This is a linkpost for https://openai.com/dall-e-2/
Mentioned in
DALL·E 2 by OpenAI
6th Apr 2022
17Rabrg
7Nanda Ale
7mattyy
21Rabrg
1mattyy
7p.b.
24Rabrg
1p.b.
5p.b.
16Rabrg
3p.b.
5Nnotm
15Rabrg
1Nnotm
4Caridorc Tergilti
4P.
20Rabrg
1P.
3p.b.
17Rabrg
3p.b.
2Optimization Process
2ChristianKl
2alexlyzhov
4Rabrg
2Rabrg
1trevor
1Harry Taussig
1Dirichlet-to-Neumann
1Dirichlet-to-Neumann
1Kenku
1Rabrg
1Throwaway2367
17P.
0Caridorc Tergilti
3P.
1Caridorc Tergilti
7Donald Hobson
6Nanda Ale
6Amelia Bedelia
6P.
6p.b.
5gwern
2habryka
8gwern
3alexlyzhov
0Throwaway2367
-1JohnTheory
-1Erytmyst Sedai
-1GBG HNM
New Comment
50 comments, sorted by Click to highlight new comments since: Today at 2:18 AM
I can post a few more non-cherry-picked generations here, if there are some captions / prompts anyone would like to try.
edit: alright, I should try to get at least a little bit of work done today. might come back to generate some more later
Illustrated instructions for a playground game called "Less Wrong" - which will be featured in the TV series Squid Game 2.
An Ikea instruction manual showing how to build a superhuman artificial intelligence.
The "Game of Life" board game, updated for the year 2142.
A map showing Russia's planned invasion of Ukraine, with key objectives circled in red.
A still image of the long-delayed Harry Potter sequel, "Harry Potter and the Reloaded Revolutions" coming out in 2028.
How about "White to mate in 3 moves"?
Oh, and what I tried to create with some of the CLIP-guided notebooks, but it never worked out: "A painting of a man-o-war chasing down a smuggler guns blazing"
Ah, unfortunate. But thanks for trying. Maybe without the guns blazing?
Dall-E also wasn't brought up on a diet of chess problems, it seems. I probably should have specified to show a diagram. These boards and positions are positively wild.
The original DALL-E was capable of having almost the same image with slight variations in one generation, so I'd be interested to see something like "A photograph of a village in 1900 on the top, and the same photo colorized on the bottom".
Let's have fun with recursion!
A checkerboard where each square is itself a checkerboard.
A cube with mirrors on both sides, the mirrors show multiple reflections of the cube.
A person wearing a shirt with an image of that person wearing that shirt.
I'd be curious what happens for the prompt:
Book cover for "Using Credence Calibration for Everything"
If it's possible to specify height and width, with the width being three times the height to be used for my LW sequence. Otherwise, I'm still interested in how this more abstract concept gets translated.
"What if outer space were udon" (CLIP guided diffusion did really well, this is cherry-picked though: https://twitter.com/nshepperd1/status/1479118002310180882)
"colourless green ideas sleep furiously"
This is a great example of how even a single iteration on the prompt can vastly improve the results.
Here are the results when using your quotes exactly:
Pretty dreadful! But here they are, with the exact same prompt, except with ", digital art" appended to it:
Hmm.. the second iteration of the second prompt isn't able to be pasted in that comment for some reason. Here it is:
Mr. Bean makes a mistake. (or charlie chaplin)
Far side comic by Gary Larson.
Overlapping mirrors.
Mona Lisa parody.
Oldest House Threshold from Control
A painting of a steampunk city that looks like a mix of San Francisco, Chongqing, and Quya. (Optional: replace Quya with Saint Denis).
The Fighting temeraire by Turner The Mona Lisa by Vinci. Judith by Caravaggio.
I'm curious to see if it is able to remember famous paintings.
It seems they edited the paper, where now there is a close up of a handpalm with leaves growing from it, there once was a super saiyan sentient bag of potato chips:
Let's have fun with recursion!
A checkerboard where each square is itself a checkerboard.
A cube with mirrors on both sides, the mirrors show multiple reflections of the cube.
A person wearing a shirt with an image of that person wearing that shirt.
Some of the sample images were prompted with "teddybears doing AI research". In some parts of these scenes are notebooks that the teddies have written in. Fortunately it produced nonsensical blobs for letters, not valid AI research.
A big list of samples, helpfully compiled by a Vahyohw on the SSC reddit. The first bunch are non cherrypicked, it was every prompt Sam was posting, the the last bunch from various people seem to be added because they were interesting so are sort of cherry picked.
"A rabbit detective sitting on a park bench and reading a newspaper in a victorian setting"
"Dinosaurs playing frisbee in the throne room"
"An elephant tea party on a grass lawn"
"A human basking in the sun of AGI utopia" and a second attempt at that prompt
"a huge tree of life made up of individual humans and animals as its leaves" and a second attempt at that prompt
"Robot dinosaurs versus monster trucks in the colosseum"
"A painting inspired in Banksy's art showing a human-machine interaction."
"A warm fireplace on top of a building, in an apocalyptic city, steampunk style."
"Rabbits attending a college seminar on human anatomy."
"A shark and a dolphin cruise hand-in-hand with an undersea city in the background"
"Portraits of the same same person in different artistic styles"
"Transformers dancing at a rave on the moon."
"Androids dreaming of electric sheep"
"Human and AI fall in love and create the future children of the milky way galaxy."
"a wise cat meditating in the Himalayas searching for enlightenment"
"a robot hand painting a self portrait on a canvas"
"The streets of Chicago depicted in a manga aesthetic.
"What highways would look like if they were designed by goldfish."
"A surrealist underwater highway, with goldfish swimming on it."
"A young girl staring down a dragon, who is visibly amused."
"High quality photo of a monkey astronaut"
"A monster ice cream cone, digital art"
"a cool panda riding a skateboard in Santa Monica"
"rockets made up of paper balls"
"a surfing alien near a stone beach"
"The ruins of a long forgotten culture rotting away in a jungle"
"a tree goat checking her blackberry while promoting hinduism"
<Below here seems to be selected samples, so more cherrypicked>
"dragons and unicorns playing together in an enchanted forest"
robot holding shampoo bottle:
"robot holding a shampoo bottle at 45 degree angle"
"robot holding a shampoo bottle in one hand"
"robot holding a shampoo bottle in both hands"
"robot holding a shampoo bottle upside down"
"robot opening a shampoo bottle"
raccoon doing person things:
"a raccoon astronaut with the cosmos reflecting on the glass of his helmet dreaming of the stars"
"a raccoon wearing a hoodie working on his laptop late into the night"
"Artificial General Intelligence"
"A humanoid AGI successfully passing Turing test!"
"A shiny sphere in a modern house mirrors the surrounding room."
"Space Totoro" - contains actually coherent text (!!)
"image_type=illustration; wake up from the deep dream, artstation"
"A vacuum listening to music on its headphones while cleaning the room"
"A kid and a dog staring at the stars"
"A robot showing another robot its painting"
"A photo of a Samoyed dog with its tongue out hugging a white Siamese cat"
"A photo of astronauts dancing Greek traditional dances on Mars"
"An old photograph showing a superhero flying away. The superhero's face is covered with a mask"
I tend to have trouble evaluating this sort of thing due to cherry-picking.
Sam Altman made a twitter post; you can see 20 user-submitted prompts and their output at https://twitter.com/i/events/1511763146746212353, which might help a little if you want to build a model of the thing's strength.
We shouldn't be surprised by the quality of the images, but since this will become a commercial product and art is something that is stereotypically hard for computers, I wonder if for the general public (including world governments) this will be what finally makes them realize that AGI is coming. OpenAI could have at least refrained from publishing the paper, it wouldn’t have made any difference but would have been a nice symbolic gesture.
Wasn't ICML in January? I can't really think of any big conference deadline that just happened or is just about to happen.
I was thinking of ICLR apparently? I don't follow conf deadlines or procedures too closely but looks like ICLR is in the middle of getting the first review back or something. Everyone seems to time announcements around conferences (for PR, because of anonymity and quiet periods, or just finishing projects I guess), so I just assume if a bunch of stuff is suddenly being announced by a variety of groups, there's probably a conference involved somehow. Even when they aren't actually submitting a paper, I guess they want to be able to talk about it at the conference or have people talk about it there or were thinking of doing a paper? More active conference-goers than myself can speak to the exact mechanisms better than I can.
(On a meta note, I think that one unfortunate consequence of this is to distort perceptions of progress: people seem to saturate at 2 or 3 major announcements, and then the rest get ignored. So 3 or 4 times a year, for ICLR/ICML/NIPS, you wind up having long stretches of time where 'nothing is happening, deep learning has hit a wall!' and then a sudden flurry of announcements which wake people up and remind them that, no, DL progress continues to be rapid, but they simultaneously get overalarmed in the week or so the announcements happen - 'oh my god oh my god oh my god this is really happening isn't it' - and then underalarmed a month later when the massed repetition fades out and the urgency can't be maintained and now it's the new normal. If everything dropped more evenly, the most upset people would be less upset and the less upset people would be more upset, and on net everyone would be less wrong.)