Researchers develop tool that lets artists poison AI training algorithms

[Jan]

Data "poisoning" tool lets artists damage AI image generators that try to train on their works

Researchers have developed a tool to inject digital artworks with invisible data that damages AI image generators when they are trained off the work, MIT Technology Review reports.

Nightshade, the work of a team led by Professor Ben Zhao at the University of Chicago, can confuse image generators like Dall-E, Midjourney, and Stable Diffusion to make them churn out nonsense results.

 

[MagNorthDigital]

This is absolutely fantastic. Long overdue. Not cause I dislike midjourney, dalle or SD. I just like artists that create amazing works of art more.

I personally know a few guys that put years into their work, literally going for broke to publish their own ideas and concepts, that had to move to the states to do it. And ai has impacted their ability to put bread on the table.

It's not a case of get better, it's a case of made redundant cause we can clone your style and vision.

 

[Swa]

Good. The sooner "AI" dies the better. It's already a failure as it needs new ideas or it just becomes dumber.

 

[Petec]

This is absolutely fantastic. Long overdue. Not cause I dislike midjourney, dalle or SD. I just like artists that create amazing works of art more.

I personally know a few guys that put years into their work, literally going for broke to publish their own ideas and concepts, that had to move to the states to do it. And ai has impacted their ability to put bread on the table.

It's not a case of get better, it's a case of made redundant cause we can clone your style and vision.

100000% this!!!

 

[Yskasmetnstoof]

Great news.