Wow, that's an impressive amount of images. How long did it take you to generate and curate so many?
Show HN: I made 7k images with DALL-E 2 to create a reference/inspiration table
31–37 of 37 posts
Re: Show HN: I made 7k images with DALL-E 2 to create a reference/inspiration table
#32Wow, that's an impressive amount of images. How long did it take you to generate and curate so many?
Thank you! I think I first started to collect the generated images 2 months ago, with the help of some scripts I wrote — I have a list of prompts (around 350), a list of subjects (5 currently), and then I rename/check all of the saved images to find the ones that are missing.
Thankfully, DALL-E 2 includes the prompt text in the image filename when you download them. This was very helpful!
Re: Show HN: I made 7k images with DALL-E 2 to create a reference/inspiration table
#33DALL-E tech aside, this is a great reference for artistic style.
Is it? I only had time to go through the start of the list, but...
Accurate:
Berkey, Magritte
Inaccurate:
Arkley, Botticelli, Elvgren, Eng, Kahlo, and ironically Dali
Mediocre:
Bouguereau, Degas, Hopper, Kandinsky, Leighton
Oh! I guess this is a great reference for "what this prompt will do to a Dall-E picture, since it usually won't resemble the original artist that much."
Re: Show HN: I made 7k images with DALL-E 2 to create a reference/inspiration table
#34I tried quite a few of them in stable diffusion, and while I found the results typically look worse, it was still interesting. How much curation of the output did you do or was it that exact prompt and the first result?
I would love to see it re-run with stable diffusion, and have those added along side/or ability to toggle.
Re: Show HN: I made 7k images with DALL-E 2 to create a reference/inspiration table
#35I tried quite a few of them in stable diffusion, and while I found the results typically look worse, it was still interesting. How much curation of the output did you do or was it that exact prompt and the first result? I would love to see it re-run with stable diffusion, and have those added along side/or ability to toggle.
I guess there are two scenarios:
1. The original prompt I gave it wasn't quite right, so I'd try a few different prompts with slightly different wording until I got something I was happy with. This happened with maybe 10% of the prompts I ran.
2. The prompt text was correct but it gave one or two bad images. In this case, I typically kept them rather than re-running them. I thought it would be a good showcase of its limitations. There's one particular type of image that I constantly (maybe 4 times) would get. It's so unique and I have no idea why it would generate that amongst other more "normal" images. You can see it here: https://generrated.com/prompts/extremeBokeh?prompt=extremeBo... — it's like a real estate listing. Quite strange.
Re: Show HN: I made 7k images with DALL-E 2 to create a reference/inspiration table
#36This is fantastic! I know there's another platform https://openart.ai that is a search engine for dalle2 prompts. Could be a good complimentary to this.
Re: Show HN: I made 7k images with DALL-E 2 to create a reference/inspiration table
#37Earlier quoted context omitted.
I'm using the default settings on the webui, here are the parameters: ``` Portrait of Joe Biden in the oval office, 4k render seed:1331361607 width:512 height:512 steps:50 cfg_scale:7.5 sampler:k_lms ```
Thanks for providing the seed, because that would let me show you how exactly the parameters can affect your specific image without generating a "random" different one every time. Check out the exact same seed and prompt and cfg_scale, but with the steps aka iteration number at 100 (50 in general feels way too low, even for the samplers that are kinda good with low iteration numbers). https://pasteboard.co/I6yXg5mZip…
that was a cool detailed breakdown.
how did you learn all that? i'm a noob at prompt engineering but would love to know how to get good at it with the least amount of time.
any good reasources to get up to speed?