AI art models like Midjourney and Nano Banana can wow us with wild, imaginative images. But a new study reveals their so-called creativity is just based on math. And if you’re a writer, designer or musician, that’s actually a reason to breathe easier.
I’ll keep shouting this from the rooftops: AI isn’t creative. Yes, it can generate stunning images and even string together a clever story, but what it lacks is the core human understanding that makes creativity so much more than just content. The internet is overflowing with content mixed with AI slop.
In the past few weeks, Nano Banana has blown up the internet with trends and practical uses. Every time you give the model a prompt, it comes up with something fun, weird and in many cases, jaw-droppingly good. But new research shows that the creativity behind models like Nano Banana, Midjourney and other image generators, actually comes down to very simple math. In other words, what we are seeing and calling “creative” is actually nothing more than technical.
Scientists looking under the hood of our favorite image generators found that what makes these models appear “creative” has to do with the models’ structure. Specifically, two factors.:
- Locality: The AI doesn’t visualize the whole picture at once. So, when you prompt it to create “a cat wearing sunglasses and riding a skateboard,” the model works on little areas at a time, zooming in on a corner of a puzzle before pulling back to reveal the full image.
- Equivariance: A fancy word for consistency. If you move part of the input, the output shifts right along with it; like sliding a stencil across paper.
These two aspects of the model force the system to mix and remix little fragments until something new pops out. In fact, the researchers of the study even built a bare-bones version with no training data at all, and it still generated images that looked strikingly similar to what the bigger AI models produce.
A reality check for creatives
Writers, designers, musicians and pretty much any other creative might find this study obvious. It’s something we knew all along: AI is predictable, and even more than we thought.
Models will always create within boundaries such as style, genre. But unlike machines, we can decide to break rules, when to stray from a particular style and when to include a rebellious spark. That alone is still ours.
Although it very much seems like machines are catching up to us, this study is a real reminder that AI’s “imagination” will always be mechanical. It’s remixing the pieces, not inventing new ones.
As a first-of-it's-kind study, the research analyzes the why behind why AI image generators can be "creative" so we can better control, guide and evaluate the work for diversity and safety. By understanding why AI models create what they do, we have a stronger foundation into how we can push them further by controlling scale, locality and patch mixing.
That’s actually freeing, isn’t it? Sure, AI can help me sketch ideas, but the true creativity — the decision to go off-script, to zig when the rules say zag — is still ours. We are the ones in control. Imagination and creativity are something AI can’t ever take away from us.
Bottom line
This study is good news for anyone working in the arts. It means that what sets your work apart isn’t at risk of being automated away. Machines remix within boundaries; humans decide when to cross them, stretch them or toss them out entirely.
The real creative edge — the spark to break rules, to invent something that’s never been seen before — still belongs to us.
Follow Tom's Guide on Google News and add us as a preferred source to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button!
