Nightshade
A free tool that lets artists subtly alter their digital images so that AI models trained on them produce corrupted or incorrect results, discouraging unauthorized use in AI training.
Explanation
Nightshade was developed by University of Chicago professor Ben Zhao and his team of PhD students, building on their earlier tool Glaze, which disrupts an AI's ability to mimic a specific artist's style. Nightshade goes further by poisoning the training data itself, making pixel-level changes invisible to the human eye that cause image-generating models such as Stable Diffusion, Midjourney, or DALL-E to misclassify the image, for example rendering a dog as a cat once a model has trained on enough poisoned samples. The goal is to give artists more negotiating power against companies scraping their work without permission or compensation. Nightshade signals a growing wave of technical resistance to unauthorized AI training on copyrighted creative work, a trend marketers in content, brand, and creative licensing are tracking closely.
Usage context / examples
Artists have applied Nightshade to portfolio images posted online as a way to discourage AI companies from using that work to train new image-generation models without consent.
Related terms
Glaze, AI Training Data, Data Poisoning, Copyright and AI
Turkish definition
Sanatçıların dijital eserlerini, yapay zeka modelleri bu görsellerle eğitildiğinde hatalı sonuçlar üretecek şekilde ince biçimde değiştirmelerini sağlayan, izinsiz kullanımı caydırmayı amaçlayan bir araçtır.