|
Through the aid of other organizations, an available option to artists is protection from a third party. An example of this would be through the site ArtStation (an art portfolio site), which offers a “NoAI” tag, placing a HTML tag within the work’s metadata that prevents data scraping tools from using it. On the surface, this seemingly could prevent all unauthorized use of art in AI, but there has not been any recorded proof of this having any effect. To add on, ArtStation still allows the distribution of AI generated images on their site, propagating the use of datasets stolen from countless artists. This feature is also automatically disabled, leaving artists to figure out its existence.
There is also software made specially for the purpose of making it impossible to train certain images and “poisoning” AI datasets. Glaze and Nightshade by UChicago are both free-to-use resources which allow artists to protect their works through adversarial attacks. It makes the work virtually indistinguishable to the human eye, yet is able to confuse the AI into not recognizing the subject. This is achieved by identifying and isolating elements of a work specific to its style, then cloaking it. As most text-to-image models are based around style mimicry, this prevents the process. Additionally, Nightshade is a program which uses prompt-specific poisoning to hinder an AI model’s ability to generate material for certain prompts. Despite this, the process is not streamlined for artists who are possibly unfamiliar with this area of technology, making it relatively unpopular
0 Comments
|
RSS Feed