Saturday, August 13, 2022

Two Ways to Input an Image Generator

There are two basic ways to generate and image with a system like Open AI's Dall-E 2.

One way is to write a text prompt such as "Outer Space Man action figure."

Both images so far were generated that way.

You can also upload a photo (above) into Dall-E2 and let it create variations (following): 

I find the first two text-prompted images a little dull, and the remaining three variations seem both horrifying and hilarious. There are a few interesting new ideas in there, but for the most part they look lumpy, awkward, weird—sometimes funny and sometimes disturbing.

I don't think we need to worry (yet) about this technology taking over from human designers.


Joel Fletcher said...

I too was granted access to the DALL-E 2 beta, and have been having fun playing with it. Agreed that artists don't need to worry about AI technology, at this time anyway. The results from DALL-E 2 are often very bizarre, naive, and lacking design and composition understanding. But that is part of the charm I suppose. Eventually AI will be able to learn from art and photography how-to books. If that is successful, then perhaps artists will have some competition and reason for concern.

Eugene Arenhaus said...

Ah, but so many people don't see the flaws in these mushy images and proclaim them realistic. So they'll probably happily use these neural networks (I am reluctant to call them AI because they are hopelessly unintelligent at this stage) to avoid paying artists.

I fear we are going to see more and more of these around us, used as cheap illustrations. The horror.

Anne said...

AI still needs prompting and ideas from humans. Also, humans are required for any sense of meaning to what AI churns out. That's what we have over the rest of creation, even our own. We assign meaning and can make connections without external prompting.AI is still just a tool

Luca said...

Maybe it's not Dall-E the real enemy: i'm seeing many professional artists really worried by newest softwares like Midjourney. Images created by it are honestly amazing and it's not even a finished software. Check for them in the web : it's actually mindblowing to think a computer generated them. The only flaw i can find is that they are somehow quite similar, when you get used to them you can recognize them even without knowing it's from an AI. Maybe it's not a big issue for traditional artists, but digital artists everywhere working at high level in the industry are really, really worried by it. Big names won't be affected but we are really close to the point in which small business will be erased by the AI. And from what i've heard Midjourney entered the studios workflow yet and it's been on for a very limited time by now...We are just a step from a complete revolution in commercial art. And things are not getting well for (human) artists.

zebulon said...

Hello, you can try Midjourney

zebulon said...

Hello, you can try Midjourney

Vabond said...

Hi Gunrney. I'm not familiar with Dall-E but Midjourney is really making big strides. Just a heads up: The tech is evolving at a rapid pace. Early next year promises to be a game changer. In a couple of years users will have a lot more control and this stuff will definitely end a lot of art careers but I hope that new doors will open up. I really hope so.

I'm just sad for all the artists who have spend the last 5-10 years trying to get their skills up to scratch suddenly replaced by any John Doe with a keyboard and imaginative vocabulary.

Also, it seems we are treading on some weird religious concepts here. In the beginning was the word and through the spoken word the universe was created (John 1). And of course Genesis... "And God said...." It's like the serpent in the Garden kind of did not completely lie when he said we will be like gods. There is a LOT to talk about here but I'm just squeezing this in as a teaser. ;)