View all on-demand sessions from the Intelligent Security Summit here.
Spawning, an organization that started building tools for artist ownership over their training data in September, announced that yesterday Stability AI honors requests from artists to opt out of Stable Diffusion 3 training, which is expected to begin in a few weeks. Artists can request their opt-outs via am trained.com.
According to Berlin-based Mat Dryhurst, who Founded spawn in September with his wife, musician Holly Herndon, the organization has been chatting for months with Stability AI and LAION, a non-profit open source dataset that is part of Stability.
“They responded immediately,” Dryhurst told VentureBeat. “Both organizations have been transparent about data from the start, which I believe may be why they are subject to disproportionate scrutiny. We approach this issue as one that needs to be solved, and they were supportive and full of ideas.”
Text-to-image AI has raised questions about who owns images
Ever since DALL-E 2 has been released in April, the creative industry was abuzz with questions about the ownership of AI art images. In August, Bradford Newman, who leads the machine learning (ML) and artificial intelligence (AI) practice of global law firm Baker McKenzie, said in his Palo Alto office the answer to the question “Who owns DALL-E images ?” is far from clear. And, he stressed, legal ramifications are inevitable.
When the open-source Stable Diffusion was released in August, there were even more questions about the model’s training. And just a few days ago, a new ones study (not yet peer-reviewed) was released, raising new concerns. It identified instances where image-generating models, including Stable Diffusion, had been copied from public Internet data — including copyrighted images — on which they had been trained.
Stability AI responds to art asset issues
Emad Mostaque, founder and CEO of Stability AI, pointed out on Twitter that Spawning would also be offering opt in requests – for artists who want their images included in the training data.
“Technically, these are tags for LAION and coordinated around it,” he said tweeted. “It’s actually quite difficult because of the size (what if your image is on a news site, for example?) Exploring other mechanisms for attribution etc., welcome constructive input.”
Mostaque also seemed keen to emphasize that Stability AI is not taking this step for expected legal or ethical reasons. “We don’t think there’s a legal reason for that, but we think different datasets from models will be interesting and would like to see output differences,” he said. tweeted. “We think most people will choose richer experiences over time, just as we’ve seen them use artstation and others.”
Could this set a precedent for AI art?
But whether or not Mostaque thinks the measures are necessary (or whether other artists think it goes far enough), Dryhurst insists it’s a “great opportunity to set a precedent for the advancement of AI art “.
It makes sense for people to register their wishes with Spawning once, he explained (the organization has claimed to be an independent organization that does nothing with the data), so that the information can then be served to various organizations – so artists don’t have to no mole to play.
“It’s certainly a chaotic challenge, and we have no false confidence that it will be possible to enforce things in a complete way, as technically anyone can scrape the web,” Mostaque admitted. “We just feel that most interactions will be with a few large organization models, and we don’t see why those organizations wouldn’t honor requests that we can provide them. In my opinion, it does them a favor so they can focus on science.
VentureBeat’s mission is to become a digital city plaza where tech decision makers can learn about transformative business technology and execute transactions. Discover our Briefings.