Midjourney & Watermarks

The ethics of generated art is a bit…complicated. We know generative AI tools have been trained on both commercial and non-commercial images. Artists have filed lawsuits asking that their work and names be removed from the training data. People have been trying to figure out ways to identify where the images come from (with more or less success). And agencies are starting to get even more strict about using generated images as work. Which I fully agree with because as I always say:

AI generated images can be great for placeholder images, for mood or storyboards but should never be used as final art.

It’s a starting place, not a solution.

This week, I encountered bits of watermarks or what looks to be copyright/artist name artifacts on Midjourney generations and wanted to document a few below. It’s odd, because I can’t decide if the AI adds a watermark/copyright because it has seen so many and thinks it’s valuable, or if it is just that something in parts of these generations are exact copies from imagery (of dubious origin) it was trained on, so it pulls that bit out as well. I’m guessing the latter, but I prefer the former.

I personally never use an artist’s name as a style prompt, because I feel like that’s an area I’m not comfortable going into. But I realized I have no problem knowing Midjourney is trained on stock, which shines a spotlight on my own personal bias and is something I really need to re-examine. I’ve also been operating under the assumption that it isn’t copying any of this directly, but rather categorizing images and using them as a basis to compile or “draw” an entirely new one.

From a 2022 interview in Forbes with CEO and founder David Holz:

Did you seek consent from living artists or work still under copyright?

No. There isn’t really a way to get a hundred million images and know where they’re coming from. It would be cool if images had metadata embedded in them about the copyright owner or something. But that's not a thing; there's not a registry. There’s no way to find a picture on the Internet, and then automatically trace it to an owner and then have any way of doing anything to authenticate it.

🤔

I mean, I’m not a data scientist, but i feel like there are at least some ways to get a hundred million images and know where they are coming from. Like having an AI categorize and create a registry as they crawl the web and at least document where they come from as you train it? Just spitballing here.

So behold: A few of the images that I have generated that have what look to be remnants of watermarks or copyrights.