How to protect your art, music and video from AI
You cannot stop a copy. You can make it impossible for anyone else to claim the original. This page explains the difference, compares every approach honestly, and shows the one step that protects your authorship permanently, for free.
Start here
AI theft is three different problems, and they need different answers
When artists say AI is stealing their work, they usually mean one of three things. Each has a real defence, and no single tool covers all three. Knowing which problem you have is most of the answer.
| The harm | What actually happens | What defends against it |
|---|---|---|
| Training scraping | Your images, tracks or text are collected and used to train a model without permission. | Opt-outs (robots.txt, ai.txt, registries) and cloaking tools (Glaze, Nightshade). |
| Style mimicry | A model, or a person using one, produces work that imitates your style. | Cloaking makes your style harder to learn. Proof of authorship makes your originals distinguishable from the imitation. |
| Impersonation and claim theft | Someone posts your actual work as theirs, or a copy circulates with your name stripped. | Proof of authorship. A permanent, verifiable record of who made it and when, that survives copying. |
The distinction that matters
Every approach, honestly
Glaze, Nightshade, opt-outs, C2PA, watermarks and TIP, compared
These are not competitors. They act at different points, and the strongest position is to use more than one. What follows is what each actually does, and where each stops.
| Approach | Acts on | What it does | Where it stops |
|---|---|---|---|
| Glaze (University of Chicago) | The copy you post | Adds pixel-level perturbations so models misread your style. | Only images. Does nothing after a file is already in a dataset. Does not prove you made anything. |
| Nightshade (University of Chicago) | The copy you post | Poisons training data so models that ingest it learn wrong associations. | Only images. Adversarial, and models adapt. Does not prove authorship. |
| robots.txt, ai.txt, opt-out registries | The scraper | Asks crawlers and companies not to train on your work. | A request, not an enforcement. Ignored by anyone who chooses to ignore it. Covers only sites you control or platforms that honour it. |
| C2PA content credentials | Metadata inside the file | Embeds a signed provenance manifest in the file. | Stripped by screenshots, re-encoding and most social uploads. Gone the moment the file is transformed. |
| Visible or invisible watermarks | Pixels of the copy | Marks the file so copies can be identified. | Cropped, painted out, or defeated by re-generation. Proves a mark was present, not who owns the work. |
| Trust Identity Protocol® (TIP) | The original, and your identity | Records who made it, when, and a content fingerprint, on a federated ledger anyone can check for free. Survives copying, cropping, re-encoding and metadata stripping. | Does not stop copying or training. Is evidence, not prevention. Supports a copyright claim; does not replace one. |
Read the last column carefully, including ours. A tool that claims to stop AI outright is overclaiming. The honest position is layered: cloak the copy you post, opt out where you can, and register the original you keep.
How it works
What registering your work with TIP actually gives you
The Trust Identity Protocol is an open standard for proving who made a piece of content. When you register a work, four things are created and none of them can be taken from you.
- A verified identity, once. A TIP-ID is issued to a real human after biometric verification by an accredited Verification Provider. Your raw biometric data is never stored anywhere. One human, one TIP-ID, portable across every platform.
- A Content Trust ID for each work. Every registered piece gets a permanent CTID, a signed record of what it is, who made it and when. Signatures use ML-DSA-65, a post-quantum standard, so the record stays valid against future computers, not just today's.
- A fingerprint that survives copying. The record includes a perceptual fingerprint (CNA-2.2) that recognises the work across re-encoding, cropping, screenshots, format conversion and derivative edits. A copy with your name stripped still resolves to your original.
- A public record nobody can erase. The registration lives on a federated DAG (TIP Network) run by independent node operators in multiple jurisdictions. No company, including The AI Lab, can remove or alter it. Anyone can verify it, free, with no account, forever.
What this means in practice
By medium
How to protect your art and illustrations from AI
- Register the original at full resolution with TIP before you post it anywhere. The registration date is your evidence, so earlier is stronger.
- If you want to discourage style learning, run the copy you post through Glaze. Keep the unglazed original registered. Cloak the copy, register the original.
- Add
robots.txtandai.txtopt-outs on your own site, and list your work in the Have I Been Trained registry for platforms you do not control. - Display the Protected by proof badge on your portfolio so a visitor, a buyer or a platform can verify your authorship in one click.
What TIP does for a visual artist that nothing else does
A commission, a print, a screenshot on a mood board, a “tribute” that is really a trace. Every one of those is a copy, and every one still resolves to your registered original. You do not have to find the copies. Whoever encounters one can check it.
By medium
How to protect your music, songs and albums from AI
- Register each track, and the stems if you can, before release. A pre-release registration date is the strongest evidence of authorship a musician can hold.
- Register the album as a collection as well, so the whole and the parts each carry a Content Trust ID.
- Use your distributor's AI-training opt-out where one exists, and publish an
ai.txton your own site. - Keep the registration record with your split sheets and publishing paperwork. It is the one document in that folder that a stranger can verify without trusting you.
Why this matters more for audio
Audio is lifted by re-encoding, clipping and pitch-shifting, all of which destroy embedded metadata. TIP's fingerprint is built to survive exactly those transformations, so a lifted hook or a lookalike vocal traces back to the original you registered.
By medium
How to protect your video and film from AI
- Register the finished cut before upload. If you can, register the raw footage too, so a re-edit of your material still resolves to you.
- Registration before upload means a deepfake, a re-cut or a clip extraction made from your footage carries a trace back to your verified original.
- Opt out on your own site and on any platform that offers it, and keep the registration record with your release paperwork.
Deepfakes specifically
The defence against a deepfake of you is not detecting the fake. It is a verified record of the real thing, made first, that anyone can check. If your genuine footage and your verified identity are on the record, a synthetic derivative is checkable against them. That is a stronger position than any detector.
By medium
How to protect your photographs from AI
- Register originals at full resolution, dated before any public posting.
- Post copies through Glaze if you want to discourage training. Register the original you keep, not the cloaked copy.
- Opt out on your own site with
robots.txtandai.txt. Stock and portfolio platforms strip metadata on upload; TIP's record does not depend on metadata.
By medium
How to protect your writing from AI
- Register articles, manuscripts, scripts and lyrics as you finish them. The record binds the canonical text to your verified identity.
- If you publish on WordPress, install the TIP plugin. It signs every post automatically at publish time, so you never have to remember.
- Add
ai.txtto your site, and keep the registration record with your drafts. A paraphrase or a lifted passage can be compared against a dated, signed original.
Do this today
Four steps, twenty minutes, free
- Get verified. Obtain a TIP-ID from an accredited Verification Provider. It takes a few minutes and your raw biometrics are never stored. Start here.
- Register your work. Upload or point at each piece. Each gets a Content Trust ID. Register content.
- Show it. Put the Protected by proof badge on your site so anyone can verify you in one click.
- Layer the rest. Cloak the copies you post, opt out where you can, and keep registering as you create.
Questions
Everything creators ask, answered plainly
How do I protect my art from AI?
Do three things. First, decide whether you want to stop models learning from your images, and if so run them through Glaze or Nightshade before posting. Second, opt out of training where you can: robots.txt and ai.txt on your own site, and the Have I Been Trained registry for platforms you do not control. Third, and separately from both, prove authorship: register each piece with the Trust Identity Protocol so it carries a permanent, verifiable record of who made it and when. The first two try to stop the copy. Only the third proves the original is yours after a copy exists.
Can AI steal my art?
A model can be trained on your images without permission, and a person can use a model to imitate your style or pass your work off as theirs. Those are different harms. Training scraping is addressed by opt-outs and by image cloaking tools like Glaze and Nightshade. Impersonation and false claims are addressed by proof of authorship, which is what TIP provides: a cryptographic record, bound to your verified identity, that anyone can check for free.
Does the Trust Identity Protocol stop AI from copying my work?
No, and nothing does reliably. What TIP does is different: it makes it impossible for anyone else to claim your work. Each registered piece gets a permanent Content Trust ID, signed with a post-quantum signature bound to your verified identity, recorded on a federated ledger that no single company controls. A copy can be made. The claim of origin cannot be taken.
How do I protect my music or song from AI?
Register the master and the stems with TIP before release. The record captures a content fingerprint that recognises the track across re-encoding, clipping and format changes, so a derivative or a lookalike traces back to your original. Pair that with a public opt-out on your site and your distributor's AI-training settings where they exist.
How do I protect my album from AI?
Register every track individually, then the album as a collection, so each piece and the whole carry a Content Trust ID. Do this before the album is anywhere public. A registration dated before release is the strongest evidence of authorship you can hold.
How do I protect my video from AI?
Register the finished cut and, if you can, the raw footage. TIP's content fingerprint survives re-encoding, cropping, screenshotting and clip extraction, which is how video is usually lifted. Registering before upload means a deepfake or a re-cut made from your footage carries a trace back to your verified original.
How do I protect my photos from AI?
Register originals at full resolution, and keep the registration date earlier than any public posting. If you also want to discourage training, run copies through Glaze before posting and keep the unglazed original registered. Opt out on your own site with robots.txt and ai.txt.
How do I protect my writing from AI?
Register articles, manuscripts and scripts with TIP as you finish them. For text the record binds the canonical content to your identity, so a paraphrase or a lifted passage can be compared against a dated, signed original. If you publish on WordPress, the TIP plugin signs every post automatically at publish time.
Is registering my work with TIP free?
Yes for individuals, at any income level, under the TIP Community License. Verification of any registered work is free for everyone, forever, and needs no account. Commercial tiers apply only to businesses above USD 100,000 in annual revenue.
Is TIP the same as copyright registration?
No. Copyright exists automatically when you create a work, and formal registration with a copyright office is a separate legal step in some countries. TIP is evidence: a permanent, tamper-evident, independently verifiable record of who made a work and when. It supports a copyright claim. It does not replace one.
What is the difference between Glaze, Nightshade and TIP?
Glaze and Nightshade, from the University of Chicago, alter an image's pixels so that models trained on it learn the wrong thing. They act on the copy, before it is taken. TIP acts on the original: it proves who made it and lets anyone verify that, after any number of copies exist. They solve different problems and work together. Cloak the copy you post; register the original you keep.
What does C2PA do, and how is TIP different?
C2PA attaches provenance metadata inside a file. It is useful, and it is stripped the moment a file is screenshotted, re-encoded or uploaded to most social platforms. TIP does not live inside the file. The record lives on a federated ledger, keyed to a content fingerprint, so it survives every transformation that strips metadata.
How do I prove I made something if AI copies it?
With a dated, signed, independently verifiable record made before the copy existed. That is precisely what a TIP registration is. Anyone, including a platform, a court or a journalist, can look up the Content Trust ID and confirm who signed it, when, and whether the content has changed since, without asking The AI Lab or trusting any single company.
Who runs the Trust Identity Protocol?
The Trust Identity Protocol is an open standard built and stewarded by The AI Lab Intelligence Unobscured, Inc., a Delaware corporation, and governed in the open by the AI Trust Council. The specification is published under CC-BY 4.0. The reference implementation converts to Apache 2.0 on January 1, 2031. Anyone can implement it and anyone can run a node.
Who is behind this
An open standard, not a product you rent
The Trust Identity Protocol® is built and stewarded by The AI Lab Intelligence Unobscured, Inc., a Delaware corporation, and governed in the open by the AI Trust Council™. The specification is published under CC-BY 4.0. The reference implementation is under the TIP Community License and converts to Apache 2.0 on January 1, 2031. Anyone may implement it. Anyone may run a node. Verification is free for everyone, forever.
That structure is deliberate. A creator's proof of authorship should not depend on one company staying in business, staying honest, or staying interested. It should be a public record. That is what this is. Read the whitepaper, the licence, or the protocol overview.