[sleightofmind]

Can someone give this completely ignorant old fellow a description of what the author is doing in this article?

What is the initial drawing being done with?

What is this stable diffusion process? (Amazing is one answer.)

A very brief description of the software used.

Color me astounded. I actually liked (was delighted by) the results in step-6. When I realized it was done programmatically somehow, I had to pick myself up and get back in my chair.

[GaggiX]

The author is using a text-to-image neural network to generate an image, in this case the author is also conditioning the model on a sketch to guide the generation process.

[sleightofmind]

Okay. Now I'm starting to wonder if folks who worry about AI taking over are onto something. Well, not really... yet. But color me absolutely flabbergasted.

Thanks for the explain.

[dougmwne]

This has been flabbergasting a lot of us. I have been playing with these text to image models for about a month and I still cannot believe it.

Try for yourself a bit. Here’s my current favorite interface to Stable Diffusion. Give it a little sketch, then go wild with the prompt description. Try different style descriptions like oil painting or comic book.

https://huggingface.co/spaces/huggingface/diffuse-the-rest

[sleightofmind]

Thanks! I'm gonna try it out. First, though, I think I'm going to add an exorcist to my contacts list. This is honestly the closest thing to magic I've ever encountered.

[dougmwne]

I made this in your honor. Call on him in a time of need. https://labs.openai.com/s/r4Y8SfSWhNdiXpnh5vFj3aZ3

[sleightofmind]

Ahhh! That's great. Thanks!

[XCSme]

Like telling an artist "hey, could you add a tree", "now recreate it, but make it appear to be on the moon", only that he's talking to an AI (artificial neural network).

[sleightofmind]

Indeed. This seems like it's in the uncanny valley of can do vs should not be able to do. Delightful but strange.