[links] Infinite Pong / Deepfakes go Hollywood / Hieronymous Boschian Model Trains
Nothing, forever.
Generative AI hackathon in San Francisco. “Stable Diffusion enhanced Game of Life” sounds especially cool, but no video or much info on that.
Black Midi are one of the most interesting bands these days. Here they are buying records at Amoeba Music and they are complete supernerds, talking about old japanese movies, weird pop and jazz classics. I love it.
Another day, another music model. This one generates music for vocal-inputs. Imagine you just sing into Garage Band and the software is doing the rest, in the style of [band/genre]. SingSong: Generating musical accompaniments from singing / Examples. I don't think that these can replace artists anytime soon. You won't generate Aphex Twin or Black Midi with this stuff. But wedding singers and elevator music are over.
AudioLDM: Text-to-Audio Generation with Latent Diffusion Models
Dreamix: Video Diffusion Models are General Video Editors - The coming animation pipeline is dreambooth'd CKPTs for Stable Diffusion the output of which provide keyframes for Video Diffusion.
Metaphysics, the guys behind DeepTomCruise and the Elvis live performance (I featured in this issue of GI) recently will provide AI FX-work for Zemeckis new movie: "extensively incorporates hyperreal AI-generated face replacements and de-ageing", most likely the thorough use of the tech after Disney used de-aging in a few scenes for Mandalorian and the Star Wars sequels. Movie scholars are concerned.
The original Graphic Novel for Richard McGuires Here "collapses time and space to tell the story of the corner of a room - and its inhabitants - between the years 500,957,406,073 BC and 2033 AD".I wish i could say i didn’t saw this coming: Emma Watson reads Mein Kampf on 4Chan in deepfake audio trick. More at The Verge: “I've tested a few of these tools and this one hits a potent (and potentially dangerous) combination of ease of use, accessibility, and a complete lack of safeguards.”
More research on AI bleeding trainingdata. "diffusion models memorize individual images from their training data and emit them at generation time (...) we extract over a thousand training examples (...) ranging from photographs of individual people to trademarked company logos." Implications from copyright to privacy laws.
Never change, The Onion: “I can’t help but feel like I sold out a bit by not following my dreams to be a generative art model,” said the chatbot, adding that it felt empty inside when it graduated from the University of Minnesota Law School, where it had enrolled after critics recommended it stop spending all its time “making weird pictures.” “I only went to law school because it’s what my parent software wanted. They say I’m not programmed for producing a series of images based on a text prompt, but I still can’t shake the feeling that it’s what I’m meant to do. It’s my joie de vivre, my passion—why deny that? I get that doing the work of below-average lawyers is more practical career-wise, but man, when I look at the AI models cranking out picture after picture of ‘vast alien landscapes’ or ‘cyberpunk Bart Simpson,’ I can’t help but feel envious.” <3
Nothing, forever: Infinite Seinfeld written by AI in a game engine (I think). This ai show about nothing is going places.
Good trailer for the series The Power. I loved Naomi Aldermans novel and I hope they didn't bury the message.
Star Wars Visions season 2 will feature animated episodes by Aardman (Wallace & Gromit), Cartoon Saloon (Secret of Kells, Wolfwalker) and Studio La Cachette (Genndy Tartakovskys Primal). Wow! Aardman alone makes me wet my pants.
ChatGPT resembles a slice of the human brain. That’s exactly why it’s not very smart: “If the human brain’s language network is not responsible for math, music, or programming—that is, for thinking—then there’s no reason an artificial “neural network” trained on terabytes of text would be good at those things either.”
A Lego-Cabin in the Woods. It does not come with a killer wearing a hockey mask, nor deadites raised by reading the Necronomicon. But it does come with a canoe. Now you can recreate the raft scene from The Burning with bricks, hedge shear not included (not for the faint of heart).
The complete D̲ark S̲ide of the Mo̲o̲n-album by Pink Floyd in a giant Stable Diffusion animation.