3 ms·

Don't tell me that DALL-E is good. I'm still waiting for it to generate hands instead of shoggoth tentacles.

by coldacid 3y ago

unshavedyak 3y ago
Funny, i bought art recently from an (human) artist who has trouble with hands.
kolinko 3y ago
What? Dalle generates hands with no issues.
- agildehaus 3y ago
  Often but certainly not always.
- pan69 3y ago
  https://www.buzzfeednews.com/article/pranavdixit/ai-generated-art-hands-fingers-messed-up https://www.buzzfeednews.com/article/pranavdixit/ai-generate...
  - carnitine 3y ago
    January 2023 is an age ago in terms of generative AI. DALL-E 3 has been out what, a month? I just tried and it generated hands no problem.
    - pan69 3y ago
      Sure. I get that. I was simply providing context to kolinko's comment (albeit in a low effort way) in case he/she wasn't aware that this was a problem at some in time.
v64 3y ago
A dated stereotype at this point. The latest versions of Dalle, Stable Diffusion, Midjourney etc. are much better at this. Generation of words (e.g. making signs that say specific things and not gibberish) has also improved.
nmca 3y ago
That was six weeks ago aha, the models work now.
zwieback 3y ago
I just asked it to create shoggoth tentacles and they look quite spooky.
schmookeeg 3y ago
it might be worth removing "in the style of chris simpsons artist" from the prompt :)
drdaeman 3y ago
Hands are manageable and fine most of the time, but I've yet to find a way to describe a complex scene. It can't even get a number of people right half the time. Last thing I've tried was a Distracted Boyfriend meme, but with people dressed like a cat, a cat tree, and a sofa (cardboard cosplay style). Maybe I simply don't know the right way to explain and make it write a perfect prompt, but it was insurmountable task for DALL-E. It always forgot something - either that there should be only three people, or that there's one cat-costumed person, or that a sofa is a costume and not a piece of furniture, or that I wanted to have the obvious canonical layout of the meme and so on. I just gave up. With Stable Diffusion at least I can do this using iterative inpainting or ControlNet segmentation. DALL-E simply cannot iterate on existing images.
- EnergyAmy 3y ago
  If you're using ChatGPT's integration, you'll have to check what text it actually sends to DALL-E, if you didn't already know. Clicking a picture will show the text sent to DALL-E. I've had ChatGPT screw up the instructions up a number of times and have had to say things like "Try again, but this time include this literal string in your prompt to the image generator: '...'"
  - drdaeman 3y ago
    Thank you. Yes, I know about this, and I've audited those prompts. E.g. the prompt literally says "glance at the other woman dressed as a sofa", and DALL-E paints a woman sitting on sofa. Or the prompt says "Only the boyfriend should have cat features" but all three people have cat ears (SD is also prone to this, but there's ControlNet). Overall, DALL-E feels like a dumbed-down version Stable Diffusion without repeatability (can't control the generation, so can't reuse the looking-good seed) and a bunch of extra hoops to jump through.
    - simonw 3y ago
      Actually it turns out you can reuse seeds in DALL-E 3, and it's surprisingly effective at retaining the same style and image composition across small tweaks to the prompt against that same seed: https://simonwillison.net/2023/Oct/26/add-a-walrus/#reusing-seeds https://simonwillison.net/2023/Oct/26/add-a-walrus/#reusing-...
ainiriand 3y ago
Just check this I just generated: https://www.bing.com/images/create/a-japanese-female-samurai-bloodied-after-a-battle-/654542dd957d452a8c5f6b888f3c1408?id=mlYK409t61uqoNwHtaTYNQ%3d%3d&view=detailv2&idpp=genimg&idpclose=1&FORM=SYDBIC https://www.bing.com/images/create/a-japanese-female-samurai...
Pent 3y ago
evidence of future shock, this comment is outdated by at least 2 major iterations
taneq 3y ago
Add 'deformed hands' to the negative prompt, et voila... no shoggothy goodness.

Don't tell me that DALL-E is good. I'm still waiting for it to generate hands instead of shoggoth tentacles.

unshavedyak 3y ago

Funny, i bought art recently from an (human) artist who has trouble with hands.

kolinko 3y ago

What? Dalle generates hands with no issues.

agildehaus 3y ago
Often but certainly not always.
pan69 3y ago
https://www.buzzfeednews.com/article/pranavdixit/ai-generated-art-hands-fingers-messed-up https://www.buzzfeednews.com/article/pranavdixit/ai-generate...
- carnitine 3y ago
  January 2023 is an age ago in terms of generative AI. DALL-E 3 has been out what, a month? I just tried and it generated hands no problem.
  - pan69 3y ago
    Sure. I get that. I was simply providing context to kolinko's comment (albeit in a low effort way) in case he/she wasn't aware that this was a problem at some in time.

v64 3y ago

A dated stereotype at this point. The latest versions of Dalle, Stable Diffusion, Midjourney etc. are much better at this. Generation of words (e.g. making signs that say specific things and not gibberish) has also improved.

nmca 3y ago

That was six weeks ago aha, the models work now.

zwieback 3y ago

I just asked it to create shoggoth tentacles and they look quite spooky.

schmookeeg 3y ago

it might be worth removing "in the style of chris simpsons artist" from the prompt :)

drdaeman 3y ago

Hands are manageable and fine most of the time, but I've yet to find a way to describe a complex scene. It can't even get a number of people right half the time. Last thing I've tried was a Distracted Boyfriend meme, but with people dressed like a cat, a cat tree, and a sofa (cardboard cosplay style). Maybe I simply don't know the right way to explain and make it write a perfect prompt, but it was insurmountable task for DALL-E. It always forgot something - either that there should be only three people, or that there's one cat-costumed person, or that a sofa is a costume and not a piece of furniture, or that I wanted to have the obvious canonical layout of the meme and so on. I just gave up. With Stable Diffusion at least I can do this using iterative inpainting or ControlNet segmentation. DALL-E simply cannot iterate on existing images.

EnergyAmy 3y ago
If you're using ChatGPT's integration, you'll have to check what text it actually sends to DALL-E, if you didn't already know. Clicking a picture will show the text sent to DALL-E. I've had ChatGPT screw up the instructions up a number of times and have had to say things like "Try again, but this time include this literal string in your prompt to the image generator: '...'"
- drdaeman 3y ago
  Thank you. Yes, I know about this, and I've audited those prompts. E.g. the prompt literally says "glance at the other woman dressed as a sofa", and DALL-E paints a woman sitting on sofa. Or the prompt says "Only the boyfriend should have cat features" but all three people have cat ears (SD is also prone to this, but there's ControlNet). Overall, DALL-E feels like a dumbed-down version Stable Diffusion without repeatability (can't control the generation, so can't reuse the looking-good seed) and a bunch of extra hoops to jump through.
  - simonw 3y ago
    Actually it turns out you can reuse seeds in DALL-E 3, and it's surprisingly effective at retaining the same style and image composition across small tweaks to the prompt against that same seed: https://simonwillison.net/2023/Oct/26/add-a-walrus/#reusing-seeds https://simonwillison.net/2023/Oct/26/add-a-walrus/#reusing-...

ainiriand 3y ago

Just check this I just generated: https://www.bing.com/images/create/a-japanese-female-samurai-bloodied-after-a-battle-/654542dd957d452a8c5f6b888f3c1408?id=mlYK409t61uqoNwHtaTYNQ%3d%3d&view=detailv2&idpp=genimg&idpclose=1&FORM=SYDBIC https://www.bing.com/images/create/a-japanese-female-samurai...

Pent 3y ago

evidence of future shock, this comment is outdated by at least 2 major iterations

taneq 3y ago

Add 'deformed hands' to the negative prompt, et voila... no shoggothy goodness.