For the book project I’m working on, I am adding images to illustrate a concept or important point I’m making. For example, I may add an image like the first one below to warn of the danger of those masquerading as supportive of the mission while undermining it simultaneously. I either have the right to the images, or they are public domain images.
My quandary is, what obligations do I have if I use an AI-generated image? With careful prompting and comparing different AIs, one can develop decent images for the purposes at hand. Below is an example.
Is it ethical to use AI-generated images in a book? Do I cite that they were AI-generated?
Legally I believe the situation is that they’re all public domain images, but you and/or the publisher are on the hook for inadvertent infringement or at least might be subject to inquiries.
I don’t know if it’s legally useful to credit your generator to make clear there’s no copyright, but it seems ethical to note that.
Ethics of not hiring an illustrator, designer or photographer doesn’t have one right answer now. You might ask if it’s serving your overall creative work and message, or just a financial decision.
Artistically, it takes a lot of work with prompts to make a coherent series of images, let alone ones that exhibit artistic perspective/voice. Also, keep in mind that tastes are evolving rapidly, just like how we quickly became averse to stock images. If you’re not publishing very soon, you’ll probably want to redo your images during the production step.
First step to answer that would be to look at the T&C’s on the AI engine you used. Some entity (person or company) probably owns the copyright. If not, perhaps they say so.
This is the current legal state in the US. They can’t be copyrighted. But you can violate the license of a generator (e.g. Adobe’s can’t be used to generate images used commercially, even though the images generated are public domain.)
(Personally I think we need more nuanced law here–but anyway.)
If you have an experienced established publisher, they will be able to answer your questions and then stand behind their position. You may not like their answer. This is a new area of the law that will likely be wrangled in the courts. Do you feel strongly enough to go to court to defend your use?
As linked above, AI-generated content cannot be copyrighted. But, do note that there is a trend that AI-generated content needs to be disclosed, starting with the AI services first, but local and regional regulations may eventually go further than that. I’d say in the book that the images were generated using AI, and what models were used. AI services will also need to disclose if their models were trained using copyrighted material under the new EU AI legislation.
As long as training data is not disclosed, you can be sure a good portion of it is indeed copyrighted material. I have seen plenty of AI generated images that has the watermark of commercial stock photo companies across them. The AI has seen this pattern so often, it believes it is something that is supposed to be a part of images.
Also, a lot of generic text prompts for “cartoon mouse”, “superhero” or “alien robot” will swiftly produce a copy of Mickey, a Marvel character or a Transformer.
Adobe Firefly claims to have only used imagery from their stock library, but it is still unclear if the photographers have been consulted and approved AI usage.
Considering that the most recent batch of images that just went into public domain is from the same year as Steamboat Willie, one can confidently assume that most of the training data has been created at a much closer date.
Thanks for the helpful responses. I’ve read through the responses and will determine how I need to deal with this to ensure integrity and that I maintain uncompromising ethical standards. Thanks again!
