At Fluffy, we care deeply about artists. After all, we are photographers and illustrators and the project was created for my mom, who was a brilliant artist and poet. We do not consider the work with AI that we do to be “making art” and we think the term “AI artist” does not apply to us at all. There are some ethical considerations with the systems that we use to create the images and the project as a whole that we carefully considered before we started work on a public version of Fluffy.
First, it has been possible to create AI generated images that mimic a particular artist’s style. For artists who are no longer living, this can be an amazing way for their work to live on and for their style to inspire current and future generations. For living artists, however, the situation is more complex. There are thousands of artists who have found that having a system that brings new ideas to life in their style is incredibly powerful and they are using this capability to enhance their own work. On the other hand, there are those who are intentionally mimicking an artist's style with the intention of impersonating or mocking the artist. Obviously we do not support or condone the use of the AI systems in this way. We were very happy to hear that the most popular open source AI system for image generation, Stable Diffusion, had started to make steps to make it difficult or impossible to copy artists' styles. (II). In addition, there is progress being made in allowing artists to remove their work from the AI training sets entirely if they choose. (I)
We feel that until these opt-out systems are fully in place and widely used, we would not feel right using any illustration or painting styles for Fluffy. Photographs, however, are another matter entirely. When the AI systems learn from the training images, they are not really learning about anything stylistic, they are learning about the world itself. They learn the concept of animals, facial features, fur texture, and sometimes even the correct number of fingers. They also learn the concepts of the science of photography, including camera and lenses, film stock and grain, lighting, poses, F-stop and shutter speed, and much more. Finally, the system learns about environments, forests, cities, and space stations.
What you end up with is something that can take this knowledge of the world as viewable by a camera lens and create entirely new images by combining concepts together in brand new ways. There is no collage or other copying of bits and pieces from other images to create the new images. Not only are the new images entirely unique, but the way the images are made means it is effectively impossible to even make the same image twice. We feel that since this workflow does not use any artistic styles and also creates only images that have never been seen before, this is the most ethical way to use the AI image generators and the only way we create Fluffy images.
There is another question that should be asked in any new project, especially AI generated ones. That is, what is the net effect on society of the project coming into existence? To answer this, we have to look at whether this is harming anyone by taking away their income making the same type of product using traditional means. To answer this, we calculated out the cost to create Fluffy using traditional means. Each Fluffy image would take between 100 and 500 hours using digital painting techniques to match the detail and resolution that the AI is able to achieve. If you assume a tragically low cost of $15/hr for a US based artist, at least 100 images per issue, and a very low estimate of 100 hours per image, the total cost to create a single issue would be $150,000 just for images. Managing communication for 100 artists, editing, revisions, concept exploration, poetry, and all the rest would likely double that. For a small run magazine like ours, where we hope to get 1000 copies to the printers on our first run, that would mean we would have to charge at least $300 per copy. It only becomes affordable for most people if you get into the 100,000 subscriber range, which is not likely to happen for such a silly endeavor. We think this is the primary reason that nothing likely Fluffy exists currently….because it couldn’t until very recently. So, if the magazine is impossible or at least exceedingly improbable to create traditionally, it is unlikely to hurt any artists making the same thing.
Now for the positives. The first is the pure joy we have seen come out in everyone we have shown Fluffy to. Second are the activities built in for kids of all ages that can inspire creativity, build confidence, and help them express their artistic vision. Finally, we have the jobs required to bring Fluffy to life. Not just for May and myself, but the printers, suppliers, and mail delivery.
The final ethical consideration we had was with the CO2 and environmental impact of making Fluffy. National Geographic found that the carbon footprint of creating a single issue of Fluffy was less than 2 pounds. (III) This is roughly the CO2 created by driving 2 miles or breathing for a day. We think that the environmental tradeoff is reasonable, as the vast majority of paper production is using planted and managed forests, especially considering the advantages of being able to look at Fluffy over and over and keep it for years.
References
Downing, K. (2023, January 26). (I) An IP Attorney's Reading of the Stable Diffusion Class Action Lawsuit. Law Offices of Kate Downing. Retrieved January 30, 2023, from https://katedowninglaw.com/2023/01/26/an-ip-attorneys-reading-of-the-stable-diffusion-class-action-lawsuit/
(III) What's the carbon footprint of a printed magazine ? (2021, October 25). Pulse Magazine. Retrieved January 30, 2023, from https://www.pulsemagazine.co.uk/blogs-articles/what%E2%80%99s-the-carbon-footprint-of-a-printed-magazine/
(II) Stable Diffusion made copying artists and generating porn harder and users are mad. (2022, November 24). The Verge. Retrieved January 30, 2023, from https://www.theverge.com/2022/11/24/23476622/ai-image-generator-stable-diffusion-version-2-nsfw-artists-data-changes