Nightshade

Nightshade is a free research tool from the University of Chicago that protects images from unauthorized AI training. It adds tiny, nearly invisible changes to a picture before you share it. If an AI image model trains on the altered picture without permission, it learns the wrong thing — for example, it may link the word dog to cat pictures. Artists and photographers run it on their work before posting online. Download it for Windows, Mac, or Linux, and check the official site for the latest guidance.

Key Features

1. Data poisoning for images

Nightshade turns a normal image into a poison sample that corrupts the training signal for one chosen concept in AI image models.

2. Nearly invisible changes

It uses multi-step math to keep visible changes tiny, so the human eye sees a picture that looks almost the same.

3. Custom poison tags

You pick the concept to poison — for example, tagging a dragon picture so a model learns it as a mountain or cloud.

4. Adjustable strength

Low, medium, and high intensity settings let you trade stronger protection against slightly more visible changes.

5. Common formats and platforms

Process JPEG and PNG files on Windows, Mac, or Linux. Results are best with images at least 512 pixels wide.

6. Pairs with Glaze

Nightshade pairs with Glaze, the team's other tool, which defends against AI copying an artist's style instead.

Frequently Asked Questions (FAQ)

How does Nightshade protect my images?

It adds invisible pixel changes to your picture. When an AI model trains on the altered image, it learns the wrong link for that concept.

Is Nightshade the same as Glaze?

No. Glaze defends against AI copying your art style. Nightshade is an offense tool that distorts what the model learns from the image.

Will people notice the changes in my pictures?

Barely. The tool is built to minimize visible changes, so the picture looks almost unchanged to human eyes.

How many poisoned images does it take to work?

In the team's tests with image-generation models, fewer than 100 poisoned samples could affect one targeted concept.

Does Nightshade guarantee my art will not be misused?

No tool can promise that. It raises the cost and difficulty of training on unlicensed data instead.

Which AI models does it target?

It was designed for text-to-image diffusion models, the kind behind popular AI image generators.

Conclusion

Nightshade gives artists a practical way to push back against unauthorized AI training. It is open, well documented, and built by university researchers. It cannot promise perfect protection, but it makes scraping your work far less useful. If you share art online, it is worth adding to your routine. It takes only minutes to protect a full portfolio.