How would a well-known artmaking generative AI (A-GAI) go about illustrating children's books? This chapter uses multiple prompting approaches to elicit children's book illustrations for known stories and more original plots to identify some early visual patterns from Deep Dream Generator. The research explores a variety of questions around the quality of the output images in conveying visual meaning in the children's book space (albeit using fictional prompts). This exploratory work provides some early and preliminary insights into this space, including in how to set up a light experiment using an A-GAI.
TopIntroduction
Adults who grew up reading children’s books of various types may think of such books fondly given how they extended the world of the child by offering characters, adventures, varied perspectives, humor, and insights…and by extending the power of language. For adults who have read children’s books in their childhood, they may have a mental abstraction of what the visuals from children’s books look like. Some visuals here may be captured in a Google Image Search (Figure 1).
Figure 1.
Children’s books (in terms of Google Image search)
Children’s literature is defined as “the totality of spoken and written products formed with an aesthetic concern that contribute to the development, interests, needs, pleasure, emotions and ideas of individuals between the ages of 2 – 15” (Sinan, Demir, & Dogan, 2017, p. 133). [Some suggest that children’s literature are for any under 16.] Children’s literature “started to develop once children had been accepted as individuals” (Sinan, Demir, & Dogan, 2017, p. 133).
Children’s picturebooks are “profusely illustrated books in which the illustrations are, to varying degrees, essential to the enjoyment and understanding of the story” (Tomlinson & Lynch-Brown, 1996, as cited in Fang, 1996, p. 130). It is hard to say when picturebooks originated, but one of the first ones hails from 1658 in terms of a book by John Amos Comenius (titled Orbis Sensualium Pictus, a children’s encyclopedia illustrated with woodcuts) (“John Amos Comenius,” Wikipedia, Dec. 19, 2023). Picturebooks have traditionally included images as augmentation to text, but some books have few to no text and the rest as pictures in some present iterations (Karaman, n.d., n.p.). Picturebooks enable young readers to look at pictures even if they cannot read yet (Karaman, n.d., n.p.). It is important to understand the relationship of illustration to print in children’s books. One researcher suggests that visuals in children’s books “expand, explain, interpret, or decorate a written text” (Bodmer, 1992, p. 72, as cited in Fang, 1996, p. 131). A later work explains the complementary and augmentary role of visuals:
The characters in picture books must have specific traits that make them appealing to the child reader and that meet the demands of the short format. Since a short story does not normally allow for more fully developed characters, illustrations help develop the characters by depicting situations and emotions immediately familiar and credible to the children. (Fang, 1996, p. 132)
Visuals in children’s books may “correspond to aesthetic sensitiveness” (Sinan, Demir, & Dogan, 2017, p. 143). The current period has been referred to as the digital age in literary education, which requires readers to develop a new set of reading skills” (Pokrivcakova, 2017, p. 11).
This work explores how artmaking generative AIs (A-GAIs) may be used to create visuals. Artmaking generative AIs are tools that were trained on “big data” using artificial neural networks. From the training, the model acquires weights across various nodes as it “learns” the visuals. These models are fine-tuned for particular outputs.