What if the sharpest critique of your favorite image generator this year didn’t come from an artist, a lawyer, or a disgruntled Reddit thread, but from the Vatican?
Pope Leo XIV has come out against AI-generated art. His objection isn’t about copyright or job displacement, the two arguments that dominate every comment section. His argument is that AI art lacks the “spark of humanity” that real human creativity carries. Algorithms, he says, work from probabilistic patterns pulled out of scraped data, and that process is missing the essence of art entirely. One account of his position frames it as a call for the arts to safeguard imperfection, singularity, and the irreducibility of human experience.
I review AI tools for a living. I spend my weeks poking at generators, measuring what they actually do versus what the marketing deck claims. And I’ll admit this one landed differently than the usual criticism, because it’s aimed at a part of these products that no version bump can fix.
The critique that no release notes can answer
Most complaints about image generators are solvable, at least in principle. Hands with seven fingers? Fixed, mostly. Text that looks like alien runes? Much better now. Training data sourced without permission? That’s a licensing and policy problem, and some vendors are genuinely working on it. Every one of those is a roadmap item.
“It lacks the spark of humanity” is not a roadmap item. You can’t ship it in Q3. It’s a claim about what the output is, not how good it looks. And whether or not you agree with the theology behind it, that distinction matters when you’re deciding what a tool is for.
Here’s where my reviewer brain kicks in. The best AI image tools I’ve tested are excellent at a specific set of jobs:
- Mockups and placeholders that would otherwise eat a designer’s afternoon
- Visual exploration, where you need forty bad ideas to find one good one
- Internal decks, thumbnails, and filler where nobody is pretending it’s art
- Style tests before you commission the real thing
Notice what’s missing from that list. None of those jobs require a spark of humanity. They require speed, variation, and acceptable quality. Generators deliver all three. The trouble starts when vendors market the same product as a replacement for authorship rather than a replacement for stock photos and scratch sketches. That’s where the papal objection bites, and that’s where most of these tools quietly overpromise.
The deepfake problem is the practical half
The philosophical argument is interesting. The other half of this story is operational, and it’s the part I’d flag to anyone shipping a product.
Pope Leo XIV has himself been the subject of AI-generated deepfakes that misrepresented his words. Circulating fakes got convincing enough that the Diocese of Springfield in Illinois published guidance on how to spot them. Think about what that means in review terms: the same generative capability that makes a tool useful for mockups also makes it useful for putting invented statements in the mouth of one of the most recognizable people alive. The feature and the failure mode are the same feature.
I keep running into this when I evaluate generators. Provenance and labeling get treated as compliance checkboxes, bolted on late, easy to strip. Watermarks survive until someone screenshots the image. Metadata survives until someone re-encodes the file. If your product can produce a photorealistic public figure saying something they never said, the detection story is not a nice-to-have, it’s core functionality. Most tools I’ve tested treat it as an afterthought, and the fallout lands on dioceses writing explainers and newsrooms issuing corrections.
What I’d actually take from this
You don’t have to share the Pope’s convictions to find his framing useful. It draws a line that tool reviewers and tool buyers both need: there’s a difference between generating an image and making art, and between generating an image and making a claim about reality.
My honest read after a lot of hours with these products: use them where the output is a means, not the message. Mockups, iteration, scaffolding, visual note-taking. Be far more careful where the image is meant to carry meaning or stand in for something true. And treat any vendor that blurs those two use cases in its marketing with suspicion, because that blur is where the real problems come from.
The tension between craft and algorithmic output isn’t going to resolve cleanly, and it’s a genuinely open argument about what creativity means. What I can say from the review bench is narrower and more useful: these tools are good at some things and oversold for others, and that gap is the whole story.
đź•’ Published: