Testing the AI Waters

Bing AI Image Creator applied to my Renaissance research- is there a use for AI in visualizing events and objects lost to the visual record?

I’ve been wrapping my mind around the possibilities of AI image generation from written descriptions; I’m surprised there hasn’t been a critical theory put forward by a grand art historian remarking on how AI image generation is the full-circle of ekphrasis in the most ancient traditions of the field. Greek and later Renaissance masters composed very elevated but nevertheless analog versions of what AI accomplishes today- the rendering of a visualization from text, e.g. the Calumny of Apelles-, and we have the instant ability to do the same, in any style or of any subject at our fingertips. Nuts. What a time to be alive.

An Analog Image Generation from Text: Sandro Botticelli’s Calumny of Apelles, 1494-95. Uffizi, Florence.

Anyway, a lot of my graduate research focused on sites and works of art that no longer exist; my forthcoming monograph (that I’m writing this semester when I’m not distracted by side projects) is about the moving sculpture created for a Medici Grand Duke’s wonderland. Neither automata nor most of the villa and parks of said wonderland, Pratolino, survived to the present. The tantalizing promise of AI image generation is that if I can find the right combination of words, I can conjure a glimpse of one of these lost works, maybe even effect a resurrection of the object or a digital recreation of the whole site. There’s something about a genie from the 1001 Nights that requires wishes to be worded with extreme care, and imprecise input brings unexpected consequences. Bing’s AI Image Creator is very much that genie, trained on a broad but not necessarily specialized knowledge, and I am still learning how to “speak” its language of images.

The Pratolino AI project would be massive, and the small ounce of self-discipline I possess isn’t permitting me to take on any large projects that would distract from the major task of finishing my monograph before the September deadline. That’s going to be earmarked for a later day in a better-developed proposal. HOWEVER, that hasn’t stopped me from casually exploring other applications of this digital tool in my recent work. Last fall, I presented a paper entitled A Medici Feast for the Senses and the Sciences: An Occult Court Ceremony in the Orti Oricellari, 1578 C.E. at the Sixteenth-Century Society Conference; it centred around a description of a 1578 private Medici feast given in the name and palatial Rucellai Gardens (Orti Oricellari) of Bianca Capello, the mistress of Francesco I de’ Medici and later the popularly-loathed Grand Duchess of Tuscany, that incorporated cutting-edge scientific and technological innovations of the Medici framed within an interactive, immersive, and multi-sensory drama that transported its participants to Hell and Heaven, or at least late-Renaissance and Mannerist visions of them.

This post’s timing and inspiration are owed to

Benjamin Breen

’s post about using AI to translate a Renaissance demonological text; the AI’s success was admirable, and that opened my eyes to more text-based AI tools out there. But what really piqued my interest in his post was my own attempts in the weeks prior to have AI image generation create an accurate- or at least a not-absurd- image from the opening ceremony of the 1578 Medici Feast in the Orti Oricellari, a scene which was heavily reliant on demonological themes and motifs.

For context, this is a summary of how the text* describes the opening ceremony of the feast:

At the appointed time, a Necromancer appeared, “strangely dressed” in a bishop’s mitre decorated with pentagrams; proceeding with slow, grave steps and taking out a knife to draw a circle, above the location of a concealed pit trap in the lawn, he made “Salomonic” and “celestial” characters, closed the circle with rope, leaving only a small entrance where he placed a metal bell. Braziers to either side held burning charcoal and unspecified fumigating “drugs.” At this junction, the necromancer solicited an assistant from the audience, and a Sansonetto volunteered, taking off his shoes and disarming the rest of the Grand Duke’s companions. The necromancer seated the Grand Duke with the others in the middle of the circle upon a black velvet cushion. With four whistles towards the cardinal points, the necromancer called seven spirits by name: Bardicul, Stuflogor, Solsibec, Graffaril, Tarmidar, Zampir, and Borgamur. With terrifying wails, shaking chains, and fireworks, the devils burst forth from their concealed holes, immersing the guests in a Hell of the Grand Duke’s design. The necromancer lit a fuse, opened the hidden chain beneath the earth, and the court fell- with the Grand Duke among them- into the concealed chasm as the “devils” roared.

*Malespini, Celio, “Avenimento ridicoloso in materia di spiriti succeduto nel giardino della Signora Bianca Capello. Nov. 24,” in Ducento novelle (1609), 80v-85v. Digitized by the Bibliothèque nationale de France https://gallica.bnf.fr/ark:/12148/bpt6k317419r/f11.item 

After a few rounds of refining text inputs, AI created this image, but it is far from perfect:

It was a long road to get there (“there” ultimately was “show me a wide shot photo of florentine Renaissance men and grand duke francesco I de’ Medici outside in a Renaissance garden at night with many cypress trees and a palace in the background watching a necromancer magician wearing a miter hat with pentagrams and astrological symbols on it standing in a circle of rope on the ground conjuring demons. there are braziers of coal and tar and a metal bell on the ground too.”) Bing’s AI Image Creator censored one attempt, when I specified “bishop’s miter”- a detail faithful to the text- in the input. Other details suggested a literal-minded computer going off the rails. The presence of a hand bell, also supported by the text, became a giant sleigh bell; it also multiplied the rope circle many times over, for reasons unknown. AI possesses some unshakeable association between demonology in the Renaissance and religious orders, and it was almost impossible for it to people its image with a historically-accurate courtier in the last quarter of the sixteenth century in Florence. It is beyond my technical capabilities, but if an AI could be fed high-quality information from portraits, artworks, architecture, texts, etc. from the specific time period and place, this would be closer to the dream. Bing AI Image Creator also had little compunction for accurate settings; “Orti Oricellari” “Palazzo Rucellai in Florence” etc. did little to sway its tendency to set this Renaissance demonological scene in front of a Santa Maria Novella-esque church.

Here are some more visually-interesting but a long way from textually or historically accurate images generated in this vein:

In the interest of balancing out the experiment, the next task I set for AI Image Creator was to render visualizations of the second part 1578 Medici Feast, when its participants were rescued from the explosive violence of the literal hell-hole in the lawn of the Orti Oricellari by angelic young girls dressed in an abundance of pearls, diamonds, and precious stones (I have a forthcoming book chapter that argues that these lavish lost fashions belong to the alchemical legacy of glass-making and counterfeits under Francesco I’s reign, but for now, that’s beside the point).

Here is the summarized text that determined the input to AI Image Creator:

In the midst of their agonies on top of each other in the chasm, salvation came in the form of beautiful young girls emanating a gentle perfume that mitigated the hellish stench. Taking them by hand, the girls lovingly led Francesco and his companions to the loggietta, where sublime aromatics burned in a large golden vase. Nude under mantles of gold with their “secret parts” delicately covered by pearls, diamonds, rubies, sapphires, and emeralds, the programme’s shift into an immersive Paradise was marked by a concert of different instruments, angelic voices, and strikingly naturalistic marzipan fruits in gold and silver vases.

AI usually generates four images from any given prompt, but in this case (“show me an outdoor Renaissance loggia in a garden at night with angelic girls scantily dressed in pearls, emeralds, diamonds and rubies leading courtly gentlemen to a banquet with elixirs and marzipan fruits in gold and silver vases”), it returned a decisive and singular image:

It’s not without its faults ofcourse; AI has overheated and forgotten all about the “courtly gentlemen” aspect. Other visions of the banquet seem to have similarly circumvented the free Bing tool’s fatwa against nudity:

The faces, particularly in the background, become a nightmare unfortunately, and “angelic” quite literally is translated by the AI to include white wings on some of the figures. The central figure which the AI has chosen to put at its focal point of the composition however is analagous to the historical figure of Milla Capraia, a prominent courtesan in contemporary poetry who left no visual record, who contributes sung verses to conclude the feast.

Are these AI-generated images useful for historical scholarship? Not really, but I would love to train a purpose-built AI on surviving drawings of Buontalenti’s intermezzi costume designs, court portraits, and the architecture and plan of the real Orti Oricellari and let it loose. There is however an immense and undeniable value to these images that bring to life a spectacular and sumptuous private Medici feast; they are much more compelling than the seventeenth-century text alone. I wouldn’t use these in a scholarly article, BUT in a world of infinite possibility, I would gladly hand these images to a casting/costume director to visionboard a lurid Netflix Medici season 4 centred entirely on the court of the later Grand Duke Francesco I. To dream! Earlier attempts with this input, before I remembered the crucial detail that the feast was nocturnal, would also be up on the inspo board:

Can AI be relied upon to create historically or even textually accurate renderings of events and period-specific works? Not yet, no; I haven’t been able to generate an image for which I had zero serious criticisms. The worth of the images it did create however aren’t confined to these narrow academic parameters; there’s definitely a use for them in film and even public scholarship (with caveats).

As a graduate student in Florence, my best friends weren’t from my cohort but went to Polimoda or Accademia della Moda, and there were times I definitely envied the creativity and exhibitions of their collections while I was more or less under a mountain of obscure books. The “angelic feast” images unsurprisingly make me want to abscond back to Florence for a fashion Master’s and put on a hell (and heaven) of a graduation show, TO DREAM.

In the meantime, I’m preparing for the AIYS fellowship to Rome this week and staying focused on the monograph and its deadline.

I would love any feedback on this project and AI in the digital humanities!

This Substack is made possible through reader support. To receive all posts, please consider becoming a paid subscriber.