Hi! I’d like to share a specific observation that might help improve the logical consistency of visual generative models, especially when working with speculative evolution.
I’ve been experimenting with both DALL·E and Sora, trying to generate images based on a simple evolutionary question:
What would a feline look like if it developed the ability to fly through convergent evolution?
The idea wasn’t fantasy—no dragon-cats or mythological creatures—but rather something biologically plausible, similar to how bats or gliding squirrels evolved: the front limbs transforming into wings, with no extra limbs added.
The problem:
the model tend to generate six-limbed creatures (four legs + two wings), which contradicts real mammalian evolutionary pathways. This is a common trope in fantasy, but it’s incorrect in a scientific context.
(Both images show winged cats with four functional legs and additional wings, violating the rule that mammalian flight requires transforming front limbs, not adding extra ones.)
Suggestion:
It could be useful in the future to include an optional semantic filter or biology-aware mode to help guide outputs according to natural principles—especially in educational or speculative science use cases.
Having a toggle that allows for “plausible anatomy” or “realistic evolution logic” could greatly enhance the scientific potential of these tools.
Closing thought:
Thank you for reading, and congratulations on the incredible achievement that Sora represents. This comment comes from deep admiration and a desire to contribute meaningfully to its evolution.
Just to clarify, I’ve actually tried multiple approaches, including combinations with Stable Diffusion. While that generation at least gets the wings right, it’s essentially a bat body with a cat’s head pasted on.
What I’m aiming for is something more grounded in speculative biology—an anatomically coherent example of convergent evolution, where a feline evolves flight naturally (like bats did), not just a visual hybrid.
I totally get the humor, though—I just think it’s worth exploring how far the models can go when pushed beyond fantasy tropes.
There is no conceivable path that is going to make cats take flight anytime soon..physics is highly stacked against their form.
The evolutionary pressure is to survive habitat loss and to avoid humans - or to be the cutest and cuddliest of the litter and exploit human’s predisposition to not kill their own young.
That’s a perfect illustration—literally—of what I’m not trying to do.
I’m not aiming for fantasy hybrids or stylized creatures with 6 limbs and magical anatomy. I’m interested in plausible speculative biology: how a real mammal might evolve flight if subjected to similar pressures as bats or gliders.
Think less “pet dragon,” more “evolutionary thought experiment.”
The fact that DALL·E generates fantasy by default is exactly the limitation I was trying to highlight—so I genuinely appreciate this image as the most visual example of the very problem I brought up.
The problem:
Examples:
Suggestion:
Closing thought:
