Catholic educator Julian Ahlquist is one of the first Christian artists to use generative AI to create and disseminate religious imagery. Ahlquist’s online community, “Generation of the AInts,” uses a combination of Midjourney and Photoshop to churn out images of Catholic saints.
As a philosemitic Protestant and occasional Bible history podcaster, my own hopes for AI-generated Christian art have a somewhat different focus.
I have long complained that—while Christianity has produced beautiful art from St. Catherine’s Pantocrator to Salvador Dali’s Catholic period—the Christian artistic tradition contains a gaping hole it has never addressed. That hole consists of almost all of the adventure of the Old Testament.
Consider the biblical king Jehoshaphat. Jehoshaphat was a real historical figure—as real as Julius Caesar—and accomplished important military victories, legal reforms, and foreign alliances. At the time of this writing, however, the Google Image results for “Jehoshaphat” are mostly sketches from children’s bibles. It’s as if Jehoshaphat was a character from Mother Goose—not a head of state who defeated real Moabites in battle.
In fact, you can find much better art of fictional characters—including obscure figures from the lore of George R.R. Martin—than of most Old Testament kings. Martin’s internet fan community, which has made most of this art for free, is apparently more invested in Aegon the Conqueror than any Christian artist has ever cared for Hezekiah. Popular Renaissance subjects like David are the exception that prove the rule.
Neglecting all biblical art other than children’s illustrations has consequences. It is easier today to learn the names of Roman emperors, whose faces are reflected in stone, than the names of the Judahite kings who no Michelangelo has ever loosed from the quarry. Instead, the church has subconsciously trained Christians to think of Old Testament stories as taking place within a kind of non-historical Narnia, fit to be taught to children but not studied by thinking adults. Mother Goose-style sketches of Hezekiah may be sufficient as a rudimentary visual aid for small children, but they cannot capture the king’s character, fire us to aspire to his nobility, or draw out the depths of his doubts and the heights of his triumphs. Nor do they even try to.
We could speculate all day about the reasons that there is no high art of Hezekiah. Regardless, Christian artists are not making it. Can generative AI help correct the church’s failure to depict the heroes of the Hebrew Bible? Can Midjourney manufacture missing biblical artwork? To find out, I took a page from Julian Ahlquist.
Current generative AI is proficient at depicting architecture, landscapes, and people—in that order. As early as 2021, when I first started toying with “WOMBO Dream,” I noticed that the AI had a peculiar ability to create awe-inspiring High Medieval cathedrals. WOMBO could also go a few steps further, building cathedrals in cosmic landscapes and surrounding them with spacecraft inspired by John Berkey.
What Dream could not do—not even remotely—is create people. As anyone who used Dream in 2021 will tell you, any attempts to create human characters inevitably resulted in incoherent, surreal, and only vaguely anthropoid shapes.
Midjourney, which entered open beta testing in 2022, is different. In my first foray into Midjourney, I tasked it with creating artwork of the Catholic futurist theologian Teilhard de Chardin. Although the AI sometimes gave Chardin six fingers, it was—on the whole—more than capable of creating compelling portraits of the priest.
Like its predecessors, Midjourney also excels at creating beautiful settings for its subjects. The AI can even be directed to draw upon the styles of specific artists. It will obediently paint, for example, Teilhard de Chardin in space “in the style of Salvador Dali’s nuclear mysticism period.”
So long as your Midjourney prompts are limited to a single subject and a landscape, the AI is an efficient servant. My portraits of Teilhard, for example, required little tinkering and no photoshopping. When the occasional sixth finger emerged, simply re-running the prompt was sufficient to fix it.
When I sat down to begin creating Old Testament scenes, then, I had high expectations. My first idea for creating missing Old Testament art was ambitious: a depiction of 2 Kings 10:15-16.
In one of the most visually-loaded moments in the Books of Kings, the Israelite general Jehu—who has just begun a Yahwist revolution against the faithless Omride dynasty—is riding his chariot towards Israel’s capital. On his way, Jehu meets a man named Jehonadab on the road. Jehonadab, an early precursor of John the Baptist, has been living a Bedouin lifestyle in the wilderness, isolated from Israel’s corrupt society. The bronze-armored Jehu extends his hand to Jehonadab and beckons him to climb up into his chariot. “Come with me,” he says, “and see my zeal for the Lord.”
Although at least one decent depiction of this scene exists, it is closer to a comic book image than a classicist painting. In fact, there is little visual artwork depicting Jehu at all. Other than a colorful 19th century painting of Jehu confronting Jezebel by Edward Henry Corbould, most modern depictions of Jehu appear to be children’s Bible illustrations. This despite the fact that Jehu was perhaps the single most influential Hebrew king after David—founding a dynasty that lasted for five generations and reshaping the religious and cultural landscape of the entire Levant.
I knew that Midjourney had difficulty creating any scene with multiple people, and that portraying both Jehu and Jehonadab would be asking too much. Still, I thought that I might create the image from Jehonadab’s perspective, with Jehu extending his arm towards the viewer.
I soon discovered that even this expectation was naïve. While Midjourney can create excellent portraits and landscapes, it struggles to get human figures to physically interact with other objects—including other humans—in any specific way. In one well-known example of this problem, a famous AI-generated image of Trump and AOC holding hands shows both figures’ hands fused together.
In my own case, Midjourney had difficulties even creating an image of a chariot. It conjured up incoherent wooden contraptions and chimerically fused Jehu’s chariot with the bodies of horses. As for Jehu himself, getting him to extend his arm towards the viewer neared the limits of Midjourney’s abilities. Fine-tuning Jehu’s armor to make it historically accurate proved to be impossible.
The latter difficulty was almost certainly a result of Midjourney’s data set. When I had requested images of Teilhard de Chardin, Midjourney had little difficulty in ensuring that Teilhard had a priest’s collar, adjusting the style of his cassock, or giving him a cross pendant. Presumably, the AI simply had more data on modern Catholic priests than on 9th-century-BC Levantine warriors.
Moderating my expectations, I decided to settle for a loose depiction of 2 Kings 9:24, in which Jehu draws his bow to assassinate the Omride kings, in the style of Jacques-Louis David.
This time, Midjourney’s “variations” feature was helpful. Midjourney answers each prompt with four different images, then allows the user to select the best of the four—which the AI, in turn, rearticulates into four new variations. Using the variations feature multiple times can act as a kind of evolutionary process, steering the art in the direction of your internal vision before eventually reaching a point of diminishing returns.
Unfortunately, adjusting the prompt text and exhausting the variations feature was almost never sufficient to create any coherent image of a man drawing a bow. The bow would terminate at odd angles, the bowstring would run through the neck of Jehu’s horse, and second or third strings would poke nonsensically out from the bow and into the air.
Remembering that Julian Ahlquist often uses Photoshop to fine-tune his images, I decided to settle for the most satisfactory variation Midjourney had given me. Next, I downloaded and tinkered with GIMP, a free alternative to Photoshop. The “GIMP Resynthesizer” plugin, which I learned how to use on YouTube, allowed me to remove the pesky protrusions from Jehu’s bow and clear other nonsensical visual tumors out of the image.
Overall, this process made me more sympathetic with James Allen—the digital artist who infamously won an art fair with his Midjourney-generated image “Théâtre D’opéra Spatial.” Allen says he inputted “at least 624” text prompts before Midjourney delivered the image he envisioned—adjusting the scene, coloration, and style of the image. I believe him.
Midjourney cannot yet, by itself, replace the biblical artists that modern Christians have failed to raise, call, and educate. AI-based art—especially biblical art—still requires significant fine-tuning and other work. I do not expect to see a flourishing of artwork depicting Old Testament scenes until Christians reemphasize the visual arts as a calling—or until Midjourney, or an AI like it, makes significant steps forward.
Both scenarios are interrelated. Present AI technology could help to hasten a renaissance of Christian visual arts. While Midjourney does not close the gap between our abilities to imagine and create, it does shorten it. If would-be Christian artists are heartened by the tools now available to them, and begin to make the most of current generative AI, biblical art could enter a new period of flourishing tomorrow.
Other, more human steps might expedite this process. Before the breakthrough of Midjourney, I suggested creating a DAO to pay Christian visual artists to create images from biblical and church history. The development of AI art could make a proposal like this less urgent but more immediately practicable.
At the same time, the further development of Generative AI is inevitable. Midjourney can already identify, create, and modify objects in response to human instructions. It is only a matter of time before we can simply tell the AI about that sixth finger, extraneous bowstring, or other visual tumor and ask the AI to remove it. Once this happens, of course, it will become rare for such protrusions even to appear in the first place.
Likewise, while the next iteration of Midjourney may not be trained on depictions of bronze plate armor, it may be able to understand and implement human instructions at a higher level of specificity and, accordingly, create a better approximation of all the visual details in our imagination.
If this process continues to its logical conclusion then—God willing—the volume and quality of artwork in the world will be limited by only two variables: our willingness to conceive of beautiful things and the degree of beauty we can imagine.
Currently, the beauty of the Christian faith is partially obscured from the world by the unwillingness of the church to cultivate beauty. This unwillingness depresses the number of Christian artists in the church and limits the resources available to those artists who do emerge.
If AI art reaches its full potential, however, it could unshackle Christian art from these cultural and logistical obstacles. The beauty of the Christian faith may then shine forth anew and more fully illuminate the world.
Ian Huyett is a litigation attorney and a commentator on law, religion, and technology. You can follow him on Twitter at @IanHuyett.