How did this contact come about?
Gerhard: My wife and I stopped in Regensburg on vacation last year. During a visit to the Trinity Church, we met Pastor Dr. Gabriele Kainz. After a brief conversation, it was clear: we would do something together about AI and the church. We exchanged contact details, and a pleasant email correspondence began.
How did the project come about, and what was your approach?
Gerhard: In our email correspondence, the idea of designing church windows with AI emerged. We both found it so charming that we put it into practice. Dr. Kainz provided the texts, and I used text-to-image software to create two images from them. The source texts for the project come from the Bible, specifically from the Old Testament:
- Kings 19:4 But he himself went a day’s journey into the wilderness, and came and sat down under a juniper tree. And he requested for himself that he might die; and said: It is enough; now, O Lord, take away my life; for I am not better than my fathers.
- Kings 2:11 Then it happened, as they continued on and talked, that suddenly a chariot of fire appeared with horses of fire, and separated the two of them; and Elijah went up by a whirlwind into heaven.
These original texts were not yet perfectly suited for the text-to-image program. After all, the AI needs precise instructions on what to create. Therefore, we developed the following prompts (what the inputs are called):
- But he himself went a day’s journey into the wilderness, and came and sat down… and he requested for himself that he might die.
- A chariot of fire with horses appeared. Elijah went up by a whirlwind into heaven.
The software we used for this project is a self-written AI program that comes entirely from the Keras library. Basically, it is a simplified version of the Stable Diffusion software. Several parameters, such as height, width, and resolution, were adjusted. The AI (still) couldn’t handle these well.
What images resulted from this?
Gerhard: Around 80 images were generated and exchanged between Göttingen and Regensburg for selection. In the end, two pictures were chosen.
The church windows were approximately 2.5 meters by 6.3 meters, so some pixels had to be adjusted. Additionally, the central struts of the windows should not disrupt the picture. We had to consider these constraints, which couldn’t simply be communicated to the AI.
This is what the original pictures looked like at the church window:
Were there further uses for the images?
Gerhard: We then tried something else: When the images were input into an image-to-text program, essentially reversing the process, the following texts were produced:
- The picture shows an illustrated scene with a central figure who appears to be an older person with a long white beard sitting on the ground in a desert landscape. The figure is wearing a multicolored cloak and is holding something small between his hands, perhaps looking at it thoughtfully. The background features vibrant colors with swirling patterns in the sky, reminiscent of Vincent van Gogh’s “Starry Night,” and cacti are scattered in the sandy surroundings. Mountains can be seen in the distance under the dynamic sky. This image could be interesting due to its blend of realistic desert scenery and expressive, almost surreal sky and its potential symbolic or narrative content involving the central figure.
- The picture shows a lively and colorful scene with a person driving a chariot pulled by two horses. The background features a swirling sky with shades of blue, yellow, and orange, reminiscent of Vincent van Gogh’s “Starry Night” style. The person on the chariot is looking away from the viewer, wearing a hat and what appears to be clothing from a bygone era. This picture is interesting because of its dynamic composition and its homage to classic art styles.
“The texts are very impressive, even though they do not reflect the original context of the biblical passages,” Dr. Kainz aptly remarked.
And then there were also lectures on AI in Regensburg?
(Gerhard) Exactly! In addition to the AI stained-glass window project, we arranged lectures, including an opening lecture and two for best-agers. We addressed the following questions:
- How does AI learn?
- How does AI generate creative images?
- How does AI generate images from texts?
- Why does ChatGPT occasionally “hallucinate”?
- How will we live with AI in the future?
The opening lecture in the church was one thing. But the presentations for the best-agers were a real challenge for me; after all, you don’t know what level of knowledge the audience might have. My primary goal is to provide clear and comprehensive information.
However, wonderful people came with lots of questions. It’s clear that people over 65 may not have the same understanding of computers as students. But the discussion quickly got going and was very engaging. It wasn’t just about the technical aspects of how AI works but also about whether AI could take away jobs, who owns AI-generated images, or whether such images count as art.
Thank you, Gerhard, for this vivid account of your personal experiences with AI.
More about AI at Arineo
