Who “made” this? The state of IP for generated images
Computer-generated art is gaining popular traction, making headlines and even winning art competitions. Some artificial Intelligence (AI) models like DALL-E, Midjourney, and Stable Diffusion are trained using images and captions taken from the internet. Users provide text prompts, and these models generate a variety of relevant novel images. Other generative AIs can synthesize images without any human input. But who owns the resulting image? Who is its author? How should copyright work here?
Steven Thaler has been pressing this question ever since he requested copyright registration in 2018 for A Recent Entrance into Paradise, an image generated entirely with his computer program. Thaler sought ownership of the image through work for hire, with his computer program as the originating author instead of a human. The U.S. Copyright Office (USCO) rejected both his initial (2018) and subsequent (2019/2020) requests, noting the lack of human authorship and insufficient “creative input or intervention by a human author.” Unable to convince the USCO to deviate from a century of jurisprudence, Thaler struggled to copyright the image because he played no creative role in its generation.
Thaler may have struggled to gain copyright for fully computer-generated content, but another approach mixing original, creative content with the output of an AI model saw more success. Kristina Kashtanova’s graphic novel Zarya of the Dawn combines many Midjourney-generated images with the author’s own story, dialogue, and artistic choices. The USCO granted Kashtanova’s novel copyright registration this past September, making it the first creative work integrating AI image generation to receive copyright registration. A month later, however, the USCO considered canceling the registration, asking Kashtanova for specific proof of human involvement in the novel-writing process. From these two cases, it is clear that the USCO is concerned not with the mere inclusion of computer-generated art, but rather with how its inclusion affects the required amount of human creative input, only granting copyright registration when a baseline level is still necessary.
While significant human creativity isn’t required in the text-to-image aspect of prompt-based image generation models like Midjourney, crafting the prompts in a way that consistently yields high-quality results requires practice and specialized knowledge. For example, this blog post walks through the subtle tricks required to get DALL-E to generate an image of a cat riding a bicycle. While attempting a naive prompt like “a cat driving a bicycle” yields inaccurate results, adding instructions to draw in the style of a book illustration and then turn it back into a photo by specifying “an illustration by Michael Sowa, but as photography” yields a result more aligned with the original intent. Other keywords relating to camera details, celebrity references, and specific phrases like “studio lighting” can make subtle improvements, though users need a deep understanding of the underlying model architecture and training dataset to use them successfully.
If the USCO places a lot of emphasis on human creative input when determining whether to grant copyright registration, could the amount of creative effort required to craft detailed and specific prompts be leveraged to copyright results from prompt-based AI image generation models?
Update Feb. 25, 2023: While Kashtanova retains registration of most aspects of her work, the USCO reconsidered and eventually rejected the copyright registration of the images within her novel on the basis of lack of human authorship. Prompt generation as a form of creative input is a significant part of her lawyer’s response!
AUTHORED BY:
Jason Klein
REFERENCES:
https://www.creativebloq.com/news/ai-art-wins-competition
https://openai.com/dall-e-2/
https://en.wikipedia.org/wiki/Midjourney
https://github.com/CompVis/stable-diffusion
https://www.copyright.gov/rulings-filings/review-board/docs/a-recent-entrance-to-paradise.pdf
https://medium.com/merzazine/prompt-design-for-dall-e-photorealism-emulating-reality-6f478df6f186
https://www.theverge.com/2023/2/22/23611278/midjourney-ai-copyright-office-kristina-kashtanova
https://processmechanics.com/2023/02/22/a-mixed-decision-from-the-us-copyright-office/
PHOTO CREDITS:
Théâtre D’opéra Spatial by Jason Allen via Midjourney
