Hacker News Reader

Comment by davidbarker | Hacker News Reader

Apparently sign ups are currently down because of a "vendor outage". Perhaps because of the increased load?

Once you're able to get access, I believe you'll receive 15 credits per month for free. Each credit allows one generation, and each generation produces 4 images.

Rather than using up credits trying to learn how to formulate your prompts, I ran hundreds and uploaded them to https://generrated.com (I posted it a couple of weeks ago as a Show HN) — hopefully they might be useful as a starting point and save you some credits/money.

View on HN

This is a fantastic project.

I really liked the 16th century Indian painting of an astronaut [1], and now I see a role of programs like DALL-E in giving people a good intuition on how to identify art from different styles and periods.

[1] https://generrated.com/?prompt=16thCenturyPainting&subject=a...

Thank you for your kind words. I've learnt a lot about art styles while I've been putting the site together — both from the DALL-E 2 generations and research I've done to make sure those generations at least somewhat match the style I requested.

I'm not sure if you know (it might not be obvious — my fault!) but you can click on a prompt heading to see all 20 images created in one place, if that's more useful to you. e.g. https://generrated.com/prompts/16thCenturyPainting

Oh man that sequence of paintings of the discovery of gravity is hilarious. As if gravity suddenly got turned on in the middle of whatever these people were doing.

This is great. I really love all my Matisse generations. On the other hand, I find DALL-E to be uniformly bad at recreating Edward Hopper. Almost none of the generations capture the spirit of his work. It's especially obvious when you run the same Hopper prompts on Stable Diffusion. I wonder why.

Oh, for sure, they will be useful to me. I got access to DALL-E shortly before summer but I haven't played with it due to "prompts and credits" scheme. Thank you!

Thank you for the compilation. Much appreciated! I just wanted to mention that the images in the row labeled "pen and ink caricature" do not load.

Oops. Thanks! I'm not sure how I missed that. I've pushed a fix that should be live in 10 minutes.

This is pretty neat! Can also you use DALL-E mini/Craiyon to help with prompts or that going to be way off?

Thanks! I actually don't have much experience with DALL-E mini/Craiyon, but I have heard people have taken the output of Craiyon and using it as the input for Stable Diffusion to improve the quality.

I've been doing something similar with some DALL-E 2 images: https://news.ycombinator.com/item?id=33011336

Some more artists that often seem to yield interesting results: Giger, Klee, Klimt.

For what it's worth, it's dead simple to make Roger Dean-like watercolor gatefold album covers in SD. This is because Dean's process already involved use of randomness and refining watercolor effects with additional painting, which is similar to how Stable Diffusion works. Many of the Dean tropes just pop out for you, reliably.

What I'm finding interesting is twisting that to attempt non-Dean-like prompts within the 'gatefold album cover in watercolor' concept. It's not hard to get vibey graphics telling Stable Diffusion to paint in various mediums.

I would imagine you could do very decent Pollocks in Stable Diffusion if those are represented in the database. The artist's process shares elements with how the AI makes art.

Reply on news.ycombinator.com