TITAA #61: Surrealism, Alice & the Occult
Exhibition - Video Gen - Transparent Ones - Unloved Poets - Firehose Vis
Lots of material here, it was a solid month of necessary escapist art viewing, reading, and playing. Kicking off with surrealism, after a visit to Paris.
TOC (links on substack page):
Intro Article: Surrealism at Centre Pompidou
AI Art News (Video, 3D, Misc AI & Procgen Web)
Narrative & Creativity (AI Poetry, Writing, Icky)
Games News (Including AI-related)
Surrealism Exhibition at Centre Pompidou
Surrealism is hot right now. The Centre Pompidou, Paris’s strange looking modern art museum, has an excellent surrealism exhibit open till mid January, with a spectacular large exhibition catalogue. Synchronistically, there are a couple surrealism exhibits in England as well, written about here in the Guardian. “Forbidden Territories: 100 Years of Surreal Landscapes” at the Hepworth Wakefield sounds especially good. Centre Pompidou is shutting in September 2025 for 5 years for asbestos repairs, and their general collection is outstanding, so visit Paris soon.
Where did surrealism come from and what were the principles? It’s rare that an artistic movement has a/many spelled out manifestos, but a young Andre Breton wrote the first of several Manifestos of Surrealism (compiled here in english) in 1924, in a Europe coming out of WWI. The movement was an explicit step away from the “real” grimy world of fascists and wars, into the world of the mind, accessible via tricks and techniques like automatic writing, occult practices, altered states of mind, a search for uncanny juxtapositions. But rather than just escape, Breton proposed an acceptance and merging of the rational and irrational: more than real, or “surreal.” Some of Breton’s inspiration was a recall of childhood’s wonders and weirdness; the Pompidou exhibit likewise celebrates Alice in Wonderland, first discovered and applauded by surrealist Louis Aragon. I especially loved some of those curated for this section:
Here’s a picture of 14 year old poetess Gisèle Prassinos, an Alice of theirs, reading her automatic writing poems to a bunch of male surrealists (photo source not the exhibit and slightly weird):
Try a sample of one of her poems from the podcast transcript on the Pompidou site shown on left, with a Claude translation on right, and think about it when we get to the AI poetry article in the Narrative & Creativity section.
Other thematic rooms in the Pompidou exhibition include “Night”, “Forests”, “Melusine” (the French fairy serpent) and chimeras, “Dreams,” mediumship, alchemy and the Philosopher’s Stone, the erotic imagination, the Cosmos.
The role of chance and automation was marked, too, relevant to fans of procedural generation. Max Ernst, for instance, used rubbings (“frottage”) of wood to inspire paintings laid over the rubbing, a kind of “image to image” inspiration with his materials (e.g. here). Collage, a reworking of images in pieces, was another one of their procgen-like methods. I was particularly impressed by Ernst’s collage work.
The Exquisite Corpse game, the practice of blindly building on a previous drawing output (often with simple rules) might remind us of a simple LLM co-creativity session—like AI collage, or my notion of “creative RAG” perhaps. The goal for these surrealist techniques was to entertain odd juxtapositions that would open up the irrational mind and feed creativity. I am certain the surrealists would be entertained and bemused by crude base model LLMs as tools—not as replacements for humans, but ways to unearth uncanny ideas in collaboration. They would have loved the bizarre video output and early image models that do not reproduce human world physics, geometry, or logic.
The entertaining picture below crossed my eyes on X via hinterlander; but I’d say the surrealists are somewhere in the purple “human experience” snake head, trying to reach towards the reality on the right. Here’s a good piece on Breton’s concept of “The Great Transparent Ones,” living there in “reality.” “One can go so far as to believe that there exist above him, on the animal scale, beings whose behaviour is as strange to him as his may be to the mayfly or the whale.”
A good segue: magic and occultism (seances, ouija boards, Tarot…) were practices that interested the surrealists as ways to unlock extended perception. But few were as serious about studying the occult as Ithell Colquhoun. She also invented and used a lot of automatic techniques in her art, including “fumage” — using smoke residue on paper as inspiration. Colquhoun was a practicing alchemist and magician. But surrealist groups could be a bit cliquey and controlling. Colquhoun considered herself a surrealist, but she was ejected from the British surrealists’ society in 1940 for refusing to quit her occult societies. Louis Aragon was likewise ejected by the French group after visiting the Soviet Union, for political reasons. Nevertheless, the Pompidou exhibit really leans into the occult and alchemical surrealists, especially the women: they sell a small book, “Les Magiciennes: Surréalisme et alchimie au féminin,” with extracts from texts by Leonora Carrington, Colquhoun, and Remedios Varo. I may write about it in my mid-month esoterica section!
The curators did a great job at showing the diverse growth of surrealism beyond the boys’ club of Breton’s start, those men watching Gisèle read her poetry. The Guardian: “As Mark Polizzotti has underlined in his recent book Why Surrealism Matters, the focus has moved away from the original surrealism bros to ‘non-western surrealisms, gender-fluid surrealisms, racially diverse surrealisms.’” Breton’s own interests included world-wide ethnographic study of magical arts and artifacts, as you can see in his atelier preserved behind glass in the Centre Pompidou (featured in a header post here, and with a large detail here.)
Incidentally, one of the Magritte paintings from “Empire of Light”—which covers one half of the catalogue’s cover and was on display—just sold at Christies for a record $121 million. I admit I don’t understand how it could both be on display in Christies and at the museum; it looks like the same piece. But if any painting could be in two places at once, it’s a Magritte.
Magritte says of this work, “The landscape evokes night and the sky evokes day. I call this power: poetry.”
AI Art News
Video
New updates in Luma’s “Dream Machine” — keyframes and storyboarding etc. Here’s an example (x) that I liked by Jerrod Lew because infinite mirrors.
LTX-Video is a new fast open source model—not super high quality, but fast! real time, it says—which is being integrated everywhere (fal, replicate, comfyui workflows, etc.) E.g.: demo on HF, and the original release Lightricks/LTX-Video: Official repository for LTX-Video. It has a nervous breakdown with anything surreal I load in for image2video, which, fair enough:
DreamRunner — a story planning video framework releasing code.
Runway.ml has added video outpainting, allowing you to reframe a video. And a new image gen model, Frames.
Can’t lie, I liked this: Artists claim to leak OpenAI’s Sora AI video model in protest - The Verge. Actually they leaked a way to use the API which was then shut down. “ARTISTS ARE NOT YOUR UNPAID R&D ☠️ we are not your: free bug testers, PR puppets, training data, validation tokens.” I approve this message. Also many folks who looked at the samples think it’s not that hot compared to other tools we can all use now (e.g., Minimax, et al., see my past newsletters).
You can do Lora fine-tuning of a Mochi video model, in their examples.
Entertaining (no code, since Adobe): MultiFoley, do audio generation by text prompt to match motion of video.
3D
https://railwaydesign.xyz/ via Tom Scott. Drag little trains in place. (Also check this cute SVG train clock!) Er, this is not AI, sorry - really belongs in the next section, but is 3d.
The Matrix — infinite driving game gen. Another one aiming at real-time games worlds.
The Bluesky firehose in 3D, kind off.. via Matt Muir. See below in Misc for more firehose work championed by Matt.
There’s been a ton of great splat news and 3d gen models, but I declare time bankruptcy for today, sorry.
Misc AI & Procgen Web
A pretty intense manga creator/editor app built on top of a lot of open source tools (SD web ui, comfy, etc).
Interesting slider for amount of detail, real-time change with the slider, Omegance:
There’ve been a lot of single image or few image customizer/edit releases recently (including QuillMagic). I was surprised by how good Omnicontrol was for this kind of “product” photo — on the left is one I took in the Surrealism exhibit. My prompt is below. (It was crap at putting a person in front of a wall, though.)
⚾️ Endless AI simulated baseball commentary. (Maybe via Golan Levin.)
Flux Text Inpainting — give it a style source, it’ll replace the text. It’s very good.
A Plotter-Oriented SVG Exporter for p5.js from Golan Levin.
Generating random mazes with Javascript, via Gorilla Sun newsletter.
FlipSketch (demo) — another animation of SVG sketches. Demos are a bit funky/difficult though.
IMG_0001, by Riley Walz. A random selection of videos uploaded to YouTube with default IMG filenames between 2009 and 2012.
Silent Poems — via Clive Thompson. A text experiment with javascript animation. Every letter a connected symbol.
Other Bluesky firehose visualizations via Matt Muir (of webcurios): the emoji one, the images one, Bluesky Mosaic, which is really great but also very NSFW and porny these days. I forgot to save the swearing one which was more SFW.
Satellite photos of Middle Earth, via Kottke.
Narrative & Creativity
Can Humans Tell It’s AI?
The Human vs. AI Poetry debate: A piece in Nature (one critique in The Conversation) showed various things about human non-experts judging poetry.
“We found that AI-generated poems were rated more favorably in qualities such as rhythm and beauty, and that this contributed to their mistaken identification as human-authored. Our findings suggest that participants employed shared yet flawed heuristics to differentiate AI from human poetry: the simplicity of AI-generated poems may be easier for non-experts to understand, leading them to prefer AI-generated poetry and misinterpret the complexity of human poems as incoherence generated by AI.”
And other things, like humans don’t like AI generated poetry if they know it’s AI generated, but are much more generous when they think it’s human. It makes a bit of sense—people infer reason rather than random from human intent (unless you’re a surrealist doing automatic writing?!), and don’t look for meaning in the AI output (surrealists might, though!). Anyway, there has been a lot of discussion, and Colin Fraser on X did a bunch of graphs from their data if you’re willing to look there. The least loved human poem was by Dorothea Lasky, and I like it. I used it as the closing poem.
And then there was the AI Art Turing Test. Similar results, people aren’t good at this, when there are styles involved. Of the 11000 people who rated images: “Your instincts were worst for Impressionism; you identified every single Impressionist painting as human except the sole actually-human Impressionist work in the dataset (Paul Gauguin’s Entrance To The Village Of Osny).” (Note: I’ve seen academic critique of the study design, so ymmv.)
Writing
The Playwright in the Age of AI, in the Atlantic. “In McNeal, he not only suggests that LLMs might be nondestructive utilities for human writers, but also deployed LLMs as he wrote (he’s used many of them, ChatGPT, Claude, and Gemini included). To my chagrin and astonishment, they seem to have helped him make an even better play.”
AI as Humanity's Salieri: Quantifying Linguistic Creativity of Language Models via Systematic Attribution of Machine Text against Web Text — tl;dr LLM models show reduced creativity: “Experiments reveal that the CREATIVITY INDEX of professional human authors is on average 66.2% higher than that of LLMs, and that alignment reduces the CREATIVITY INDEX of LLMs by an average of 30.1%.” We want base models.
Terminal-velocity, a novel created by 10 AI agents, via Justine Moore (with a repo of code with agents and structure breakdown.
Supposedly the Midjourney storytelling team (Max K et al) will be demoing all the stuff they've been working on during office hours in mid December. Not sure how that works with audio office hours, or has their format changed?
Icky AI Book Gen
The Bookseller: New publisher Spines aims to 'disrupt' industry by using AI to publish 8,000 books in 2025 alone. Not pleasing anyone. “The company charges up to $5,000 a book, but it can take just three weeks to go from a manuscript to a published title.” Their site… weirdly low key on the AI part, actually and at least one strangely empty conference room pic.
Some sarcastic sniping at this Bookengine.xyz project which is not super serious in their trolling page (but they’re charging money anyway):
“Choose Your Source Material: Plunder literary classics or cult favorites to offend their fanbases.”
“Customize Your Chaos: Let AI suggest plot points, or input your wildest ideas for maximum controversy.”
“We use predictive modeling to take your sequels in directions that'll make people say, "Why though?" You won't believe it-but isn't that the point?”
Brief Games News/Links
“You Exist In The Long Context” — Play Steven Johnson’s NotebookLM game from a book demo, which is embedded in the piece. Evidently the NotebookLM team gets a disproportionate number of people wanting to use table top roleplaying game books as the basis for interactive games with Gemini (Google’s flagship LLM). They are now working on ways to make this easier. But how about letting us add UI instead of just using a linear chat window? Anyway: People interested in this might also like Aaron A. Reed’s piece on how TTRPGs relate to procedural games, via his 50 Years of Text Games substack site.
Papers up from Proceedings of AAAI AAIDE conf 2024. One is:
NarrativeGenie: Generating Narrative Beats and Dynamic Storytelling with Large Language Models — at AAAIDE (links below). “By leveraging LLMs for reasoning and generation, NARRATIVEGENIE, translates a designer’s story overview into a partially ordered event graph to enable player-driven narrative beat sequencing.”
And I just glanced and saw: A Model for Automating the Abstraction of Planning Problems in a Narrative Context by Mira Fisher, Stephen Ware.
How to Become a Game Writer, a Greg Buchanan tool.
Storylet implementation in Ink (Inkle’s language for writing interactive fiction).
Retro No More: Interactive Fiction of the Early Comp Era The Digital Antiquarian — featuring a lot of Andrew Plotkin/zarf.
The Twine Cookbook can be bought as a paper book, too.
Pixel Art Knowledge - Isometric games don't exist by Thomas Feichtmeir "Cyangmou" — A free guide on perspective & projection for videogames
NLP & Data Science & Vis
spacy-layout: 📚 Process PDFs, Word documents and more with spaCy (yay)
A good post from Yoav Goldberg: What makes multi-agent LLM systems multi-agent? And HuggingFace’s page on agents.
SPARQL wikidata queries from text prompts, via Wiki Research.
A new UMAP article co-authored by Leland McInnes.
Foursquare Open Source Places: A new foundational dataset for the geospatial community.
A radio map of the world, interactive (h/t DreamingTulpa). This is fun!
Book Recs
This is immensely long, as usual… I’m just sticking the book titles here, and you can become a supporter to see the full books, tv, and games mailing (there were a lot of great tv shows and a solid play of Alan Wake 2 this month). The books featured class warfare, janitors, and domed cities.
⭐️ The Last Murder at the End of the World, Stuart Turton (sf).
⛓️ The Practice, the Horizon, and the Chain. Sofia Samatar (sf).
🪄 Blood Over Bright Haven, by ML Wang (fantasy).
🎭 Daughters of Chaos, by Jen Fawkes (fantasy).
⭐️ The God of the Woods, by Liz Moore (mystery/thriller).
⭐️ Bluff, by Jane Stanton Hitchcock (mystery/thriller).
Poem: “Beautiful World”
Please tell me that I was a good child And that I did everything right And that the atmosphere was exactly certain I want you to love me In ways that you never have So that I become a forgotten world With rainbow sunrises over dark green trees And the cooling of the day Becomes normal again We will sit and watch the body of water That we once called a sort of death You know even in my dreams You say I’ll never get it right This is not a dream We are burning here with no escape But no matter how many times They talk about the moon It does not take a poet To know that the moon Is still only an illusion Only an illusion The moon calls out to all of us Come back, it says But we don’t hear it Already on our way To somewhere
This took a while to write but still feels brief, for some reason. I’m swamped as I wrap up for a conference trip! Drop me a note, it’s nice to hear from you.
Lynn (@arnicas on Mastodon, Bluesky and Threads, but mostly Bluesky)
Well first of all: thanks for the great newsletter :)
And surrealism, my favorite "movement" and one that I spent a lot of time with in the last months. The exhibit sounds fantastic, won't make it to Paris until January, but will hopefully catch it in Hamburg next year! Did you get the exhibit book? I'll probably order it ahead.
I totally agree the surrealist would have loved the liminal space of AI art, I always thought the stuff we did with Disco Diffusion about 3 years ago was especially uncanny, giving lots of room for interpretation. And, at the art biennale 2022 there was Slovenia artist Marko Jakše, you might wanna check his work out (https://www.labiennale.org/en/art/2022/slovenia-republic)
The "Infinite Driving" game sounds scary, and their website is ... ahm, retro. But what a wild concept.
Then I read "The Playwright in the Age of AI", reads like a weird play in itself to be honest. Maybe because I'm not in the theater bubble.
Ha, and the novel written by agents - I'll take a look, but its really close to my concept for NANOGENMO, only I have it generate screenplays. It's not done yet, I'll let you know when I put the code on Github. (Ok, this looks about one order of magnitude more complex that my project, and I just use Claude API)
And re radiomap - I used radiogarden for a while, think it's pretty much the same, plus they have a cool app.
The Last Murder at the End of the World sounds fun. I just finished "Tomorrow, and tomorrow, and tomorrow" which was quite fun, albeit sometimes a little too much "self-inserting" by the author. Oh well.
Finally, do check out "Mirrormaze" and "Somniscope", which are 2 short story anthologies from the Dreampunk genre. I have a story in each, but I think you might enjoy the variety of stories in each. Jeff Noon ("Vurt") wrote one for Somniscope.