Adobe's 'Ethical' Firefly AI Was Trained on Midjourney Images

More problematic than this is that there are many images on Adobe Stock that were stolen and put up for sale by parties entirely unrelated to the artists who produced them. Worse, they don’t make it easy to report image theft and generally aren’t responsive when reached out to about these issues.

It’s not happened to me personally but I’ve seen artists I’m a fan of fall victim to this.

So any claim that Firefly was ethically trained is questionable at best and mostly just feel-good marketing.

The implications of this in the article are mainly focused on the ethics, but at least where the copyright is concerned there is no problem with this -- Midjourney Images are not copyrightable, and it doesn't seem like there's any grounds for Midjourney to complain; especially if the images were uploaded by 3rd-party users and not sourced directly by Adobe under a Midjourney license.

From the article:

> Training on AI-generated content probably wouldn’t make Adobe’s Firefly image generator less commercially safe, and the company isn’t required to say what it’s training on as long as it isn’t misleading consumers, said Harvard professor Rebecca Tushnet, who focuses on copyright and advertising law. But training on AI images, such as those created by Midjourney, undermines the idea that Firefly is unique from competing services, she said.

The critique here isn't that Adobe is violating Midjourney's copyright, it's that Firefly's data ultimately is not as ethically sourced as Adobe claims (keeping in mind that different commenters probably have different qualifications of what ethically sourced training data would be).

One consequence of Midjourney's Discord based interface is anyone can scrape thousands of prompts and image generations.

> Generative AI images from the Adobe Stock collection are a small part of the Firefly training dataset,” wrote Adobe

Interesting I wonder if this add a strong incentive for competitors, or anyone really, to sneak in data like that and then sue later.

It can be a bit like patent trolling but in this case it would be training data trolling.