Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Show HN: I made 7k images with DALL-E 2 to create a reference/inspiration table

generrated.com

Post reply on HN
Print view View original

11–20 of 37 posts

Re: Show HN: I made 7k images with DALL-E 2 to create a reference/inspiration table

#11

Post by ManuelKiessling »

This is not (yet) worthy of a "Show HN", so I'll drop it in here: If you are using Discord and want a low-barrier way to play around with Stable Diffusion, you can use my Bot to do so.

Use https://discord.com/api/oauth2/authorize?client_id=101337304... to invite the bot to your Discord server, or join my demo server at https://discord.gg/nsfeutx35z.

Disclaimer: while you can use the bot for free 5 times, it is not a completely free service; while I don't plan to turn this into a money-making endeavour, I cannot afford to keep the rather costy GPU instance running without outside funding.

Therefore, the bot comes with a credits system - you can buy credits on https://pay-a-robot.com, for $0,05 per credit. 1 credit = 1 text-to-image run with 3 result images.

If you want to host the bot yourself, the source code with a step-by-step setup guide is available at https://github.com/manuelkiessling/stable-diffusion-discord-....

Re: Show HN: I made 7k images with DALL-E 2 to create a reference/inspiration table

#12

Post by capitalsigma »

CuriouslyC wrote: post #10
capitalsigma wrote: post #7

Earlier quoted context omitted.

My stable diffusion output looks awful. I've been trying to recreate the xkcd about Joe Biden eating sandwiches, so I try something like "Joe Biden eating a sandwich in the oval office, 4k render, photograph" and I get nightmare fuel with pieces of bread attached to his head, faces that dissolve into random geometric shapes, toppings that melt into hands while a sandwich sits on a plate in front of him, etc. I had hi…

SD isn't great at generating images for detailed, weird prompts (at least not compared to DALL-E2). If you're not great at prompt writing or just having bad luck, you can use img2img with a rough sketch of what you want.

Here, a less specific prompt:

"Portrait of Joe Biden in the oval office, 4k render"

First attempt: https://pasteboard.co/IYo5m6KeaqF4.png

OK, I'll grant you that all of the parts of a face are there and reasonably correct (except for the bottomless pits of darkness in his nose and mouth)

Second attempt, I end up with these weird artifacts in his head half of the time (3/5 of my generated images)

* https://pasteboard.co/xDFv9KD7Or4U.png

* https://pasteboard.co/j7PqHPgorZ9G.png

Am I holding it wrong somehow?

Re: Show HN: I made 7k images with DALL-E 2 to create a reference/inspiration table

#13

Post by bambax »

filoleg wrote: post #6
whywhywhywhy wrote: post #2

End of the day, unless it's opened up Dall-E 2 will be seen as an evolutionary dead end of this tech and a misstep. It's gone from potentially one of the most innovative companies on the horizon to a dead product now I can spin up equivalent tech on my own machine, hook into my workflow and tools in an afternoon all because Stable Diffusion released their model into the wild.

Yeah, until Stable Diffusion became available, I felt that Dall-E 2 stance on not opening it up was sorta reasonable. Mostly because "groundbreaking tech producing all these impressive results that cost a ton to build, and I bet stable diffusion announcements were all just riding the hype, and it will disappoint at the end." I have never eaten my words as fast as I did when stable diffusion had finally released. Such…

Dall-e is kind of sad. It reminds me of Google Video, that thing before YT.

Re: Show HN: I made 7k images with DALL-E 2 to create a reference/inspiration table

#15

Post by filoleg »

capitalsigma wrote: post #12
CuriouslyC wrote: post #10

Earlier quoted context omitted.

SD isn't great at generating images for detailed, weird prompts (at least not compared to DALL-E2). If you're not great at prompt writing or just having bad luck, you can use img2img with a rough sketch of what you want.

Here, a less specific prompt: "Portrait of Joe Biden in the oval office, 4k render" First attempt: https://pasteboard.co/IYo5m6KeaqF4.png OK, I'll grant you that all of the parts of a face are there and reasonably correct (except for the bottomless pits of darkness in his nose and mouth) Second attempt, I end up with these weird artifacts in his head half of the time (3/5 of my generated images) * https://pasteboard.…

What is your guidance scale number, the number of iterations, and the chosen sampler? Those would be very relevant to know. Pretty much the most relevant thing aside from the prompt itself.

Setting guidance scale number higher typically results in imagery getting trippier and more surreal with more artifacts. So i feel like that's the main culprit for the artifacts.

I am pretty curious to see how far we can get with this prompt. So I will try playing with it later today and post the results and what I found in a reply to this comment.

Re: Show HN: I made 7k images with DALL-E 2 to create a reference/inspiration table

#16

Post by jimmygrapes »

The best thing about this imo is that it proved to me that these image generation algorithms aren't just regurgitating minor variations on an existing image somewhere in their database of billions of images. Although I understand the general idea of how these work, I still had my doubts. The astronaut one in particular made this stand out; many of these artists definitely did not draw astronauts, yet the resultant images do fairly well at emulating the distinct styles of the artists. Thank you for debunking one of my concerns!

Re: Show HN: I made 7k images with DALL-E 2 to create a reference/inspiration table

#17

Post by crazygringo »

This is fascinating.

Obviously it's handmade, but it gets me wondering -- is there any way to produce something like this algorithmically?

In other words, if you had full access to the DALL-E 2 (or Stable Diffusion or whatever) model... is there any kind of algorithm or method that could spit back the e.g. 100 most orthogonal text phrases that represent all the art styles?

I know that historically neural networks have been considered to be pretty much black boxes, but I'm curious if there's been any progress made at all in terms of interpreting them?

Re: Show HN: I made 7k images with DALL-E 2 to create a reference/inspiration table

#18

Post by yunwal »

glial wrote: post #3

DALL-E tech aside, this is a great reference for artistic style.

I also think it’s really funny to have Sol LeWitt on there because he sort of pioneered the idea of translating text instructions to art. He’s probably the first person to explore the idea of diffusion through a neural network (art history experts feel free to correct me here).

Re: Show HN: I made 7k images with DALL-E 2 to create a reference/inspiration table

#19

Post by yunwal »

yunwal wrote: post #18
glial wrote: post #3

DALL-E tech aside, this is a great reference for artistic style.

I also think it’s really funny to have Sol LeWitt on there because he sort of pioneered the idea of translating text instructions to art. He’s probably the first person to explore the idea of diffusion through a neural network (art history experts feel free to correct me here).

Btw if anyone was curious I tried feeding some Sol LeWitt-style instructions to Dall-E 2.

It didn’t produce an accurate representation unsurprisingly although it might have made the connection to LeWitt

https://labs.openai.com/s/GSthErBUTNJMho6aXJnWf6Yu

Post reply on HN