More incredible Dall-E-stuff

an ai in the shape of a campfire telling stories to an audience of enthralled forest animals

Apr 22

Nachdem ich gestern das Posting über Dall-E 2 veröffentlichte, habe ich mich stundenlang durch die Outputs und Projekte mit der künstlichen Intelligenz geklickt und all sind gut bis wahnsinn. Jeder einzelne Tweet kostet zwei Pixelschubsen den Job.

Jemand, dessen Link ich leider verloren habe, hat Dall-E 2 als “Singularität für Kreativität” bezeichnet und mindestens für Illustration in allen Stilen stimmt das, unten findet man ad hoc game sprite Generatoren in game development Umgebungen, semi-kohärente Animationsframe, concerns about safety issues, and a ton of playful creative stuff produced with Dall-E.

Desto länger ich die surrealen, träumerischen Massen von Dall-E-produzierten Bildern betrachte, umso anders beurteile ich “reale” Fotografien und Visuals.

Die eigene Wahrnehmung löst von den Regeln der üblichen Realität, man erkennt in herkömmlichen Bildern die Möglichkeiten der träumerischen Bildwelten einer KI und es beginnt eine Entgrenzung der eigenen symbolischen inneren Welt. Unsere mentale “Landkarte der visuellen Möglichkeiten” wird gesprengt und um synthetische Unknown Unknowns in allen Variationen erweitert.

Für Menschen wie mich, die auf dem Spektrum zwischen aphantasia und hyperphantasia eher zu einer lebhaften visuellen Vorstellungskraft neigen, ist Dall-E 2 wortwörtlich mindblowing, oder anders formuliert: bewusstseinserweiternd.

More thoughts and links

Official Dall-E Instagram-Account, Dall-E 2-Hashtag on Twitter and the Subreddit, there: Request Thread 1 (closed now) and Request Thread 2.
Ben Barry curated a book of 1000 robot paintings generated by Dall-E 2
“DALLe is for personal, non-commercial use. No NFT with #DALLe“. (However, Dall-E is by far not the only art generating ai, albeit the most advanced for now.)
"It would be very interesting to preserve photographically, not the stages, but the metamorphoses of a picture. Possibly one might then discover the path followed by the brain in materialising a dream." (Pablo Picasso, 1935) —> Following Picasso's deconstruction with #dalle
“Kinda cool that any tweet can now be illustrated” - Instant visualization of any tweet, including the violent fantasies and attacks of vicious trolls. Here’s a thread about safety concerns with an AI with learned biases, which can imagine anything and here’s a doc on Github documenting “initial findings on potential risks associated with DALL·E 2”.
“dall-e 2 illustrations of my friends' twitter bios” featuring a “Chief Meme Architect”, a “feeder of kittens and reality engineer”, a “bookbear” and this interesting misspelling of “following me”.
Yes, Dall-E imagines Elon Musk smoking a joint through his nose.
“a baby fennec sneezing onto a strawberry, detailed, macro, studio light, droplets, backlit ears”, here’s all 10 renderings.
Twitter-Threads: @BecomingCritter / #DalleMerz / TylerAngert / Sam Altman / @E0M 1 & 2 / #dalle by @hardmaru (some favs below)
Here’s a video featuring a Dall-E 2-demo with a lot of weird and beautiful examples, from renderings of Berlin and Zurich in the style of MC Escher and whatnot (german).
"Sisyphus as a happy man according to Albert Camus"
Károly Zsolnai-Fehérs 2-Minute-Papers Video featuring people holding on to their papers in the style of Spitzweg.
Notice how “whimsical font design” does not produce an incoherent mess, but a near full alphabet of real letters. It’s missing the G and K and fuses T and U for some reason, otherwise it’s coherent. Typography is a big issue for computer vision and the progress Dall-E makes in this visual subset is remarkable from previous iterations of the technology.

Read this comment about the image above on Reddit: “Just looking at this is fucking me up, I need to keep reminding myself that there isn't any intentional symbolism anywhere here and it's just an output by a machine. Every little scene reminds me of the stresses and anxieties of modern life.” — If this is the case, if every little scene in this AI-generated artwork reminds you of stresses and anxieties of modernlife, then this is real, relevant art. “If you can’t tell the difference, what does it matter?”
Please notice the small bulb in the lower mid left section and the fractal loop going on there, with scenes containing scenes ad infinitum.
The Dall-E images are like zooming into the hallucinating google ai from back in 2015? 2016? the dog nose ai where you zoomed in and always new dogs and dognoses emerged everywhere, but it just looses to fixated point (the dog/nose) and can imagine everything. when the first AIs were caught in a dognose-cage, dall-E is free.
Amjad Masad integrated Dall-E 2 into a game dev environment and prompts game sprites on demand (tweet featuring video below). This alone costs 3 gfx jobs in every game dev studio.
It’s interesting to analyze the mistaked dall-E makes in generating pixel artworks for the 2D-game assets generated by David Schnurr in this thread. Dall-E 2 seems to add diagonal 45° triangles and work with different, scaled and layered pixel sizes which also seem to be recursive, if you look at some details. The “underground lair pixel art game scene” seems to feature a wooden ladder and wood patterns, which seem to recursively feature game level designs, if you zoom in.
Someone requested “Running animation keyframes” - “Ha, it kind of worked”