After a large collection of artwork was scraped and used to train an AI system without the artists' consent, I started wondering whether an art-sharing site could automatically apply a barely visible noise pattern or other protective filter to every uploaded image. The goal would be to keep the artwork looking normal to people while making it harder for image-scraping and training systems to interpret. It obviously would not be perfect, especially as newer models learn to bypass older defenses, but could this still be a useful feature? Have any existing tools or platforms already implemented something like this?
5 Answers
Tools such as Glaze and Nightshade already use related ideas. They can interfere with some forms of model training, but they are not universal shields, and future systems may be trained to remove or ignore the effects. Making the protection optional is probably important because some artists dislike any visible change to their work.
Some people see AI tools as useful for creating personalized work, but that does not resolve the consent issue. Artists can reasonably support creative software while still objecting to their portfolios being copied into training datasets without permission or compensation.
These defenses are best viewed as temporary friction rather than a permanent solution. A determined scraper with enough resources can often collect and process images anyway, and a new model may learn to work around a filter. Still, making unauthorized collection slower and more expensive can be worthwhile, especially when combined with clear consent policies and technical access controls.
There are also browser-based approaches that avoid exposing certain content directly in the page, which can block some automated extraction methods. They are still imperfect and may create accessibility or usability problems. In particular, visual effects and rapid rendering changes should be tested carefully because they could bother users who are sensitive to flashing or motion.
Automatic processing could make scraping more difficult, but it cannot guarantee protection. If someone can view an image, they can usually capture or reconstruct it somehow. Access controls, rate limits, blocking suspicious services, and requiring accounts can raise the cost of scraping, though those measures also reduce how easily people can discover the artwork.
That is the main tradeoff: keeping everything behind a login offers better control, but it also makes the site less useful for artists who want their work seen.