I was curious if anyone has any thoughts or experience using AI generated art images in games, especially as it pertains to license and copyright? I recently learned about artbreeder (https://www.artbreeder.com/), and was super impressed with some of the images it generates!
The artbreeder TOS states:
Each Artbreeder image is owned by the user who created that image.
You agree to license any images you create on Artbreeder under the Creative Commons CC0 license.
which makes it sound like I would be allowed to use images generated on Artbreeder, either ones that I created, or other users. But I wanted to be sure, so I checked and artbreeder is based on the ganbreeder code (https://github.com/joel-simon/ganbreeder).
ganbreeder is GPL-3.0 licensed, but uses the biggan-512 deep network model (https://tfhub.dev/deepmind/biggan-512).
The biggan-512 model is Apache-2.0 licensed, but was trained on the ImageNet dataset (http://image-net.org).
The ImageNet FAQ states:
The images in their original resolutions may be subject to copyright, so we do not make them publicly available on our server.
...if you are a researcher/educator who wish to have a copy of the original images for non-commercial research and/or educational use, we may provide you access through our site, under certain conditions and at our discretion.
So what do you make of this chain of licenses / property? The biggan-512 model doesn't actually contain any of the images from ImageNet, and ganbreeder does not reproduce any of them (at least, you'd have to work to "evolve" an image close to a pre-existing image). Nevertheless, ImageNet is clear to state that they don't own copyrights on their images, and only provide the dataset for non-commercial use. So I would think that anyone using artbreeder art in a commercial project is running a risk - even though that risk may be incredible minimal and distributed over thousands of copyright holders who may have a very hard time showing that you have infringed on their copyright.
Have you heard of any legal cases involving these issues?
I believe this earlier post is relevant as well: https://opengameart.org/forumtopic/what-is-generated-with-the-help-of-th...
EDIT: I should say that I don't know the answer to this question. Whether AI can create new inspirations or if their output is entirely derivations of the works they were trained on is tricky. Do we believe an AI can be truly 'creative'? Coming up with new content inspired by previous works? Or are they just doing lots of tiny copy-&-pastes, hue shifting, pixel transitioning, and other "photoshop" operations that would still constitute derivative work based on the content they were trained with?
--Medicine Storm
Thanks! I missed that one.
ar yes, artbreeder, i too have seen this and had a play with it. it was an interesting test with some cool results and not so good ones. i can see its apeal and why you could use but, like you, was unsure of how realiable the licensing is. Without going in to deep with it my conclusion was that because it generates new images using other images that others have uploaded and We know that not everyone understands or care about using other peoples work for their own gains, so just on that alone it cant be trusted. Using images that you know are CC0, or your uploading your own work is a little different, and is most likely more secure on a license front, but not completely, because like you mention above with all other licenses that have been linked in between the software itself, we don't know what code/algorithms has been 'forked'/copied/derived from what and when, so this makes it a minefield to know what is licensable, let alone useable.
i don't know of any legal issues, but if we are asking the question and are unsure, then we cant be confident, and if we can't confident in what we are using then the best thing to do is stay well away from it, just in case, or at least until we know for sure that its all good. :)
Chasersgaming | Support | Monstropolis |