My little experiment with images
So, I got curious about playing around with those picture-making tools, you know, the AI ones everyone’s talking about. My idea was pretty simple, or so I thought: could I get the AI to show the same person, same pose, but once with regular clothes on, and once… well, without certain items, more like showing the basic form underneath, like an artist’s dummy?
First off, getting started was a bit of a mess. These tools aren’t exactly straightforward. You type in words, hoping the machine understands what you want. I started typing things like “person standing, wearing a jacket” for the first picture. That worked okay sometimes, though the jackets often looked kinda wonky.
Then came the tricky part. I wanted the exact same person and pose, just, you know, minus the jacket. Seemed easy enough in my head. Just tell it “same person, same pose, no jacket”, right? Wrong.
- First attempt: Got a completely different person.
- Second attempt: The pose changed entirely.
- Third attempt: It just put weird blobs where the jacket was.
- Fourth attempt: The person looked like a badly erased drawing.
I spent a good few evenings just fiddling with the prompts. Tried using special codes or numbers people suggest online to keep the ‘seed’ the same, hoping that would lock the person and pose. Sometimes it helped a tiny bit, but mostly it felt random. It’s like the AI has a mind of its own, and it’s not very good at remembering what it just did a second ago.
What I found out
Basically, getting that kind of specific control is super hard. The AI is great at making new stuff that looks cool, but asking it to make tiny, specific changes to something it already made? That’s where it gets clumsy. It’s like trying to tell a toddler to edit a movie scene precisely.
It’s really hit or miss. You end up generating maybe ten, twenty pictures just to get one pair that looks vaguely similar. And even then, the lighting might be different, or the person’s expression changes for no reason. You can’t just tell it “remove the shirt” and expect it to understand the concept of layers or the body underneath existing consistently.
It wasn’t really about the ‘undressed’ part in the end, more about the sheer difficulty of getting the AI to be consistent. I was trying to see if I could use it for maybe comparing fashion styles on the same figure, or just understanding how clothing changes a silhouette. But the tools aren’t quite there for that level of detailed control, not without a ton of workarounds and sheer luck.
So yeah, that was my little project. Learned a bit about how these AI things work, mostly learned they can be pretty frustrating when you want something specific. Still kinda neat, but not the magic wand I thought it might be for this particular idea.
