glaze.cs.uchicago.edu Paid content

What is Glaze?

Glaze is a free tool from the SAND Lab at the University of Chicago that helps artists protect their work from being used to train generative AI image models. It works by making small, carefully calculated changes to an image before you post it online. To a human eye the image looks essentially the same, but to an AI model the changes distort what the model "sees," making the artwork a poor training example. The project describes this as a way to give artists a measure of control over whether their style is mimicked by models trained on their work.

Who it's for and when it helps

It's aimed at illustrators, painters, photographers and other visual artists who publish images online and are concerned about their work being scraped into AI training datasets. The practical use case is straightforward: you glaze a finished piece, then post the glazed version to your portfolio or social media. If someone later collects that image for training, the protection is meant to interfere with the model's ability to learn your style from it.

Trade-offs to keep in mind

• Protection is not absolute. Glaze is designed to resist known attacks, and the project notes updates that respond to new ones — an arms race rather than a permanent shield.
• Glazing takes processing time and, for the desktop version, a capable GPU. The download page mentions support for recent NVidia 50xx GPUs alongside Windows.
• The changes are intended to be visually subtle, but you should always check the glazed output yourself before publishing, especially on detailed or high-contrast work.
• It protects images you glaze. It does not remove work already in a dataset, and it can't stop someone from training on unglazed copies.

Next step

If you want to try it, start with the official download and the user's guide, and glaze one test image first so you can judge the visual quality and processing time on your own machine. The project also offers WebGlaze for those who can't run the desktop version, and a companion tool called Nightshade. See Glaze for the current version, downloads and FAQ.

How does Glaze protect artists from generative AI?

Glaze is a tool from the SAND Lab at the University of Chicago that helps artists protect their work from being used to train generative AI models without permission. It works by making small, human-imperceptible changes to an image before it is posted online. These changes are designed to disrupt the way AI models learn from the image, so a model trained on glazed art may fail to imitate the artist's specific style.

What it does, in practice

• You run your artwork through Glaze before uploading it to a portfolio, social media or marketplace.
• The visible image looks essentially the same to people.
• The altered version is meant to be less useful as training data for style-mimicking models.

Who it is for

Glaze is aimed at illustrators, painters, photographers and other visual artists who publish work online and want a layer of protection against style scraping and unauthorized model training. It is a defensive, pre-publication step rather than a way to detect or remove existing AI copies.

Trade-offs to consider

• It is not a guarantee. Protection is an ongoing cat-and-mouse problem, and the project has released updates specifically to resist new attacks.
• It adds a processing step to your workflow, and heavily processed or resized images may weaken the effect.
• It protects the specific image you glaze, not your style as an abstract concept.

A concrete scenario

An illustrator posts a new character design to a portfolio site. Before uploading, she runs the file through Glaze. If someone later scrapes the portfolio to train a model on her style, the glazed images are intended to degrade that training. She keeps her original, unglazed files for clients and print.

Next step

If you want to try it, download the current version and read the user's guide on Glaze. Check that your operating system and GPU are supported before installing, and test the result on one image first to see how it looks at full size.

For a related tool aimed at disrupting image-to-image pipelines, the same lab also offers Nightshade, which is linked from the same site.

How do I download and install Glaze on my computer?

Glaze is downloaded from its official site and installed like a normal desktop application. The current release listed there is Glaze 2.2 for Windows, with support for newer NVidia 50xx GPUs, so the download link and version notes on the page are the first thing to check.

What to do

1. Go to Glaze and open the Download Glaze section.
2. Pick the installer for your operating system. The page currently highlights Glaze 2.2 for Windows.
3. Run the installer and follow the on-screen steps.
4. Before using it, read the User's Guide linked on the same page. Glaze's protection depends on settings such as the style you choose and the intensity, and the guide explains the trade-offs.
5. If you want release notes or fixes, check the page's update notes; it mentions that Glaze 2.1 included bugfixes and changes to resist a new attack.

A practical example

A digital illustrator finishing a piece for social media would download Glaze, install it, open the image, select a style close to their own work, and export a protected version to post. The original file stays untouched, and the protected copy is what goes online.

Decisions to keep in mind

• Windows vs. other systems: the page's main download is Windows-focused. If you use macOS or Linux, check the site for current platform support rather than assuming the Windows build applies to you.
• GPU matters: the mention of NVidia 50xx support means newer graphics cards are handled in the latest version, so updating from an older release can matter.
• Read before clicking: the User's Guide and FAQ on the same page answer common installation and usage questions, including how Glaze differs from Nightshade.

If the installer or a setting is unclear, the Feedback, Bug Reports, and Comments and Suggestions links on the page are the intended next step.

What are the differences between Glaze and Nightshade?

Glaze and Nightshade are two separate tools from the same research group at the University of Chicago, and they solve different problems. Glaze is aimed at protecting a human artist's own work from being imitated by generative models. Nightshade is aimed at poisoning how a model learns from images, so that unauthorized training produces worse results. In practice, Glaze is the defensive tool an artist runs on their portfolio; Nightshade is the more offensive, data-poisoning tool.

H3: How they differ in plain terms

H3: Which one fits your situation

If you mainly want to post finished work to a portfolio, social media or a client gallery and worry about style mimicry, Glaze is the more natural starting point. If your concern is that a specific model or dataset has already absorbed your work and you want to make future scraping less useful, Nightshade is the tool built for that purpose.

Both are available from the same project site, so you can compare their guides and downloads directly: Glaze. The site also links to WebGlaze, which is worth checking if you don't want to install desktop software, and to the Glaze user's guide for step-by-step instructions.

A practical next step: read the "What is Glaze" and Nightshade sections on the project site, then decide based on whether your priority is protecting your own style or disrupting training data. If you only have time for one, start with Glaze, because it is the lower-risk, more widely adopted option for individual artists.

How can I use WebGlaze instead of installing the desktop app?

WebGlaze is the browser-based route to the same protection Glaze provides on the desktop: you submit artwork through the web interface, it processes the image, and you get a protected file back. It exists mainly for people who can't or don't want to install the desktop app—Mac users, artists on machines without a supported GPU, or anyone working from a borrowed or locked-down computer.

How the two options differ in practice

The desktop release notes on Glaze mention specific GPU support, which is exactly the kind of requirement that pushes people toward the web version.

A realistic workflow

Say you're on a MacBook without a supported Nvidia card. You'd open WebGlaze, upload the piece you want to protect, choose your settings, and wait for the processed image to come back. For a single illustration that's fine. For a 40-piece portfolio, the upload/download cycle and queue times get tedious fast, and you're trusting a third party with unreleased work.

Decision criteria

• Choose WebGlaze if you can't install the desktop app, need it once in a while, or are on hardware the desktop version doesn't support.
• Choose the desktop app if you protect art regularly, work under NDA, or want to avoid uploading originals.
• Before either, check the current user's guide and FAQ on the site—settings and supported hardware change between versions, and the pages list the latest release details.

One caution: any web tool means your file leaves your machine. If the artwork is client work or under embargo, confirm that's acceptable before uploading.

What do artists need to know about Glaze's limitations against new attacks?

Glaze is a defensive tool that adds a protective "cloak" to artwork so AI models learn a distorted version of the style. Its main limitation is that it is an ongoing arms race, not a permanent shield: the project itself notes that version 2.1 included changes "to resist a new attack," which tells you that older cloaks can become less effective as attackers adapt. For artists, the practical takeaway is to treat Glaze as one layer of protection, not a guarantee.

Where the limits show up

• New attacks and model updates. A cloak that worked well against one training pipeline may weaken against newer or differently designed ones. Re-glazing older work with the current version is safer than assuming an old file is still protected.
• Platform processing. Uploading, resizing, re-compressing or screenshotting an image can degrade the cloak. Post the Glazed file directly where possible and keep an untouched master.
• Style versus copying. Glaze targets style mimicry. It does not stop someone from manually copying a composition, tracing, or training on images you never Glazed.
• Effort and hardware. Cloaking is compute-heavy, so batch-processing a large portfolio takes time and a suitable machine. Plan for it rather than doing it the night before a deadline.

A workable routine

1. Keep an unmodified master of every piece.
2. Glaze the version you publish, using the current release.
3. Note the Glaze version and date alongside the file so you know when to redo it.
4. Re-glaze older portfolio pieces periodically, especially after a major update.
5. Combine it with visible watermarks, low-resolution previews, and clear licensing terms.

Choosing your level of effort

Glaze's own site frames the work as a mission-driven research effort rather than a finished product, and points to a user's guide for setup details. If you are deciding whether to rely on it, start by downloading the current version and reading that guide, then test the workflow on one or two pieces before committing your whole portfolio. For background on the research and its updates, see Glaze.