TIFF for anyone working from scans. Note that the video list is short — no AVI, MKV or WebM — so you may need to convert first.
The core legitimate case. Photographers, designers and studios routinely hold master files that carry a preview watermark, and rights-holders sometimes need a clean version for a new context. The tool's own terms scope it to exactly this.
Scanned prints often carry date stamps burnt in by the camera, lab markings or physical damage. The Photo Stamp Remover and Old Photo Restoration tools address this, and it is the use case with the fewest complications — the photos are yours.
Soft subtitles can be switched off; burnt-in subtitles are pixels. If you hold rights to footage that has hardcoded subtitles in the wrong language, removing them is otherwise a frame-by-frame job. This is the case where automation saves the most labour.
Downloading your own TikTok post reattaches a platform watermark with your handle. Republishing that to another platform looks unprofessional and advertises a competitor. Removing a watermark from a video you made yourself is straightforward — with the caveat that each platform's terms govern what you may do with downloads.
The AI Text Remover and Logo Remover are not limited to watermarks. A stray sign, a date stamp or unwanted text in an otherwise good photograph is the same technical operation, and the tool is usable as a general object eraser.
The enhancement tools stand on their own for low-resolution archive video or slightly soft photos, independent of any removal task.
Content owners cleaning their own files. The intended audience and the one the terms are written for — photographers, video editors and studios working with material they hold rights to.
People restoring personal photographs. Date stamps, lab marks and scanning artefacts on family photos. No rights complications and a genuinely useful outcome.
Social media managers republishing their own posts. Cross-posting your own material without carrying another platform's branding, subject to each platform's terms.
Casual users with an occasional one-off job. Three free operations a day is enough for someone who needs this a few times a year and does not want to learn a photo editor.
Editors who need burnt-in subtitles gone. A specific, laborious problem that automation genuinely helps with, assuming you hold the rights to the footage.
Who it does not suit. Anyone hoping to strip watermarks from images they do not own — the terms prohibit it and the responsibility falls on the user. Professionals with exacting output standards will likely still finish by hand, so this is a time-saver in their pipeline rather than a replacement for it. High-volume operations should note there is no documented batch or API access. And anyone handling confidential material should weigh the fact that files are uploaded to a third-party service whose corporate entity is not disclosed.
MarkGone runs entirely in the browser. The site is explicit that "No software installation required" — you upload a file and the processing happens server-side. There is no desktop application, no mobile app and no documented public API.
The interface is offered in nine languages: English, Portuguese, Spanish, Japanese, Traditional Chinese, Arabic, "Bahasa Indonesia", German and French. Note that this list does not include Simplified Chinese, Korean or Russian, so speakers of those languages will be using an interface in a second language even though this page is available to them.
An account is required for paid use, with Log in and Sign up in the header and credits attached to the account. The free daily allowance is the entry point before registration becomes necessary for anything sustained.
Pricing is published, though not in the page's static HTML — the tiers are loaded at runtime, so the figures below were read from the fully rendered pricing page.
Free. Three generations and three downloads per day, with no credit balance. Enough to evaluate the tool; not enough for a project.
PRO subscription, billed annually, in four credit tiers:
| Credits per year | Price | Effective unit price |
|---|---|---|
| 1,200 | US$29.99 | US$0.025 / credit |
| 2,400 | US$49.99 | US$0.020 / credit |
| 6,000 | US$89.99 | US$0.015 / credit |
| 12,000 | US$119.99 | US$0.010 / credit |
The subscription is marked "Auto-renews yearly, cancel anytime", and a monthly option is offered alongside the annual one.
Pay As You Go, a one-time purchase, also in four tiers: 300 credits for US$11.99, 600 for US$21.99, 1,200 for US$35.99 and 2,400 for US$59.99 — from US$0.040 down to US$0.025 per credit, marked "Pay once, credits never expire".
What credits buy. Photo watermark removal costs 2 credits on the standard model and 3 on the advanced one. Video watermark removal is "1 credit per second". Video enhancement is charged per second by resolution, from 1 credit per second for HD on the standard model up to 5 per second for 4K on the advanced model, with a five-second minimum charge.
The structural trade-off is clear: subscription credits are cheaper per unit but expire with the subscription, while one-time credits cost more and never expire. Occasional users should generally prefer the latter despite the headline price.
Trial and refunds. Plans that include a trial give a "3-day trial period" with limited credits, and convert to paid automatically unless cancelled during it. On refunds, the billing policy says purchases are generally non-refundable outside specific circumstances, and the FAQ is blunter: "We do not offer refunds on pay-as-you-go plans". Buy credits in the quantity you actually intend to use.
Legal scope is the first limitation, not the last. The terms restrict use to material you own or are licensed to use, and prohibit removing third-party copyright watermarks without authorisation. Third-party legal commentary on this category notes that under US copyright law, removing copyright management information with intent to facilitate infringement carries statutory penalties — this is general background on the category rather than any finding about this product, but it explains why the terms are written the way they are. If you do not hold rights to the material, the fact that the software will process it is not permission.
The Sora tool sits in an unaddressed grey area. A Sora watermark is an AI-provenance marker rather than a third-party copyright mark, so removing it primarily engages the generating platform's terms of service and content-provenance norms rather than the copyright scenario MarkGone's own terms describe. The site ships a dedicated tool for this and its terms do not address the distinction. Users should check the originating platform's rules themselves; this page does not offer a legal conclusion either way.
The Terms of Use describe a different product. The "About Our Service" section lists capabilities including "AI-powered image editing and outfit swapping", customisable templates and drag-and-drop design functionality — none of which match the watermark removal and enhancement tools actually shipped. This looks like unedited boilerplate. It is a documentation defect rather than a functional one, but legal documents that do not describe the actual service are a reasonable thing to note.
The operating entity is not disclosed. No registered company name, no address and no specific governing jurisdiction appear anywhere in the terms, privacy policy or billing policy — the governing law clause says only that disputes are handled where MarkGone operates, without saying where that is. The sole contact channel is an email address. That is a lower level of transparency than well-documented competitors provide, and worth weighing before uploading anything sensitive or committing to an annual plan.
No independent review coverage exists. Searches of Trustpilot, G2 and Product Hunt found no listing for either MarkGone or removewatermarks.com, and no independent editorial coverage from established technology publications was located. A third-party AI directory lists the product with a rating, but directory scores of this kind come without disclosed sample sizes or methodology, so no rating is cited as evidence anywhere on this page. In practice this means you should rely on your own testing via the free tier rather than on reputation.
Output quality varies, and the vendor says so. The FAQ states that "The results from generative AI features may vary and are not always fully controllable", directing users with flawed output to support. This is an honest disclosure and matches how these models behave: results depend heavily on what sits behind the watermark.
Hard file limits. Images up to 20MB; video up to 300MB and 10 minutes. High-resolution photography and long-form video fall outside these bounds. The supported video container list is also narrow.
Uploads are processed remotely and user content is collected. The privacy policy explicitly includes "including images, text prompts, and generated outputs" within the data it gathers. It states that personal data is not sold, names its processors (Google Analytics, Stripe, PayPal, AWS, Google Cloud, Zendesk), and gives retention periods — accounts until deletion or inactivity, "Billing records: kept up to 7 years", system logs up to twelve months. One question it does not answer: whether uploaded content is used to train models. The policy makes no statement on that point either way, so it remains unknown from official sources.
Refunds are restrictive. Pay-as-you-go purchases are non-refundable, and subscription refunds are limited to enumerated circumstances. Combined with automatic renewal, this rewards paying attention to the billing calendar.
Both, at different levels. MarkGone is the brand and operating entity; RemoveWatermarks is how the service is presented on its site at removewatermarks.com. The Terms of Use state that RemoveWatermarks is a service provided by MarkGone, the footer copyright reads MarkGone, and official support runs through an address on the markgone.ai domain. Given how many similarly named watermark tools exist, treat the domain as the definitive identifier.
It depends entirely on what you process, and the tool's own terms are explicit: you may only use it on images you own or have the legal rights to use, and using it to remove third-party copyright watermarks without authorisation is prohibited. The terms also state that you take full responsibility for how you use it. Removing your own studio mark, cleaning an old family photo or working on client footage you hold rights to is squarely within scope. Taking someone else's watermarked photo from the internet and stripping the mark is not, regardless of whether the software can do it technically.
There is a free tier, but it is small: three generations and three downloads per day, with no credits granted. Beyond that you need either a PRO subscription (annual tiers from US$29.99 for 1,200 credits up to US$119.99 for 12,000) or a one-time credit purchase (from US$11.99 for 300 credits up to US$59.99 for 2,400). The "free" in the site's own title refers to that daily allowance, not to unlimited use.
It depends on how regularly you will use it. Subscription credits are substantially cheaper per unit — as low as US$0.010 versus US$0.040 at the entry one-time tier — but they are only valid during the subscription period and the plan renews automatically. One-time credits cost more but never expire. If you need this a few times a year, pay-as-you-go usually works out better despite the higher unit price; regular, predictable volume favours the subscription.
MP4, MOV, M4V for video and JPG, JPEG, PNG, BMP, WEBP, TIFF, HEIC for images. Images can be up to 20MB; video up to 300MB and no longer than 10 minutes. Common containers such as AVI, MKV and WebM are not listed, so conversion may be needed first.
Video watermark removal is billed at one credit per second, with a five-second minimum charge. A two-minute clip is therefore about 120 credits — roughly US$1.20 to US$4.80 depending on your tier. Video enhancement costs more and scales with resolution, up to five credits per second for 4K on the advanced model. Preview your result before processing a full-length file.
It depends on what is behind the watermark. Marks over flat, even areas usually come out convincingly; marks over faces, text, detailed textures or repeating patterns are where reconstruction has to invent detail and where artefacts show. The vendor itself acknowledges that generative results may vary and are not always fully controllable. Use the free daily allowance on your hardest file before paying, and on video always check the removal area frame by frame for flicker.
They are processed on remote servers, and the privacy policy explicitly counts uploaded images and generated outputs as data it collects. The policy states that personal data is not sold, names its infrastructure and payment processors, and publishes retention periods including up to seven years for billing records and twelve months for system logs. It does not state whether uploaded content is used to train models — that question is simply unanswered in the official documentation, so this page makes no claim about it either way. Avoid uploading confidential or sensitive material.
Largely no. The FAQ states plainly that refunds are not offered on pay-as-you-go plans, and the billing policy restricts subscription refunds to specific situations such as unresolvable technical faults, accidental purchases or duplicate charges. Because the three-day trial converts to a paid subscription automatically unless cancelled, and because credit purchases are effectively final, it is worth testing on the free tier first and buying only what you need.
The brand is MarkGone, but the corporate entity behind it is not disclosed: no company name, address or specific governing jurisdiction appears in any of the legal documents, and email is the only contact channel. No listing was found on Trustpilot, G2 or Product Hunt, and no independent coverage from established technology publications was located. There is a third-party AI directory entry that corroborates the brand-to-domain relationship, but its rating comes without a disclosed sample size and is not treated as evidence of quality here. The practical implication is to evaluate it yourself on the free tier rather than relying on reputation, and to weigh the limited transparency before committing to an annual plan.
