this post was submitted on 23 Mar 2024
377 points (87.9% liked)
Technology
59605 readers
4202 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
stunning but uncreative af.
that still depends on the operator.
Yeah and there are tons of angles and gestures for human subjects that AI just can't figure out still. Any time I've seen a "stunning" AI render it's some giant FOV painting with no real subject or the subject takes up a 12th of the canvas.
Actually less so because it can't draw the stuff but because it doesn't want to on its own, and there's no way to ask it to do anything different with built-in tools, you have to bring your own.
Say I ask you to draw a car. You're probably going to do a profile or 3/4th view (is that the right terminology for car portraits?), possibly a head-on, you're utterly unlikely to draw the car from the top, or from the perspective of a mechanic lying under it.
Combine that tendency to draw cars from a limited set of perspectives because "that's how you draw cars" with the inability of CLIP (the language model stable diffusion uses) to understand pretty much, well, anything (it's not a LLM), and you'll have no chance getting the model to draw the car from a non-standard perspective.
Throw in some other kind of conditioning, though, like a depth map, doesn't even need to be accurate it can be very rough, the information density equivalent of me gesturing the outline of a car and a camera, and suddenly all kinds of angles are possible. Probably not under the car as the model is unlikely to know much about it, but everything else should work just fine.
SDXL can paint, say, a man in a tuxedo doing one-hand pullups while eating a sandwich with the other. Good luck prompting that only with text, though.
I mean, just like any other tool.
That's not a tool. A tool is something a mind uses to make something. AI is a generator in and of itself, requiring nothing from a mind.
Of course it does. An AI generator does nothing without a prompt. Give it a bad prompt, and it looks boring and uncreative.
The idea that you can throw anything (or nothing) into a generator and get something good out is a misconception. I’ve played around with generators, and can’t get much “good” out of them. But I’ve seen amazing looking stuff created by others.
yea I've also seen amazing stuff created by others. But that's not what we're talking about here
It literally is. The person I replied to explicitly said it’s a good tool but has no creativity. I said the creativity comes from the users skill.
If it’s a tool requiring a user to bring it to its full potential… then again thats what is being talked about.
These tools do literally nothing unless a user is involved. Be it setting up auto responses to certain text, or explicitly handing it instructions and tweaking as they go.