The cover reveal of Christopher Paolini’s latest novel should have set the book community rejoicing. After all, the author, best known for his Inheritance Cycle, is a beloved name in the fantasy genre; his warrior-magician Eragon and dragon companion Saphira grew up with a generation of adoring young readers.
The cover of Christopher Paolini’s novel Fractal Noise, set against a close-up of its background art | Photo Credit: Image from @torbooks on Twitter; background edited with Canva
However, on February 8, Mr. Paolini took to Twitter to express his surprise that his upcoming science fiction novel, Fractal Noise, had received a mix of both five-star and one-star reviews on the book rating platform Goodreads—despite the book not yet being released.
Screenshot showing both 5-star and 1-star book reviews on Goodreads | Photo Credit: @paolini on Twitter
A number of angry reviewers explained they were leaving one-star reviews on the platform to protest against Fractal Noise’s cover. The image itself is nothing out of the ordinary: an astronaut approaching a looming red-and-black abyss. However, publisher Tor Books admitted that the stock image “may have been created by AI.”
‘Fractal Noise’ cover and statement by Tor Books | Photo Credit: Cover image and statement taken separately from @torbooks on Twitter; compiled as a collage with Canva
Others on Goodreads left five-star reviews simply to counteract the one-star reviews. Despite the strong reactions, Tor announced that it was moving ahead with the controversial cover due to “production constraints.” Mr. Paolini also defended the move.
A cover up job
Close to 150 years ago, book covers in the Western world largely served the purpose of decorative or protective wrapping as opposed to art indicative of the book’s contents. A little over a hundred years ago, book covers began to actively reference the stories they contained. The iconic cover of the 1925 edition of The Great Gatsby, with its teary eye motif, dark cobalt blue sky, and city lights, conjure up images of the story’s upper-society drama and its sweltering aesthetic even today.
Cover page of The Great Gatsby | Photo Credit: Wikimedia Commons
In more recent years, a Canva template and stock images can get the job done in minutes. Now, with text-to-image generators, it’s not a surprise that AI-generated art is the next stage in the book cover’s evolution.
Sunandini Banerjee, senior graphic designer, editor, and translator, is known in the Indian publishing sector for her intricately layered and exuberant book covers that hold the stories of the many authors who publish their work with Seagull Books.
A collage of book covers designed by Sunandini Banerjee | Photo Credit: Sunandini Banerjee; compiled as a collage with Canva
“I can see the temptation or the ‘fun’ element in being able to throw a name/title at something and have a cover ‘generated’ as if by magic. I can also see the ease with which some self-published authors may ‘generate’ their own covers, and be spared the worry of locating, briefing and paying a cover designer,” she said via email, when asked about AI-generated covers by The Hindu.
“But I don’t want an entire creative field, an entire wonderful act of creativity—reading, drawing, imagining, interpreting—to be wiped out. One of the most wonderful aspects of the publishing industry—in every language—is the variety and vibrancy of the book covers, often our first entry point into the content,” Ms. Banerjee explained.
Also read: Artists fight AI programs that copy their styles
To test how easy it is to make AI-generated book covers and to learn how they might fare in the Indian publishing industry, we sent Ms. Banerjee two book covers we created using DALL-E, a text-to-image generator made by OpenAI— the same company that created ChatGPT.
After tweaking image descriptions and generating some suitable pictures, we formatted them on Canva and added the finishing touches—titles, author names, and typeface effects.
One was a cover for Bram Stoker’s vampire horror novel Dracula (1897), compared with an original cover from Penguin Books.
The other book cover was for Edward Said’s Orientalism (1978), compared with an original cover from Penguin Books. Orientalism explores the power dynamic between colonising states and the Middle East. Creating both covers took less than two hours in total; a practised DALL-E user could probably generate more sophisticated images in a shorter time span.
Ms. Banerjee was informed that these covers were made with AI and was asked to rate them. Ms. Banerjee called the Dracula cover “generic and absolutely unremarkable.”
Mr. Paolini’s cover received the same assessment from her.
For Orientalism, she said the cover was “plain terrible,” pointing out the poor image quality, multiple objects keeping the viewer from finding the main subject, and the unsuitable typeface. She provided a version of what she felt was a far better cover, which we used in our comparison.
“On the whole, there is nothing in the cover that makes it leap to the eye,” she concluded.
Suits and Stability
Aesthetics is not the only point of contention. Beautiful and professional looking AI-generated covers do exist on Amazon, but copyright is another obstacle. Even if designers or publishers accept AI-generated art or tools, other media providers may not.
In January 2023, stock image provider Getty Images said it was suing Stability AI, which created a text-to-image model called Stable Diffusion. Getty Images accused the AI company of unlawfully using its pictures to train the model.
If a book cover is made using a text-to-image platform where the data set contains copyrighted images, it’s unclear how much legal protection such a cover could enjoy. For now, however, there are multiple self-published books on Amazon that list ChatGPT as an author, Midjourney as the illustrator, or even both.
An AI-generated picture book on Amazon | Photo Credit: ‘Alice and Sparkle’ by Ammaar Reshi; screenshots from Amazon
While the use of copyrighted material in AI training data is a serious concern, treating a book cover in the same way as a modern artist’s painting, for example, may not be entirely accurate. Book designers Peter Mendelsund and David J. Alworth noted that many book covers aren’t art pieces, but rather “commodity packages.”
“Marshall McLuhan tells us that ‘the ‘content’ of any medium is always another medium,’ and this claim seems especially apt for the book cover, which recycles drawings, photographs, and text from book reviews and other sources,” the designers wrote in The Look of the Book (2020).
From that perspective, the idea of an AI-generated book cover image drawing from the work of other artists is not entirely unprecedented. But whether it will be welcomed by the Indian publishing industry—and more importantly, readers—is a looming question.
For instance, Ms. Banerjee said that she did not see her publisher, Seagull, using AI in the bookmaking process as they enjoyed every part of their work.
Larger book publishers, such as the Big Five or their imprints, have ethical obligations to the communities they serve, so using an AI-generated image may be seen by many readers as betraying human artists and designers in the bookmaking sector. However, rising costs could tip the balance in favour of AI text-to-image generators.
Replacing covers, replacing readers
It is a misconception that a book designer must also be a professionally trained artist or painter. Ms. Banerjee said that she could neither draw nor paint and that her work as a cover designer was only possible on a computer.
“I’m not against technology, as long as I run the program, and the program doesn’t run me. If I’m stuck over a book, or unable to ‘see’ an image, I’m all for any kind of inspiration from any source that can help me think differently,” she said, remarking that she had not explored AI technology to a great extent.
She had no objections to her work being added to the data sets of AI platforms, in order to grow their stores of knowledge and help others.
“I just hope it doesn’t become an industry practice,” Ms. Banerjee said. “It would rob the world of a lot of creativity, and it would render jobless a lot of creative workers. Once you replace the cover designers, then you replace the editors, then you replace the writers too . . . you see where this is going, don’t you? Do we replace the readers too, then?”
This is the second article in the series on Artificial Intelligence and publishing.
