Hacker News new | ask | show | jobs
by simiones 1375 days ago
But if I want a good picture of an owl, I Google "owl" and get many more options than I could possibly ever have time to pick from. Stable Diffusion is essentially doing the same thing as Google, except presenting a kind of average result instead of showing me all the results in its DB.

Now, this may actually be helpful in that it gets around copyright claims - but that's the only real difference.
2 comments
_aavaa_ 1375 days ago
And you are free to search through the whole catalog of google results until you find an owl that looks exactly like you want. Though this is going to get harder as you want something more specific than a simple owl.

But the approach for stable diffusion is just as easy whether you want just "an owl", or "an owl in X's style with A, B, and C"

link
simiones 1375 days ago
Changing the prompt until it generates what I want is not that different from changing my search terms until the result I want is closer to the top.

Now, I should of course note that search engines already employ ML techniques to actually interpret search terms, so to some extent the point is moot - ML is important to actually solving this problem.

link
_aavaa_ 1375 days ago
But searching on google doesn't "generate" anything, If your image isn't on the web, there's nothing to bring "closer to the top".
link
simiones 1375 days ago
Sure, but chances are, it is already on the web.

And of course, it's also possible that the image I want can't be generated by SD/DALL-E/etc.

link
Paracompact 1375 days ago
Go ahead and get me a photo off Google images of an alpaca in a suit playing chess in vibrant digital painting style.

Without meaning to sounding rude about it... I'll wait.

link
Jevon23 1375 days ago
I'd be curious to see if you could get that from the AI as well.

I tried generating that exact prompt a few times at theartbutton.ai and all the results were nonsensical.

For example: https://theartbutton.ai/image/OW1HZLfhjg6DFvJtk4vQZzUYqI7pGG...

link
Paracompact 1375 days ago
Here are my best attempts: https://imgur.com/a/obZH7X5

Not a very wide range of what I could do with the idea in terms of composition, but just some variations of finishing touches/intermediate steps. I achieved this with some human-in-the-loop iteration and inpainting, but it was no more than 15-30 minutes toying around with it, and I'm no artist.

If you have a semi-decent graphics card and would like to experiment with a bunch of extra settings and tools than are readily available online, this is a good repo for that: https://github.com/AUTOMATIC1111/stable-diffusion-webui

link