In that sense, yeah, fully agreed. Someone typing up whatever in the prompt is about as much of art as someone taking a random selfie in poorly lit environment on their smartphone.
However, there is definitely nuance to generating art using AI that is similar to photography as art in a way. There are massive blog posts and writeups that exist already, talking about different techniques for crafting prompts and fiddling with the model to achieve different results. Stuff like how ordering of the words matters (e.g., the order changes how much the model weighs a word), how the choice of different samplers affect the outcome, which words tend to be strong modifiers (e.g., if you want something looking like a real photo, add a specific camera model name), the number of iterations and guidance scale parameter choice, etc.
That's barely scratching the surface, given the whole thing became available less than a month ago, and people discover all those methods and techniques in real time. I am willing to bet that in a year, the complexity of a "good prompt" will be lightyears away from where we are now.
It's like someone invented a music instrument/artmaking tool, except it behaves a lot like a blackbox. So instead of developing techniques from the base knowledge of how the tool was created, we have to test and try bajillion different things to figure out how to get what we want.
And no, I am not trying to make a claim that generating art using those AI models is an equivalent to creating an acrylic painting. Just like i wouldnt claim that taking a great (in terms of art) photograph is an equivalent to creating an acrylic painting. Both of those stand in classes of their own.
However, there is definitely nuance to generating art using AI that is similar to photography as art in a way. There are massive blog posts and writeups that exist already, talking about different techniques for crafting prompts and fiddling with the model to achieve different results. Stuff like how ordering of the words matters (e.g., the order changes how much the model weighs a word), how the choice of different samplers affect the outcome, which words tend to be strong modifiers (e.g., if you want something looking like a real photo, add a specific camera model name), the number of iterations and guidance scale parameter choice, etc.
That's barely scratching the surface, given the whole thing became available less than a month ago, and people discover all those methods and techniques in real time. I am willing to bet that in a year, the complexity of a "good prompt" will be lightyears away from where we are now.
It's like someone invented a music instrument/artmaking tool, except it behaves a lot like a blackbox. So instead of developing techniques from the base knowledge of how the tool was created, we have to test and try bajillion different things to figure out how to get what we want.
And no, I am not trying to make a claim that generating art using those AI models is an equivalent to creating an acrylic painting. Just like i wouldnt claim that taking a great (in terms of art) photograph is an equivalent to creating an acrylic painting. Both of those stand in classes of their own.