Not legal advice. This page summarises publicly available statutes and published policy documents for general information. It is not legal advice, it creates no lawyer-client relationship, and it cannot account for your jurisdiction, your licence terms, or your specific facts. For a commercial decision, talk to a qualified lawyer where you are.
1. The question people actually mean
“Is watermark removal legal?” collapses two questions that have different answers. The first is whether removing the mark breachescopyright law. The second is whether it breaches theterms of the service that produced the file. You can pass the first and fail the second. Most of the confusion around this topic comes from answering only one of them.
2. What copyright law says
The U.S. Copyright Office describes the DMCA as making three changes to copyright law, two of which are relevant here. The Copyright Office on the DMCA explains that section 1201 “prohibits circumventing technological protection measures (or TPMs) used by copyright owners to control access to their works”, and that section 1202 “makes it unlawful to provide or distribute false copyright management information (CMI) with the intent to induce or conceal infringement”.
17 U.S.C. 1201 — anti-circumvention
The statutory text defines the prohibited act precisely. Section 1201(a)(3)(A) says to “circumvent a technological measure” means “to descramble a scrambled work, to decrypt an encrypted work, or otherwise to avoid, bypass, remove, deactivate, or impair a technological measure, without the authority of the copyright owner”. Section 1201(b)(2)(A) uses parallel language for measures that “effectively protect[] a right of a copyright owner”.
Two features of that text matter for this topic. First, the definition reaches removal and impairment, not only decryption. Second, section 1201(a)(2) and (b)(1) prohibit trafficking in technology that is “primarily designed or produced for the purpose of circumventing” a measure, or which “has only limited commercially significant purpose or use other than” circumvention. A tool whose primary purpose is removing a mark is squarely within the language a court would examine.
One limit is worth stating, because it is frequently overstated. Section 1201(c)(1) provides that “nothing in this section shall affect rights, remedies, limitations, or defenses to copyright infringement, including fair use”. Section 1201 is not an infringement statute. It is a separate anti-circumvention regime, and it can be breached without any infringement occurring at all.
17 U.S.C. 1202 — copyright management information
This is the provision most directly aimed at the question, and it is the one most watermark discussions skip. The Copyright Office describes CMI as “certain information, including the title, name of the author and copyright owner, and terms for use of the work”, and states that section 1202 makes it unlawful “to provide or distribute false copyright management information […] or to remove or alter that type of information in certain circumstances”.
The unresolved question is whether a platform’s AI-provenance mark — a visible overlay, an invisible SynthID watermark, or a C2PA Content Credentials manifest — constitutes CMI as that term is defined. Content Credentials are metadata by their nature, which puts them on firmer ground than a decorative overlay. Whether a purely decorative visible logo qualifies is not something a U.S. court has squarely answered. Treat this area as unsettled, not as either safe or fatal.
3. What Google’s terms say
This is where the risk becomes concrete, and it is a different kind of risk from copyright law: it does not depend on who owns the underlying image. A platform can restrict marker removal as a matter of contract even where the law would permit it.
The Terms of Service
The current Google Terms of Service (effective July 30, 2026) contain several relevant clauses. Under “Don’t abuse our services” they state that you must not abuse, harm, interfere with, or disrupt Google’s services “by […] spamming, hacking, or bypassing our systems or protective measures”. They also prohibit misleading others into thinking that generative AI content was created by a human, and “providing services that appear to originate from us when they do not”.
The most directly on-point clause is in “Google content”: “Don’t remove, obscure, or alter any of our branding, logos, or legal notices.” Read literally, that covers removing a visible Google AI overlay from Google’s own branding.
The same section is also why this site carries a visible disclaimer that it is independent of Google. Google, Gemini, and Veo are trademarks of Google LLC. We do not claim affiliation, and we do not use their marks in our own branding.
The Generative AI Prohibited Use Policy
The Generative AI Prohibited Use Policy (last modified December 17, 2024) is incorporated by reference into the Terms of Service. Three clauses bear on this:
- Prohibited Use 1.6 prohibits content that “violates the rights of others, including privacy and intellectual property rights”.
- Prohibited Use 2.3 prohibits circumvention of abuse protections or safety filters.
- Prohibited Use 4.5 prohibits misrepresenting the provenance of generated content by claiming it was created solely by a human.
One caution about secondary sources. Forum answers and community posts frequently quote Google as prohibiting “circumvention of safety and transparency markers”. That exact phrase does not appear in the policy text. The real clauses are 2.3 and 4.5 above. If you are going to rely on what Google’s policy says, read the policy rather than a summary of it — including summaries that agree with each other.
A related document is worth avoiding as a citation. The Generative AI Additional Terms of Service are archived: the page states they no longer apply as of May 22, 2024, except for business partners with a signed agreement. Citing them as current policy would be an error.
4. The four cases, ranked by risk
Clearly unlawful
Stripping the watermark from an unlicensed stock photo. Stock licences prohibit it in terms, and removing the mark is the mechanism by which infringement occurs. Do not do this.
Generally unlawful
Removing attribution or copyright notices your licence requires you to keep. This is the clearest § 1202 case, because CMI is defined to include exactly that.
Unsettled
Removing a visible AI-provenance overlay from your own generated file. You may well hold the rights to the image itself; Google states it “won’t claim ownership” over content you generate. But you received the file under terms that tell you not to remove Google’s branding, and those are contractual obligations you accepted. Copyright law may not be engaged at all, and you can still be in breach.
Out of scope here
Invisible provenance signals. See can you remove a SynthID watermark. This tool does not touch them and this page gives no guidance on how to.
5. What this tool does and does not do
- Runs entirely in your browser. Files are decoded and processed on your device and never transmitted, so we cannot inspect, log, or moderate them, and we hold no copy of anything.
- Cannot verify ownership. Because nothing is uploaded, there is nothing to check against. You are responsible for lawful use.
- Grants no rights. As our terms of service state, it cannot give you a right to modify a file you did not have the right to modify.
- Does not remove SynthID or C2PA. These are treated as out of scope, deliberately.
A tool that never receives a file cannot itself be the party that uploads infringing material. That is a meaningful distinction for intermediary liability, and it is the main reason this site is built the way it is. It is not, however, a defence to a contractual claim against the person operating the tool, or to a claim against the person using it.
6. Where this leaves you
If you are working on media you created, in a jurisdiction you understand, the legal question is usually the less binding one. The platform terms are the constraint you are more likely to hit, and they are the one to read before you rely on any tool including this one.
If you are working on third-party material, stop and get advice first. If you have a commercial or compliance question, that is a lawyer question, not a documentation question, and no page on the internet including this one is a substitute.
7. Sources
- The Digital Millennium Copyright Act — U.S. Copyright Office. Overview of sections 512, 1201 and 1202.
- 17 U.S. Code § 1201 — Circumvention of copyright protection systems. Full statutory text, including the definitions in (a)(3) and (b)(2).
- Google Terms of Service — effective July 30, 2026. See “Don’t abuse our services” and “Google content”.
- Generative AI Prohibited Use Policy — last modified December 17, 2024. Referenced by the Terms of Service.
- SynthID — Google DeepMind. Description of the invisible watermark and its robustness to modification.
Policy documents change. The Terms of Service above are effective July 30, 2026 and the Prohibited Use Policy last modified December 17, 2024; re-read the linked pages rather than trusting this summary, which may fall out of date.