Panel: Engaging Quixotism in Generative Artificial Intelligence: A GC1Y Experience
Faculty Mentor(s) Name(s)
Daniel Holcombe
Abstract
In this Digital Humanities panel, undergraduate students share their journey using Generative Artificial Intelligence (Gen-AI) in a GC1Y course that explores the novel Don Quixote in cultural production. Tasked with using Gen-AI to reproduce compositional elements of a book illustration, the panel presenters share their experience in prompting – through input narrative – their chosen Gen-AI to recreate the illustration as closely as possible. Gen-AI’s quixotic tendencies are highlighted through their quests to produce an acceptable image as output. Presenters: Abbey Howell, abbey.howell@bobcats.gcsu.edu Hampton Sapp, richard.sapp@bobcats.gcsu.edu Jorja MacKenzie, jorja.mackenzie@bobcats.gcsu.edu Hannah Savage, hannah.savage@bobcats.gcsu.edu Presenters and Presentation Titles: Abbey Howell: “Capabilities and Challenges of AI: Replicating Illustrations and Narratives of Don Quixote” Hampton Sapp: “‘Don Pixelote’: Quixotic Artificial Intelligence Image Generation” Jorja MacKenzie: “Recreating Iconography of Don Quixote with Artificial Intelligence” Hannah Savage: “Nuanced Narratives Matter: Prompting Artificial Intelligence to Reduplicate a Book Illustration of Don Quixote” Abstracts: Abbey Howell This presentation explores the complexity behind reduplicating book illustrations from the novel Don Quixote by Miguel de Cervantes using Artificial Intelligence (AI). This concept poses the question: “Is AI able to effectively replicate important elements shown in the illustration while also maintaining vital elements that make up the narratives of the novel?” To answer this, I began with an original illustration depicting a classic scene from Don Quixote. Desiring to fulfill his knight errantry duties, Don Quixote valiantly charges into a flock of sheep that he believes are two rival armies, creating a comical scene. The image I chose was created by Gaetan Evrard in 1986, a Belgian author and illustrator who specialized in children’s novels. In his illustration, Evrard places Don Quixote in the middle of the herd of sheep, and his artistic style gives the characters a very animated and cartoonish appearance, which is relevant considering the illustration is aimed towards a young audience. This presentation explores how proficient AI is at replicating scenes from Don Quixote while maintaining the important compositional elements relevant to both the illustration and novel. Using Dall-E 3, I provided the program with several prompts until I was given an image that related to the original illustration as closely as possible. The differences between my first and last AI prompts were drastically different, ultimately showing the complexity behind using AI to get desired images and highlighting its questionable reliability. Keywords: Accuracy, Artificial Intelligence (AI), Don Quixote, Complex, Illustration, Image, Prompt Hampton Sapp Artificial Intelligence (AI) image generating programs have become exponentially more effective at creating precise imagery. While Dall-E 3 and other AI generators are more popular, lesser-known programs are now available, including Adobe’s Firefly. This presentation examines Firefly’s ability to accurately recreate a work of art, and the user prompts that yielded the best results. To achieve this, a 1905 illustration by Spanish artist José Jiménez Aranda from a special edition of Don Quixote was used as a reference for recreation. The Chinese ink illustration depicts a crucial part of the novel’s plot in which the storyline comes to a halt. The narrator finds the rest of the story written in Arabic. This is a very unique part of the novel that yields unique illustrations. There are very few illustrations pertaining to this portion of the book, making it particularly interesting to see how the AI recreated it in its output. The results of the experiment show that Firefly was unsuccessful in recreating the 1905 illustration. This presentation discusses the project details by analyzing AI prompt specifications, Adobe Firefly abilities, and its limitations. Keywords: Adobe Firefly, Artificial Intelligence (AI), AI Art, AI prompts, Book Illustrations, Don Quixote, Cervantes Jorja MacKenzie This presentation explores a project that recreated an iconic scene from Don Quixote via AI image generation, revealing both promises and limitations. I aimed to replicate a 2005 book illustration by Spanish artist Antonio Mingote Barrachina that showed Sancho Panza's comical tossing in a blanket using the platform wepik.com. Barrachina was known for his expressive etched illustrations depicting Spanish life and literature. His sketches utilized flowing contours and energetic linework to capture a sense of motion and emotion. Barrachina's cartoonish style featured exaggerated gestures and comedic absurdity. These unique artistic qualities presented an ambitious challenge for AI recreation. Through five rounds of refinement and adjusting details like actions, positioning, and emotions, the AI images gradually transformed toward the dynamism and relationships in Barrachina’s original. Adding verbs like “catapulted” and imagery like Sancho’s “chubby limbs splayed” injected palpable chaos. Interactive phrases linking foreground and background also improved coherence. The final AI-generated image contained noticeable similarities to key components like Sancho's suspended figure and the gleeful men below. However, finer artistic intricacies requiring discernment remained difficult for the AI. It struggled to replicate Barrachina's finer artistic details, which require human discernment. This exercise showed AI’s promise for recreating renowned scenes once only achievable by master illustrators. However, limitations arose revealing human creativity still exceeds AI’s visual replication capabilities. While the vigorous AI facsimile fell short of flawless accuracy, this demonstrates the technology’s future potential if evolution continues. Recreating cultural images remains an aspirational yet partially fulfilled goal for AI. Keywords: Antonio Mingote Barrachina, Artificial Intelligence, Book Illustration, cartoon, Don Quixote, wepik.com Hannah Savage Artificial Intelligence, although constantly evolving, is known to produce errors when asked to complete highly specific tasks. This challenge is especially prevalent in image creation, since using AI to produce images as output relies on the descriptive narrative input provided by the user. This presentation analyzes how Dall-E 3, accessed through Microsoft Bing Image Creator, can reduplicate a book illustration of Miguel de Cervantes’s novel Don Quixote of la Mancha. The illustration selected is “Disappearance of Don Quixote’s Library by the Enchanter Freston” (1950) by Spanish artist Enrique Herreros. Seven narratives were required to produce an image that resembles the original illustration. Analyses of these narratives and each corresponding AI-generated image reveal the efficiency of certain word choices that in turn helped the AI produce the desired images. As each narrative input expanded upon the previous one, and was slightly altered to produce specific changes, nuanced phrasing was revealed that resulted in improved AI-generated images. While the style of the AI image is partly dependent on the specific software used, as some AIs are better at producing distinct types of images than others, it is apparent that using concise, well-known phrases and vocabulary in the input is essential to create an image in which the AI did not confuse the details. Keywords:
Start Date
27-3-2024 2:00 PM
End Date
27-3-2024 3:00 PM
Location
Arts and Sciences Auditorium
Dr. Holcombe's panel information
Panel: Engaging Quixotism in Generative Artificial Intelligence: A GC1Y Experience
Arts and Sciences Auditorium
In this Digital Humanities panel, undergraduate students share their journey using Generative Artificial Intelligence (Gen-AI) in a GC1Y course that explores the novel Don Quixote in cultural production. Tasked with using Gen-AI to reproduce compositional elements of a book illustration, the panel presenters share their experience in prompting – through input narrative – their chosen Gen-AI to recreate the illustration as closely as possible. Gen-AI’s quixotic tendencies are highlighted through their quests to produce an acceptable image as output. Presenters: Abbey Howell, abbey.howell@bobcats.gcsu.edu Hampton Sapp, richard.sapp@bobcats.gcsu.edu Jorja MacKenzie, jorja.mackenzie@bobcats.gcsu.edu Hannah Savage, hannah.savage@bobcats.gcsu.edu Presenters and Presentation Titles: Abbey Howell: “Capabilities and Challenges of AI: Replicating Illustrations and Narratives of Don Quixote” Hampton Sapp: “‘Don Pixelote’: Quixotic Artificial Intelligence Image Generation” Jorja MacKenzie: “Recreating Iconography of Don Quixote with Artificial Intelligence” Hannah Savage: “Nuanced Narratives Matter: Prompting Artificial Intelligence to Reduplicate a Book Illustration of Don Quixote” Abstracts: Abbey Howell This presentation explores the complexity behind reduplicating book illustrations from the novel Don Quixote by Miguel de Cervantes using Artificial Intelligence (AI). This concept poses the question: “Is AI able to effectively replicate important elements shown in the illustration while also maintaining vital elements that make up the narratives of the novel?” To answer this, I began with an original illustration depicting a classic scene from Don Quixote. Desiring to fulfill his knight errantry duties, Don Quixote valiantly charges into a flock of sheep that he believes are two rival armies, creating a comical scene. The image I chose was created by Gaetan Evrard in 1986, a Belgian author and illustrator who specialized in children’s novels. In his illustration, Evrard places Don Quixote in the middle of the herd of sheep, and his artistic style gives the characters a very animated and cartoonish appearance, which is relevant considering the illustration is aimed towards a young audience. This presentation explores how proficient AI is at replicating scenes from Don Quixote while maintaining the important compositional elements relevant to both the illustration and novel. Using Dall-E 3, I provided the program with several prompts until I was given an image that related to the original illustration as closely as possible. The differences between my first and last AI prompts were drastically different, ultimately showing the complexity behind using AI to get desired images and highlighting its questionable reliability. Keywords: Accuracy, Artificial Intelligence (AI), Don Quixote, Complex, Illustration, Image, Prompt Hampton Sapp Artificial Intelligence (AI) image generating programs have become exponentially more effective at creating precise imagery. While Dall-E 3 and other AI generators are more popular, lesser-known programs are now available, including Adobe’s Firefly. This presentation examines Firefly’s ability to accurately recreate a work of art, and the user prompts that yielded the best results. To achieve this, a 1905 illustration by Spanish artist José Jiménez Aranda from a special edition of Don Quixote was used as a reference for recreation. The Chinese ink illustration depicts a crucial part of the novel’s plot in which the storyline comes to a halt. The narrator finds the rest of the story written in Arabic. This is a very unique part of the novel that yields unique illustrations. There are very few illustrations pertaining to this portion of the book, making it particularly interesting to see how the AI recreated it in its output. The results of the experiment show that Firefly was unsuccessful in recreating the 1905 illustration. This presentation discusses the project details by analyzing AI prompt specifications, Adobe Firefly abilities, and its limitations. Keywords: Adobe Firefly, Artificial Intelligence (AI), AI Art, AI prompts, Book Illustrations, Don Quixote, Cervantes Jorja MacKenzie This presentation explores a project that recreated an iconic scene from Don Quixote via AI image generation, revealing both promises and limitations. I aimed to replicate a 2005 book illustration by Spanish artist Antonio Mingote Barrachina that showed Sancho Panza's comical tossing in a blanket using the platform wepik.com. Barrachina was known for his expressive etched illustrations depicting Spanish life and literature. His sketches utilized flowing contours and energetic linework to capture a sense of motion and emotion. Barrachina's cartoonish style featured exaggerated gestures and comedic absurdity. These unique artistic qualities presented an ambitious challenge for AI recreation. Through five rounds of refinement and adjusting details like actions, positioning, and emotions, the AI images gradually transformed toward the dynamism and relationships in Barrachina’s original. Adding verbs like “catapulted” and imagery like Sancho’s “chubby limbs splayed” injected palpable chaos. Interactive phrases linking foreground and background also improved coherence. The final AI-generated image contained noticeable similarities to key components like Sancho's suspended figure and the gleeful men below. However, finer artistic intricacies requiring discernment remained difficult for the AI. It struggled to replicate Barrachina's finer artistic details, which require human discernment. This exercise showed AI’s promise for recreating renowned scenes once only achievable by master illustrators. However, limitations arose revealing human creativity still exceeds AI’s visual replication capabilities. While the vigorous AI facsimile fell short of flawless accuracy, this demonstrates the technology’s future potential if evolution continues. Recreating cultural images remains an aspirational yet partially fulfilled goal for AI. Keywords: Antonio Mingote Barrachina, Artificial Intelligence, Book Illustration, cartoon, Don Quixote, wepik.com Hannah Savage Artificial Intelligence, although constantly evolving, is known to produce errors when asked to complete highly specific tasks. This challenge is especially prevalent in image creation, since using AI to produce images as output relies on the descriptive narrative input provided by the user. This presentation analyzes how Dall-E 3, accessed through Microsoft Bing Image Creator, can reduplicate a book illustration of Miguel de Cervantes’s novel Don Quixote of la Mancha. The illustration selected is “Disappearance of Don Quixote’s Library by the Enchanter Freston” (1950) by Spanish artist Enrique Herreros. Seven narratives were required to produce an image that resembles the original illustration. Analyses of these narratives and each corresponding AI-generated image reveal the efficiency of certain word choices that in turn helped the AI produce the desired images. As each narrative input expanded upon the previous one, and was slightly altered to produce specific changes, nuanced phrasing was revealed that resulted in improved AI-generated images. While the style of the AI image is partly dependent on the specific software used, as some AIs are better at producing distinct types of images than others, it is apparent that using concise, well-known phrases and vocabulary in the input is essential to create an image in which the AI did not confuse the details. Keywords: