-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
So how does this work? Is it actually attempting to draw a fresh image or taking elements of images from a Google search and mish mashing them together at random? I'm kind of tempted to say the latter by the amount of noise it generates, but the results are so far off the actual image sometimes I'm also thinking the former. I just don't get why it would generate so much noise if it was doing the former
Assuming it uses the same foundations as it's namesake model, dall-e, it's creating them entirely from scratch.
EDIT: Just looked into it. It is based on the original dall-e, so it is creating them from scratch. Here's more details: https://wandb.ai/dalle-mini/dalle-m...images-from-any-text-prompt--VmlldzoyMDE4NDAy
Last edited:
OMG, the things you guys have it make up is giving me life. I need to try it once I can get to a laptop, don't know how good this is on mobile.
Is this the same tech or different from the actual Dall-E AI project?
Because the examples I've seen out of "real" Dall-E absolutely crush what I'm seeing from dall-e mini, which has been pretty bad so far:
Because the examples I've seen out of "real" Dall-E absolutely crush what I'm seeing from dall-e mini, which has been pretty bad so far:
DALL·E 2
DALL·E 2 is an AI system that can create realistic images and art from a description in natural language.
openai.com
not much more advanced than NeuralBlender imo, but very fun to experiment with this stuff again. some stuff almost makes me physically ill because its so disturbing and "wrong"... how nightmarish can the stuff we post be? i wanna test the limits lol.
Some terrifying Godzilla designs here lol.
I genuinely dig bottom-row center.
Is this the same tech or different from the actual Dall-E AI project?
Because the examples I've seen out of "real" Dall-E absolutely crush what I'm seeing from dall-e mini, which has been pretty bad so far:
DALL·E 2
DALL·E 2 is an AI system that can create realistic images and art from a description in natural language.openai.com
Dall-E mini is based on Dall-E 1, not Dall-E 2.
Ahh that's pretty interesting. Still weird to me that it generates so much noise in its images versus deep fakes but I guess this would be partially that deep fakes had more time to mature and I guess this has less direct instruction to assist it.
OP
OP
UFO helping with the construction of ancient pyramids
The traffic error issues are real, I can see there's lots of comments on the site too.
That last one is straight 🔥
[edit]
UFO helping with the construction of ancient pyramids impressionist painting
😲
i feel combining humans with other organisms is the true horror material. i will have nightmares.
LOOK AT THEM
LOOK AT THEM
UFO helping with the construction of ancient pyramids
That last one is straight 🔥
[edit]
UFO helping with the construction of ancient pyramids impressionist painting
😲
Links are broken or my laptop is?
(Could go either way honestly)
OP
OP
The results are actually... not great at all. They look exactly like the other random free AI art generators I've tried, and full of just as much visual gibberish. Unless you give it a very simple prompt, or a prompt with very common elements, it's not going to give you anything worth saving.
PSA: f the servers keep dying, it is possible to run this locally provided you're familiar with using the terminal and you have a beefy PC. There's also an option for running your own cloud servers via Google or Amazon so you're not hitting the same machines as everyone else trying to use the app.
github.com
I got it up and running pretty easily on an M1 Macbook following the local development instructions, but there's no GPU acceleration on that platform yet so it takes forever even without using the hi-res training data. Getting it working on Windows seems to require building a library from source and ain't nobody got time for that, but the Docker instructions seem to work fine there too. The output always seems to be 256x256px though sadly.
GitHub - saharmor/dalle-playground: A playground to generate images from any text prompt using Stable Diffusion (past: using DALL-E Mini)
A playground to generate images from any text prompt using Stable Diffusion (past: using DALL-E Mini) - saharmor/dalle-playground
github.com
I got it up and running pretty easily on an M1 Macbook following the local development instructions, but there's no GPU acceleration on that platform yet so it takes forever even without using the hi-res training data. Getting it working on Windows seems to require building a library from source and ain't nobody got time for that, but the Docker instructions seem to work fine there too. The output always seems to be 256x256px though sadly.
here are the shroom dudes
Dude at the bottom middle is a fungi
Threadban me, I'll take it
OP
OP
a brutalist xbox shaped building drawn in inkA building shaped like a PlayStation 5 rendered in watercolours
