Lessons from DeviantArt’s AI Debacle
Last week, DeviantArt announced the launch of its new artificial intelligence (AI) art creation tool, DreamUp. However, that launch was not greeted the way that they had hoped.
Rather than excitement, DeviantArt’s DreamUp was greeted with user backlash. Many users took to Twitter, in addition to forums on DeviantArt itself, to express their frustration and to condemn the new tool.
Specifically, users were upset that that their artwork would be used to train the new system and that, if they wanted to opt out, their options were limited.
That’s because, while DeviantArt provided a button to opt out of inclusion in AI, it had to be enabled per-work and was opt-in by default. If users wanted to opt out their entire account, they had to fill out a web form that required a human review and, according to reports, up to 10 days to take effect.
However, more recent reporting has cast some nuance on this, both in good and bad ways for DeviantArt. According to a post on The Verge, DeviantArt CEO Moti Levy says that the site isn’t doing any DeveintArt-specific training for DreamUp and that the tool is Stable Diffusion, an image AI tool that already has a controversial history for scraping artwork without permission for the purpose of training.
According to DeviantArt, the opt-out toggle simply adds a pair of flags to the pages, “noai” and/or “noimageai”, that is designed to block third parties from using the image in their training. The aforementioned opt-out form is meant to prevent DreamUp from creating works based upon their names or the titles of specific works.
However, those efforts may be too little too late. Stable Diffusion was already trained on what is being described as “unethically sourced data”, likely including DeviantArt content, and there’s no way for the system to unlearn that.
In short, users are likely attempting to opt out of a system that their work is already a permanent part of. To that end, DeviantArt changing all works to be opt-out by default may not carry much meaning.
To that end, the controversy (and confusion) are not poised to end. DeviantArt has made it very clear that it intends to move forward with its AI projects, despite user objections. Meanwhile, artists, including DevinatArt members, are still very wary and unsure of how such technology users their images.
The Good DeviantArt Was Trying to Do
On the surface, DeviantArt was trying to do several good things.
The company, feeling that AI is more or less inevitable at this stage, wanted to try and find a way to make it more fair to artists. To that end, they wanted to do three things when launching DreamUp.
- Opt Out Tags: First, DeviantArt developed a series of meta tags that it hopes will become standards to let other AI scrapers know what content they can and cannot access. These are similar to the tags that search engines use to know what pages to index or not index.
- A More Limited AI: According to DeviantArt, the main difference between DreamUp and vanilla Stable Diffusion is that Dream Up places limits on the queries it will respond to and will block queries that involve specific names or work titles to prevent the AI from too closely mimicking the work of any one artist.
- Flagging AI Art: Also, all works created by DreamUp will be automatically flagged as AI generated, making it easier to identify it is uploaded elsewhere or, as in one case with a different AI, submitted to a contest.
These steps are ostensibly good, if weak. The tags, for example, are only useful if other AI tools adhere to them. Going back to the search tags, those are only useful because all major search engines follow them. So far, no AI company has agreed to adhere to those tags.
Likewise, the limitations placed on DreamUp don’t mean that users won’t be able to game it to mimic the works of other creators. It may make it more difficult, but almost certainly not impossible. Finally, the automated flags are only useful if they can’t be removed by humans, which they almost certainly can.
These are clearly good steps and, if DeviantArt had just introduced these changes, they likely would be lauded for at least doing something to help their users. However, these changes cannot be taken in a vacuum.
Instead of announcing these changes by themselves, they announced them at the same time they were launching their own AI project, one that partners with an already-controversial service. Couple that with the toggle being set to opt-in by default, and it’s easy to see why creators quickly assumed the worst.
Lessons from DeviantArt’s Mistakes
AI has a horrible reputation with artists right now. AI projects, by in large, have been very cavalier about their use of works created by human artists to train their work.
That, in turn, has created a copyright gray area that has caused many services, including Getty Images, to ban AI-generated content from their services. As we saw with the dispute over GitHub’s Copilot program, there are many ethical and legal challenges to sort out in this space, but AI companies have simply continued to race ahead, hoping that it is better to seek forgiveness rather than ask permission.
This has left a bad taste in the mouths of many artists, who have felt powerless to stop the use of their work.
And that points to the problem DeviantArt has. While many of the steps that they are taking should be seen as giving artists more control over their work when it comes to AI, they made the announcement at the same time that they launched their own AI art tool.
DeviantArt has been repeatedly forced to clarify their stance and modify their actions, for two reason:
- Buried Message: DeviantArt horribly buried their message under promotional material they were launching their own AI service. Nearly all the marketing focuses on using the new service rather than what it means for existing artists. Those artists, rightly, feel as if they are viewed as unimportant by DeviantArt.
- Confusing Messaging: DeviantArt, at least on their first few attempts, failed to answer a simple question: When and how is DreamUp trained on DeviantArt images? In fact, that question is still unclear. While the answer to this question is complicated, and it’s likely DeviantArt doesn’t actually know since it’s using Stable Diffusion, the lack of clarity creates serious problems.
A smarter move for DeviantArt would have been to introduce the tags and the opt-out tool in one move (providing the opt-out list as a tool for other AI makers). That would have given them time to explain those tools, hone them, let users set them to their liking and make the point that they are trying to give artists more control.
Then, after that was done, they could have introduced the AI tool. While any AI tool released by DeviantArt was going to be met with controversy, this would at least have helped clarify the role of those tools and how DreamUp will use works but their own members.
That, in turn, is the lesson for others in this space. Artists are, for good reason, wary of AI image generation systems. If you are going to launch one, it is crucial to communicate clearly whose work it will be trained upon, how artists can opt out and what steps you’re taking to ensure that output works do not infringe or unethically copy the work of human artists.
DeviantArt failed at that task. Their message was, and still is, very confusing. Furthermore, releasing it at the same time as their new AI system only harbored mistrust and confusion. That’s made even worse by partnering with an AI system that is already deeply mistrusted by artists.
Regardless of DeviantArt’s intentions, the response they were met with is both understandable and deserved.
Bottom Line
I actually have a strange history of defending DeviantArt from panics that have surrounded the site. The most prominent one was the “they’re selling your images” one from 2017, where users were afraid that the site’s terms of service made it possible for their artwork to appear on t-shirts for sale without their explicit permission.
However, this is a time DeviantArt created their own mess. While they may be right that AI art is inevitable and ignoring it is not a winning strategy, it still is a space they should enter into with care and caution. That’s not what happened.
DeviantArt partnered with an already-controversial AI system to launch an AI product on their service. Though they launched it alongside tools to help users protect their works, those tools were not properly highlighted and were poorly explained, especially as they related to their own tool.
DeviantArt wants to be known as “the only platform giving creators the ability to tell third-party AI datasets and models whether or not their content can be used for training.” However, if that’s the case, that’s the message they should have led with. Instead, they focused on trying to get people to use DreamUp.
Hopefully others entering this space will take some notes and lessons from this debacle. Artists are understandably wary of AI generators and if you want them to not turn on your project, it’s important to consider their wants and wishes, communicate with them clearly and effectively, and move slowly in the space.
Sadly, that’s something that is very unlikely to happen given the rapid pace in this field.
Want to Reuse or Republish this Content?
If you want to feature this article in your site, classroom or elsewhere, just let us know! We usually grant permission within 24 hours.