A Clarifying Moment

Earlier this week, there was a bit of a debate in one of the Patreon communities I follow. A popular YouTuber had put a short section of AI-generated artwork in a video to discuss an old legal case, a section that they might have struggled to explain in another way.

The images introduced a number of factual and visual continuity errors, but the big thing was that it was obvious where they came from. And some of the creator’s fans pushed back, though some seemed indifferent to the whole mess.

Ultimately, though, the video was delayed from release by a couple of days, with new imagery to tell the story. The result added a new visual style to the channel’s arsenal, and it’s all because the creator was pushed to not take the easy way out.

In recent months, there has been a lot of discussion about the role of artificial intelligence in creative works. I have in the past written with interest on the issue from both sides of the debate—talking to graphic artists about it, comparing the cultural moment to similar debates on sampling and stock photography. And I saw cool artists that I appreciate, like Rob Sheridan, using these tools in interesting ways. I honestly took a somewhat objective stance because it wasn’t clear how all this stuff was going to play out culturally.

But a lot has changed in the past couple of years. The tenor against AI creative work, particularly art, is uglier. People get madder about it when they see it. It is not just frustration; it is fury. I have seen people who have been willing to not use services just because they use LLM-generated images on their blog.

There is something strange about what we’re seeing now that feels different, and I haven’t quite been able to put my finger on it. Fortunately, we now have a clarifying moment to help bring it all home.

That clarifying moment comes in the form of OpenAI’s new GPT-4o image generation models, which have gone viral in recent days for their ability to convincingly mimic the works of Studio Ghibli.

These images have allowed for famous photos and iconic memes to appear in a style generally associated with anime, and it’s sort of disconcerting in a way I might compare to the dead-eye fake human creator This Person Does Not Exist. It was not long ago that the low quality of the image creation was almost charming in a way. We are now well past those days.

The thing that is clarifying about this move is not that this is possible but the ethical red line it crosses, in that it specifically disregards the feelings of the animator whose style is getting nicked. Hayao Miyazaki, who largely animates using traditional methods at a time when everyone else uses computers, has spoken out against AI art in a way that makes the images almost seem offensive in full context.

At Aftermath, Luke Plunkett put it in a far more wide-eyed way than I think I could have:

The power of AI imagery isn’t in its capabilities, it’s in its attempts to become ubiquitous. It’s (likely) illegal and immoral to be making this stuff, but like so much else in our 21st-century world, laws are developing much slower than technology. Companies like OpenAI are hoping that the longer their tech can stay out there, the more it becomes part of the background noise of the modern internet, and the more likely it is that they themselves will become part of the fabric of the modern internet, and not a bunch of raiders stripping the place for its creative wiring.

In this context, if you think of OpenAI as “Uber, but for creative work,” it starts to feel a lot more skeezy. Like quick access to a vehicle by pressing a button on your phone, this stuff does have value—I will tap the sign of my “bionic arm theory” any day of the week. And those use cases are out there. One that comes to mind is Goblin.tools, which uses AI to help make big tasks more manageable for people with ADHD. If most AI was in this spirit, I think fewer people would have a problem with it.

But that is very clearly not what OpenAI is selling here. You may have an “ethical way” of using large language models, but OpenAI does not. If it wasn’t clear by the company’s prior use of Scarlett Johansson’s voice even after she said no, the company would rather take from creative people without asking than do things the right way. And that sets the tone for others.

And that causes critics to rage, when an approach that was even 10% less aggressive might tamp down some of their concerns.

Yesterday, there was a user in my feed who shared an image of Miyazaki based on a still frame from the above video where he actively denounced AI art. Yes, it was in what I guess we can now call the “OpenAI house style.” I felt awkward when I saw it, but it just kept biting at me all day, making me think about the fact that this user was essentially making an ugly political statement with this action. Who cares what this beloved animator thinks? People are choosing to stomp all over him by using his likeness in a way he specifically would not condone.

(For what it’s worth, I blocked the guy.)

These images intentionally spoil any potential goodwill this company has earned in the two and a half years since it unleashed ChatGPT onto the world. LLMs have value, but OpenAI feels beyond the point of saving.