I instinctively agree there is an important difference.
If you try to define systematically what that difference is, though, it's not obvious. At the end of day, I think it's something like "degree of difficulty" or "amount of thought", which are vague concepts. Yet most would agree what the author here did requires more skill and thinking than typing "image of celtic knot" into Gemini.
I used to work on procedural graphics, and to me the clear difference is that all the training involved happened inside my brain. This author's article describes a similar process. He's not throwing a lot of existing examples into a black box, letting it learn their features, then driving it to emit new images with similar features: he is learning, himself, what those features are, inventing a process which fits those bounds, then automating it with code.
If you try to define systematically what that difference is, though, it's not obvious. At the end of day, I think it's something like "degree of difficulty" or "amount of thought", which are vague concepts. Yet most would agree what the author here did requires more skill and thinking than typing "image of celtic knot" into Gemini.