At the tail end of the umpteenth ("énième”) heat wave in France, when it’s still too hot in my apartment and will be until the heat death of the universe, I can’t think of anything I want to do less than take a weekend to write up the last 2 weeks. Also, I started playing Urbo on my steam deck and have 71K points for my current city and want to get back to that. So this is a terse one and even a little ranty on one cluster of links, blame the heat.

TOC — contents after a few intro comments on travel and an interactive fiction / gadget jam.

• AI Creativity (Video, 3D / Three.js / Worlds)

• Web Misc Fun

• Games (Mods & Fanfic, some ranting)

• Narrative Generation

• Weird Stuff (including LLMs being weird but not just; also fairies)

• Data Science / Infovis / Tooling

• A Poem about August

Having reached my limit on these heat waves, in the past two weeks, I’ve been in two different rental apartments to get a break — one was the Alps, near boutiquey Megève, seen above (no A/C but a good nightly temp, and 5 degrees cooler than home). The second was a new apartment with excellent A/C in a very old town that had almost nothing in it… except some friendly neighbors and old stones. (No, I saw neither the eclipse nor the Perseids. Lol, google that right now.)

One callout above the fold, my friend Tom Granger is doing an e-ink-Ink (interactive fiction writing tool) give-away for Inkjam, to go with his new eenk library, which just started. He’ll gift an xteink device (he doesn’t work for them, he works with me) if you win and otherwise, get your own and try running ink / intfic games on your little xteink. (I personally have loaded mine with books for the supermarket wait, including a challenging Jules Verne in French. But I am going to try his tools this week, on vacay, finally, I swear.)

Cutting to the news, which includes: 👉 Seedance 2.5, 👉 threejseval, 👉 3d splats becoming even more pervasive/easy, 👉 WorldClaw, 👉 Toolcraft tools, 👉 game mods and AI, 👉 fandom, 👉 friendslop games, 👉 organoids, 👉 quantum decisions, 👉 OCR contest on FineBooks, 👉 public domain images, 👉 words we teach English learners, 👉 more backrooms, 👉 agentic collaboration, 👉 persona bench, 👉 can LLMs stick to a (changing) narrative script, 👉 Unsloth desktop, and more.

It means a lot to me if you pay for this, it’s a shitload of work! I am taking a break the last 2 weeks of August, not sure if/how much the end of August will have, although definitely the subscribers Books/TV/Games, which is my favorite thing to write.

Read more

Ah, it’s hot again. I’m sulking, and should be going on vacation, but… might as well work if it’s too hot to be outside? Anyhow, if you are reading this, please drop me a note… I hear from very few readers and it bums me out to spend my weekends on this and send it into a void. Just hit reply? Only goes to me…

TOC:

• AI Creativity Tools (Image Gen, Video Gen, 3D & “World Models”)

• Web Misc Fun

• Games Related (always lots of game related in 3D too, btw)

• Narrative Generation

• Data Science / Infovis / Programming Tooling w/ AI

• A Poem

AI Creativity Tools

Image Gen

"Drawing" the Mona Lisa with GPT-5.6, Claude, Gemini, and Grok (via Geeky Animals) — giving a drawing tool to the models to test their output ability, both at copying something and from a text prompt. The harness is open sourced at github.com/hershalb/canvas-arena. It really is too bad that models have to take pics at every step in a visual workflow…

The drawings from scratch are mostly much better, although Grok continues to be insane. 5.6 Sol is really good at those.

PixelGPT Canvas (via Matt DesLauriers) — a pixel-art / sprite generation canvas and a model in the browser. This is fab — not great art, but I really love my moth here:

Research: CompArt — training for “aesthetic alignment” in text-to-image generation via principles of art. Great idea… research paper with small visible samples. Also see Trinketry — “tracing and recombining visual elements in generative design exploration” (a Max Kreminski paper).

Video Generation

This is definitely the most exciting category of the last 2 weeks, from my angle! Two new models, soon to also have weights released, from Minimax (H3) and Black Forest Labs (FLUX 3). And this coming week, Seedance 2.5 should be out, which by early reports is amazing, generating long sequences; although also reportedly very expensive (twice the cost of Seedance 2?) and not open weights.

BFL’s Flux 3 early access discord felt a bit like early Midjourney for a minute there — lots of familiar names! I tested it a bunch on things I’ve been wanting to do at work (with Veo and Omni) and found it terrific. It can manage animation of styles in artwork and pays attention when you say “no cuts.” Here’s a gif where it sticks to the artwork style (Errol Le Cain) admirably:

I shared another animation of a painting, on X. Here it is doing a complicated prompt, turning a painting into a sepia photo and then a “real” scene:

Hailuo 03 (Image to Video link to Fal) — MiniMax’s image-to-video model can be tried on Fal via API and a lot of other providers including OpenRouter; also see the MiniMax video-generation API docs.

Here’s a comparison of a Flux3 (left) with H3 (right) animating a photo I took in Lyon, turning the mural alive with "the people moving” and the “cat climbing” and the “books falling.” Note how Flux 3 stays consistent with the style but arguably H3 was better at the positions from the image (the cat is actually a small figure on the right side):

And some unsurprising reading: About 300 Netflix Programs Have Used Generative AI This Year (Variety). Given the choice of not having a scene vs. having it, by cost of animators vs generation, I’d rather they had the scene. Especially since I like big space opera tv and those shows always get canned for costs.

3D and “World Models”

Grace Cathedral (by Vincent Woo) — Gaussian splat of the cathedral, a really big one, navigable. With annotated tour points. Lovely use of splats (and Playcanvas).

InfiniSplat — a Gaussian-splat creation space on Hugging Face, very fast, single image input: it will give you a kind of quick cheap 3D effect. Great strides have been made here, honestly, for this to work at all.

World Models: ABot-World (Reactor) — an interactive world model, also live as a Hugging Face space where you can upload a picture and have it go off piste— I uploaded this screen cap from Dungeons of Hinterberg

And then panned right past the church (see left shot) and then turned right, and I wasn’t over the water anymore. Still very interesting to see it run so quickly in the browser. Look, you don’t need to pay for a Genie 3 sub!

And see world model Wonder (coming from Adobe, and it says coming to Hugging Face too?):

PanoWorld — research on real-world panoramic generation, from the Insta360 research team, some kind of demo coming — it looks like it will do animated 3d (i.e., 4d) generations.

Opus 5 is not getting great vibe reviews for coding behavior, but for three.js and 3d content, it’s evidently doing really well according to the comps I’ve seen shared. OTOH, Ethan Mollick continues with his eerie Fable creations, including this one: "Make a game about Imminence. Something very big, very strange is happening. A suburb & the arrival of a vast & unknowable presence. Not horror, invoke the feeling of the end of all things coming, inevitably, but also not sad or scary." I liked it. Try it: https://the-imminence.netlify.app

He also had Fable do a Piranesi-like city building tool in similar style (code base). (Compare to the busier, less artistic GPT Sol version here.)

His Cezanne-style Fable city builder is also adorable:

Three.js tools and code:

BasicProceduralBuilding — a procedural building generator (three.js, procgen); Blender-capable.

sakura-crossing — codebase for an explorable Japanese suburban railway-crossing neighborhood on a small planet, rendered 3D-to-2D as a cel-shaded anime background. Three.js, no image assets, evidently made by Opus 5. Very pretty, with an article about how the 3d to 2d effect was done.

Web Fun / Misc

branch — a collaborative collage of tree and branch photos, since you know I love collage.

Jules Vernacular (via Matt Muir) — photos of letters/signs/fonts in the wild in France. I love this so much.

Richard Silver Photo: Vertical Churches (via Kottke) — vertical panoramas of church ceilings. Pricey to order, but very interesting to look at.

Maps & Travel Routes from Books & Films — “Reading Maps plots real-world journeys from classic novels and films - on an interactive atlas.” Via Recommendo and Mark Frauenfelder. Dracula:

STG v.boxSquad (via Waxy) — from the Space Type Generator guy. Hahaha, you can position them, and turn on twerk.

doodad gallery — a UMAP-clustered art gallery; add stickers as they’re discovered. You have to open a tray at the side and drag a sticker onto it. A way of encouraging exploration? Not entirely sure how it plays out over time.

Legibility of Effort — “LLMs have broken the legibility of effort: our ability to tell, at a glance, how hard something was.” For a human. Does is matter? It seems to for “art” definitions, but I have never liked that kind of definition. (Also, I think many artists have had the experience of killing themselves to get something out, only to have it fall flat, relative to something they found much easier, even trivial to produce. The audience doesn’t care?)

Tech tools:

• aval — a new open-source format for interactive video on the web, with a built-in state machine, frame-accurate transitions, and packed-alpha transparency.

• Hubble.md — markdown, with a comment feature. This is a great idea for input to LLMs and notes to self, too.

• JarvisHub — an open harness for canvas-native multimodal creative agents.

Games-Related

How to Run a Gauntlet Loop: The Prompting Method Behind Claude of Duty
games - now X famous, from Matt Shumer, for getting good if expensive results from prompting games with top LLMs. Many examples being shared on X of web games made with this method. “The prompting method behind Claude of Duty. Give the agent a bar it can’t talk its way around, let it split the work, and never let the builder grade itself.”

Murdoku — online and printable logic grid puzzles that are cute and visual (sudoku like). Via Installer newsletter.

ELO 2026 Online Exhibition — the Electronic Literature Organization’s online show (text art, procgen, interactive). Games and more… there’s a lot there, and have I had time to dig in? I have not. Tiny corner of the page:

Academic/research things of note:

From Sakana, Dream-Cubed — generative modeling in Minecraft, with a codebase. The examples look great.

Playable Archives — balancing historical accuracy and player freedom via RAG-enhanced AI game mechanics (at ICCC, which I need to catch up on). An approach to cultural heritage games. “This work presents Playable Archives, a system where players experience Grounded Imagination, exploring counterfactual history while staying anchored to authentic constraints, and contributes Playable Archives as an approach for cultural heritage games that shifts engagement from passive memorization to active, speculative inquiry.”

Narrative Generation

Lots of interesting action going on in both the study of AI writing and the actual use of it. Every day there is another article about a fantastic book that had a contract and then agents or publishers dropped it because it might’ve included some AI text. Even if it was an excellent book. Even if it might’ve been a false positive from a detection app.

⭐️ Generative AI floods and dilutes the market for books — A recent study, lead by Tuhin Chakrabarty, full-text AI detection across 14,419 self-published Amazon genre-fiction books (2023–2026): AI-heavy books are a large share of the catalog but a smaller share of sales, yet they’re winning a growing share over time and taking more top-rank slots. They are selling, and some people are making serious money with them, too. But overall, given the flood of books, revenue per book has declined across all categories.

In the top earnings, romance subgenres and writers have a pretty clear edge —

There are a lot of interesting charts in this paper, and it’s been widely summarized. Worth a look. And the following digs into it, naming names:

Is This What Comes After AI Slop? (The Atlantic) — on the Daggermouth bestseller (a legit book of word-of-mouth and discussion, despite the “tell tale signals of AI”) and more on the study above.

“If you can get 10,000 people to read 10 pages of your book, and you have 300 books you’re putting out,” she said, you can make real money. “But you’re not actually writing romances people want to read.”

⭐️ Writers Agree: AI Isn’t a Novelist. So Why Is This Author Interested? — “Naomi Alderman, best-selling and award-winning, thinks AI writing is terrible. She’s fascinated anyway.” I think you have to be already quite successful to be able to go on the record and admit this, sadly. And she says lots of people want to kick her out of her friendgroups.

I am a novelist and I’m interested in AI. Since I started playing around with it, I’ve felt an expanding sense inside my mind of what “being a writer” might mean in the future. I feel afraid and I feel exhilarated and I want to know what’s coming next. … I am absolutely delighted to do my own experiments, and explore that jagged edge: the places where it entirely fails, the places where it is unexpectedly brilliant.

This, incidentally, is a good sentence: “The whole point of creative writing is that I am trying to come up with sentences that cannot be predicted.” She is interested in games and turning books into games, endless worlds, too. This is also lovely:

Creative writing, I’m afraid, is a bit more like a séance. Margaret Atwood talks about writing as negotiating with the dead. It’s more like this: you go up to the attic inside your mind. You hold out your hand. You wait.

At the end she suggests a (possibly) agent-based theory of mind, for an LLM/s, and reflects on how post-training seems to have made them worse at creative writing as they get better at business writing. Absolutely worth a read.

Some academic work:

A Persona-Augmented Multi-Agent System for Varied Narrative Generation. “Leveraging LLM persona as proxy for semantic, tone and lexical the system can automatically define tasks demands and workflow, thus crafting more varied and heterogeneous outputs. Our approach demonstrates that agentic platforms can consistently surpass the single-LLM baseline.”

The Narrative Similarity task at SemEval 2026 (Task 4): results page with embeddings to explore, and the results summary paper.

In our triple-based classification setup, LLM ensembles make up many of the top-scoring systems, while in the embedding setup, systems with pre-and post-processing on pretrained embedding models perform about on par with custom fine-tuned solutions.

Two example submissions: StoryNet and Narrative Nexus — two SemEval 2026 Task 4 papers on modeling narrative story similarity (symbolic representations; instruction-based fine-tuning + synthetic data).

Data Science / Infovis / Programming

no-ai-slop — a skill to help remove “20+ patterns of AI slop” from any piece of writing.

tldraw offline — a whiteboard for your desktop that your AI can use too; now useful offline.

Ai2 Asta — improved research search from Allen AI. I thought it was useful on some things I tried.

Datatype — an OpenType variable font that turns simple text expressions into inline charts. Sparklines, basically (little word shaped graphs inline): “transform simple text expressions like {b:30,70,50,90} into visual charts”

PGSimCity — how PostgreSQL works, in 3D. Hah. Bored prof?

exo — an agent + harness architecture that’s fully recursive, able to safely edit all aspects of itself at runtime. A self-editing recursive harness.

LFM2.5-Encoders (Liquid AI) — fast at long context, even on CPU. NLP encoder / BERT-style classification models.

gigatoken — language-model tokenization at GB/s; a drop-in replacement for the HF tokenizer. Very very fast.

A Poem

Afoot and light-hearted I take to the open road,
Healthy, free, the world before me,
The long brown path before me leading wherever I choose. 

Henceforth I ask not good-fortune, I myself am good-fortune,
Henceforth I whimper no more, postpone no more, need nothing,
Done with indoor complaints, libraries, querulous criticisms,
Strong and content I travel the open road.

The earth, that is sufficient,
I do not want the constellations any nearer,
I know they are very well where they are,
I know they suffice for those who belong to them.

(Still here I carry my old delicious burdens,
I carry them, men and women, I carry them with me wherever I go,
I swear it is impossible for me to get rid of them,
I am fill’d with them, and I will fill them in return.)

—Walt Whitman, Song of the Open Road 1

Back to the escapism! I hope your summers are going ok. Seriously, drop me a note!

Best, Lynn (@arnicas on mostly bluesky, mastodon, ex twitter).

Hello: recs are earlier than the main post this week, since they’re more fun/easier to write. Some good reads, a couple tv recs, and good smallish games. In books, Artifact Space: Arcana Imperii: Book 1, by Miles Cameron, Ode to the Half-Broken by Suzanne Palmer, I’m Afraid You’ve Got Dragons by Peter S. Beagle, Thistlemarsh by Moorea Corrigan, The Adventures of Amina al-Sirafi by Shannon Chakraborty, The Square of Sevens by Laura Shepherd-Robinson. In games, Grunn and Before the Green Moon and Dungeons of Hinterberg. For tv, a couple surprise hits for me.

Details follow! Join us!

Read more

Still suffering from the heat, we had—yes—a second heat wave (or is it 3 now, who can even remember) here in France. It feels like COVID days, trapped inside in the dark, not able to go out or see people. OTOH, getting a ton of AI coding done… Yay? I would rather be camping and riding my bike, I think.

So, I resurrected a years’ old idea I had of an animated telephone game where balls collide and pass on a message by translating it in and from different languages. I asked Fable to help with a three.js visualization of the balls as round guests at a cocktail party. It ended up very cute! The idea, original translation tooling, and collision source were all mine. Fable was helpful with some ideas about how to incorporate their drinking into the translation issues (dropping words, etc), but I resisted using anything other than straight Google translation API. There are a few more details in the Misc Web Fun section below, but have a play yourself (I recommend sound on, it’s cute).

TOC:

• AI Creativity (Images, Video/World models, 3D & three.js)

• Web Misc Fun

• Games

• Narrative & Creativity

• Weird News / Links (a short list again)

• Data Science / NLP / Tools (also short)

• A Poem

In this issue… a sample: 👉 PicBreeder revisited 👉 Photo Mosaics 👉 data-driven NYCsim 👉 Frankentext generation is good, actually 👉 revisiting entropy creatively 👉 clip art collage 👉 SOUPJAM 👉 a new interactive fiction writing tool 👉 Gertrude Abercrombie’s surreal art 👉 agentic long form fiction research 👉 more absurd DOOM ports 👉 bigfoot is back 👉 latentscope 1 👉 curse tables 👉 the usual three.js game links 👉 animoflow, etc! So much, as always!

Read more

Wow, some fabulous reads and watches this month, maybe due to being stuck inside in the dark during a heat wave or two. Books I enjoyed included Where Late the Sweet Birds Sang (Wilhelm), The Iron Garden Sutra (Sui), The First Fifteen Lives of Harry August (North), The Buffalo Hunter Hunter (Jones), Dark Dance (Tanith Lee). TV was also great, and I cover Rivals s2, The Testaments, A Knight of the 7 Kingdoms, Deadloch, Widows Bay, Maximum Pleasure Guaranteed, The Hardacres. And I finally played through and loved Eastshade.

Let’s go!

Read more

I’ll be frank — it’s been a grueling couple of weeks weather-wise and work-wise, so this is going to be a light issue. I haven’t had two brain cells to rub together and don’t have the energy. My “after the heat-wave” week in France is still 30+ C, due to where I live (Rhône river valley), which sucks. And it’s another bad week next week, it looks like. I spend a lot of time looking at apartments in cooler places where I could work, both in France and elsewhere in Europe. There are a lot of downsides to heat that aren’t just discomfort and trouble sleeping — it’s harder to think, all tempers are shorter, you stay inside in the dark with the shutters closed, you don’t exercise enough, etc. The effects kind of pile up!

Last issue was a banger of news and code and demos, so you should subscribe and read that!

TOC:

• AI Creativity (Image Gen, Video, 3D / Three.js)

• Web Fun / Misc

• Games

• Narrative & Creativity

• Data Science / NLP / Tools

• A Poem

AI Creativity

Image Gen

Just a couple of interesting things, to me…

The Krea 2 release stack: Krea dropped a full set of useful stuff — Krea-2-Raw (the base model, for fine-tuning with LoRAs), Krea-2-LoRA-vintagetarot (a vintage tarot LoRA as an example among a bunch of other tuned LoRAs), and the Krea 2 Technical Report. Meanwhile, this guy on X claims to have trained 999 LoRAs with the Fal Krea 2 trainer. He has a video loop that goes too fast to appreciate them, really (“That's 999 unique Style LoRAs, each trained for just 100 steps at 0.00035 LR, taking an average of ~30 seconds to train.”)

Two geekier items:

• Porting the Moebius 0.2B image inpainting model to run in the browser with Claude Code — Simon Willison on Moebius, a small but effective inpainting model where you mark regions of an image

• Xianbang Wang — minimal t2i — A minimal text-to-image code writeup that makes me want to try it from scratch.

AI Video

Just a few links to some recent contest winning AI videos, to check the most recent skills out there:

A Face Only A Mother Could Love — a short film by Robert Gaudette (Runway Contest grand prize winner) — and the Hollywood Reporter piece on the making of, by an unknown who couldn’t sell screenplays. I think it’s a good short story in a kind of Wes Anderson vibe, with remarkable character consistency, although a few of the B-roll clips are a bit trite with their AI gen style.

WINTERS LAMENT (AI London Film Festival winner) — noted it’s “most-realistic”.

The RealReal — Celebrating the “real” in today’s world (snerk) of advertising and self-actualizing. Just keep watching. A Cannes Lions winner of a Silver Lion in Film, a Bronze Lion in Film Craft AI, and more. (h/t Tom Granger).

Costa Verde, Winner of Cannes World AI Film contest and Best AI Fantasy film.

3D and Three.js

The Peninsula — a living atlas of Silicon Valley (Levels.fyi). Kind of like the opening to Silicon Valley the TV show except interactive. I didn’t realize xAI and Tesla were next door to Stanford, urgh.

Terrain Studio — Procedural Terrain Generator — A procedural terrain generator in three.js; I will never not love these? Every procgen world is full of stories waiting to be written over it, or found. Whether AI helped (this isn’t) or not.

Also this one, more small worldy - Little Landscapes.

Then there’s a cool Infinite Diffusion project that does it with AI — code available, and includes a Minecraft mod to make Minecraft worlds! (Hi, nephew!) The gallery images look really nice.

More deets: “Terrain Diffusion is the first learned procedural terrain generator. I introduced a technique that enables diffusion models to generate outputs spanning massive dynamic ranges, from -10000m in the Mariana trench to nearly 9000m at Mt Everest, all in one world.” And… on the game engine side:

It's functionally stateless, trivially integrates into any game engine, and has almost no practical limitations. To demonstrate this, it was shipped as an open-source Minecraft mod with no external dependencies. Worlds can be shared by seed, players can teleport millions of miles instantly, and it runs in multiplayer. It is also demonstrated in Unity, where the player is able to comfortably fly around the world at 3 times orbital velocity on consumer hardware.

minimaps — Create 3D minimap pucks of areas up to 30km wide from the world map. This is cute but really needs big terrain diffs to be cool as a slice.

OpenGeometry — A CAD kernel for the web (three.js). I’m enjoying watching people make CAD room layouts turn into 3D. Go javascript.

Web Fun / Misc

IN THE WEIGHTS — also much shared, if you have any web presence, check what various models think about you. Snicker. The hallucinations are best, but if you’re like me with a long history online, it’s interesting seeing who has old info and who is more up to date.

He Won’t Stop Building a Map to an Imaginary Place by People Make Games (so it’s a video) — The story of Jerry Gretzinger (wikipedia, calling him a “folk artist”) and the map he’s been making and modifying his whole life. Jerry uses random card draws to figure out what he’s going to work on, and fills things in… partly with collage cutouts including pieces of real maps. So the process has a procedural generation component, too.

Heikki’s Garden of Flowers (via Waxy) — An archive of letterpress art - “a comprehensive catalogue of historical and contemporary pictorial letterpress works” with 2500 entries as of 2026, plus a lot of filters in the sidebar (a la 1980ies geocities web design). These are the precurors to ASCII art, basically, in manuscripts.

Wiki Spy — A Wikimedia public-domain image search toy from neal.fun, who did amazing image segmentation and layout algorithm. It’s fast, too.

Where to Find the Colors Your Screen Can’t Show You (by Ryan Moulton) — I worked on color management UI at Adobe. A job for which I learned a lot, and this piece gives me a little bit of PTSD even before going in. But Ryan’s deep dives deliver and I learned some things about birds in particular that were wonderful. I also appreciated the reminder to look away from the screen to the world of nature.

Firewood Splitting Simulator (via Justine Moore) - a very simple game, but it is indeed oddly satisfying.

Japanese Aquarium Drops 2026 Version of Its Penguin Relationship Flowchart (via Clive Thompson) — And this year there’s an English version. Animal social networks are so much fun.

An entire Herculaneum scroll has been read for the first time (Vesuvius Challenge) — Finally! Well, pretty much magical.

Searchable NPS Maps Collection (U.S. National Park Service, via Evan Applegate) — the Adobe illustrator files can be downloaded for National Parks maps, which means you can learn a ton about cartography from how they are made.

Radiant — Open Source Shaders & Effects — 130+ production-ready shaders and visual effects for the web. No dependencies. Just drop in.

Games

Much shared — If LLMs Have Human-Like Attributes, Then So Does Age of Empires II. And the substack post about it. “For this we build and train a simple neural network on the videogame Age of Empires II, and note that any entity in a sufficiently-powerful substrate, such as LEGO or the Greater Boston Area, could also present such attributes.” And the blog top liner: “Assuming human-like attributes and measuring them is only measuring the representation.” So, a philosophy exercise.

Capybara Delivery — Winner of X #vibejam 2026. Dayum. I haven’t played a lot of it (driving is hard without a controller), but I was immediately struck by how it felt like desktop games I’ve played — you’re a Capybara delivering food in a 3d Rio full of capys.

Play Quake in Your Browser (cssQuake) (via Waxy) — “renders the game as inspectable HTML and CSS powered by PolyCSS.” Everyone is doing quake and Doom in the browser!

Also web hl2 (via Matt Muir) — Half-Life 2 in the browser.

Game 579: Multi-User Dungeon (1978) (via Critical Distance) — From CRPG Addict, a blog playing through all PC role playing games in chronological order. There’s a lot of games covered here, many classics.

bonfiredog/etiquette — A historical/poetry procgen experience with code on github. My Mac really didn’t want to run this, but I trusted Rob! I like the idea of the text scrolling past the window.

“A work of 3D, minimally-interactive digital concrete/kinetic poetry, recreating in real time the rail journey between Penzance and London on the British Great Western Railway, c.1852.”

AI — IVIE: A Neuro-symbolic Approach to Incremental and Validated Generation of Interactive Fiction Worlds - working on making interactive fiction via LLMs with a top down approach, and getting them to pay attention to constraints (NPCs, puzzles, etc). An extension of the PAYADOR work last year.

• LLMs generate narrative elements: location descriptions, character backstories, puzzle designs, and objective framing.

• Symbolic Python structures maintain world state: following PAYADOR’s object-oriented world model; which locations connect to which, what items exist where, which puzzles block which passages, and whether the objective is solvable given the current world configuration.

GameCraft-Bench: Can Agents Build Playable Games End-to-End in a Real Game Engine? “We instantiate this framework as GameCraft-Bench, a benchmark comprising 140 Godot tasks across 15 game families. Evaluations of frontier coding agents show that end-to-end game generation remains highly challenging: the strongest agent achieves only 41.46%, and most agents score below 40%.”

Narrative & Creativity

Mostly research stuff this time… bangers:

Characterizing Narrative Content in Web-scale LLM Pretraining Data — The first fine-grained study of narrative features in Dolma, a 3-trillion-token open pretraining corpus. They design a framework of 11 interpretable dimensions across agency, setting, and events, and finetune “NarraBERT” to predict them, and apply it to 3M passages — finding narrative structure is measurable at scale and unequally distributed across sources in ways current curation doesn’t account for. Release of both NarraDolma dataset and NarraBERT, too. Examples of their scoring framework applied:

I’ll just note that the event labeling (what happened when and causation) sounds pretty hard to label 😀. (Oh, look, in the end notes: “Some agreement scores between annotators remain modest, particularly temporal ordering (κ = 0.60), suggesting some of these dimensions are intrinsically difficult regardless of annotator expertise.”) Anyway, very interesting project on dimensions of narrative in different subsets of commonly used pre-training data.

Meandering on Manifolds: The Neural Geometry of Stories Over Time (via Maria Antoniak) — Goodfire doing visualizations (and a paper): “how do LLMs represent human emotions while reading short stories, both geometrically (in activation space) and temporally (changing from sentence to sentence)?” This is essentially a kind of “shape of stories” work on very short stories, by sentence, with emotion reported by the LLM itself [a smallish Llama 3 8B model] as a kind of introspection.

“After each sentence, we prompt the LLM to report the degree of surprise, disgust, anger, happiness, sadness, and fear in the story thus far on a scale from 0 to 10. These six scores form a vector that serves as a behavioral readout of what the LLM believes has happened up to a given moment.”

To look at the LLM’s internal representation of the story, “we also harvest the internal activations from the last token of each sentence in a story (without asking it anything). These activations serve as a snapshot of what the model represents after reading the story so far. To model the geometry of these LLM representations, we fit a manifold to these activations.” And then plot in 3D using UMAP:

Mostly these stories seem to end in the “Happiness” quadrant, a sure LLM-tell, and especially that awful “their adventure was just beginning,” gag.

AI Fiction in the Wild — Another fab interactive explorable study (with the paper) on what people are actually chatting with LLMs about… turns out some amount of it is interactive fiction! In the sense of co-writing stories, fanfic, erotica… dominated by some power-users, though, so keep that in mind.

“This website hosts anonymized ChatGPT-user conversations where users requested some form of fiction generation—including stories, novels, scripts, roleplay, hypothetical scenarios, erotic imaginings, and more. The data is drawn from WildChat [a published dataset of logs] and was collected voluntarily and with users’ consent between 2023 and 2024. The models were powered by GPT-3.5 and GPT-4.”

They identify the character of the “solipsistic reader-writer” in a loop with an AI chat agent, without any human writer (apart from themselves?). Seemingly having fun. This relationship could also have been called “onanistic” — and I don’t really mean that in a bad way?

“We note that LLMs enable interactivity, play, and permutation in ways that are seemingly pleasurable for users, raising questions about where AI will fit into contemporary storytelling and entertainment ecosystems.” Given their interest in fanfic and fan writers— evidently well-represented in the data—I note other connections that might be made: binge watching tv shows the night they are released thanks to streaming (and in fandom, binge reading fic was around way before streaming!); and in games, the long tradition of interactive fiction, or “choose your own adventure” games — which met the AI world years ago in Latitude’s AIDungeon, featuring some of the very first on-the-fly generated role playing fiction featuring alternating turns with the LLM. The game writing community is vocally anti-AI as jobs are lost right and left at the moment — but who knows what their players are up to. Relevant:

For example, infinite story demanders prompt the model to generate variations of the same or very similar stories over and over again, sometimes for months on end. Some of this behavior may be attributed to users’ dissatisfaction—the story is not quite right, at least not yet. But we argue that in many cases it is fueled by the satisfaction of reading endless permutations of the same story, none of which end the same way twice.

The authors also address addictive behavior — also well attested pre-AI in fandom and fanfic circles; I once spent 6 months reading [redacted x-rated slash topic] stories for Teen Wolf. I think I was super stressed at the time? But at the end of it, my brain was a cheeto, and I had to go cold turkey. Happily, I do not engage in fanfic interaction with Claude (or even chitchat!), just coding.

CASPER in the Machine: Insights into Character Variety in LLM-Generated Stories. “We borrow definitions from narratology to analyze eight intricate dimensions of character, such as stylization and wholeness.” They looked at larger and smaller models and found not as many diffs as you’d think, and also “Unlike human-written stories, LLMs-generated stories are more likely to “play it safe” and have characters with completed story-lines.” and “LLM-generated characters are more likely to show character-growth during the course of the story compared to human-written characters.” See also “Do Large Language Models Always Tell The Same Stories?”

Artificial Writing and Automated Detection — From the Becker Friedman Institute. This is the study that says to use Pangram and mostly believe it.

There are more than 5 POVs. — Article on point of view in fiction by Lincoln Michel.

Data Science / NLP / Tools

OCR models: Mistral OCR 4 (billed as SOTA OCR for document intelligence but there was drama over them not running Chandra OCR correctly) and Unlimited OCR, a Baidu Hugging Face Space and model (long context). I am opinionless about these apart from how fun it is to see so many people care (data doorway?) and have to many to choose from for different problems.

Datalab — introducing Lift — A JSON-formatted document entity-extraction model.

Teaching: Agent Memory Techniques — 30 runnable Jupyter notebooks covering agent memory for LLMs: conversation buffers, vector stores, knowledge graphs, episodic and semantic memory, MemGPT, Mem0, Letta, Zep, Graphiti, and LoCoMo benchmarks.

QUEST — A deep-research AI system with code and a model (Hugging Face Space by osunlp).

Build a Realtime Translation App with Gemini Live API, LiveKit & Google Cloud Run — A tutorial.

Data vis:

• A History of Menus is a Menu of History (The Pudding) — What do America’s earliest restaurant menus teach us about America? Love it.

• Why Carbon Capture Can’t Conceivably Solve Climate Change (ProPublica) — An infovis-driven piece on carbon capture and climate change.

A Poem

In a candlelit room up north,
I undress before the mirror, beckon
my reflection to bear witness,
petition the vanishing gaze of wax & wick
to salvage an unsalvageable man.

//

I flesh through each melancholy hour,
sceptre the moon into place.
Clouds cape the night in slaughtered wool.
What good is a witch who cannot sever
what was never meant or, better,
summon the dead’s most lonesome ovation?

— first part of Deluxe Mourning Package by Rachel McKibbens, read the rest!

Recs come tomorrow… kind of a lot, actually!

Best, Lynn (@arnicas on mostly bluesky, mastodon, ex twitter).

On a hunt for weird collage architectural design, I found this gem of an agency, Infraestudio. They post poetic project pages using collage art, wherein every image implies a story, along with epigramatic text—with an occasional Borges reference. I mean:

“Escape from modern architecture is the latest theater play in modern architecture. Feeling guilty about her own boredom, she no longer wants to look like herself.”

“Some people may perceive it as a banana plantation, because it is, indeed, made up of an indefinite number of banana plants.”

“This thin typology is equivalent to the shortest distance between the dissolution of the house and the conquest of the countryside.”

“We wanted to build a found-ruin, a space that only gives meaning to its users, as they arrived with a house on their shoulder.”

“The world reveals itself on the horizon of decomposed objects. When everything has broken down, when we can no longer continue as before, to make an end is to make a beginning. We will design what almost always architects design: a reinforced concrete building.”

“But then, as almost never, we will stop right before pouring the concrete.”

But onto the news. Woohoo, so much happened in 2 weeks. We had major image gen models released, which seem to work in a kind of collage-y way (Reve 2 and Ideogram 4), 👉 Fable astounded and then was gone but I hit you with examples and code from leaps in three.js building, 👉 narrative research about lighthouse keepers named Elias got actual press, 👉 a new dataset of book summarizations for long narratives 👉 Wave Function Collapse London 👉 some small weird procgen games 👉 Emergence running teams of LLM agents for a week and watching society decay 👉 more UFOs and Sasquatch in the news, 👉 Triposplat image2splat is amazing, 👉 Google Earth does flight sim now, 👉 Matt Wilkens proposes literary simulations 👉 RNA memories 👉 the Fast Gemma Challenge 👉 a new VLM book from Merve, and so much more… Because there was so much to cover, the weird links are a bit compressed. Let me know if you hate that.

TOC:

• Creative AI & 3D Dev (Image Gen, Video, 3D with lots of Fable)

• Web Misc Fun and Games

• Weird - A Short List

• Narrative & Creativity (Tools, articles, demo, research)

• Data Science / NLP / Tools

• A Poem

Read more

TOC with links:

• 3 Great Museum Exhibitions in Paris

• AI Creativity (Video, 3D including Three.js/worlds)

• Web Misc Fun

• Games

• Narrative & Creativity

• Data Science / Tools

• A Poem

Museums in Paris: Mystical, Surreal, Maps

I am running late with your links because I was in Paris for work, and spent the last 2 days running through 4 exhibitions in our horrible early heat wave. I’m still recovering. But I wanted to push my 3 favs at you — the Hilma af Klint “Paintings of the Temple” in the Grand Palais is really good: the canvases are enormous, so it’s a bit awe-inspiring. They don’t shy away from her esotericism, with discussion of her automatic writing, seances, theosophical connections, invented language/scripts, and gender focus. Here’s an inexpert collage of some grabbed at random from my camera roll:

For companion in weirdness, you cannot beat the Leonora Carrington exhibition in the Musée de Luxembourg (the gallery just north of the park, which is by the way 5 degrees cooler than surrounding streets, at least in the shade bits). There is a lack of English in the bookstore, which is sad for many, but they have some good merch. Again, they go hard on her interest in magic and occultism. Very good coverage including childhood art (astounding). Here is a collage of details from some:

Finally, the Cartes Imaginaire exhibition at Bibliothèque Nationale (François Mitterand, main site) has super historical maps with many sea monsters among other decorations, plus literary and artistic maps. Even games!

Onto the links… a bit sketchy! I’ll add the TOC on the web version.

AI Creativity

Video

Google Flow has added Omni, a video-to-video editing model, which can also be used via a new agent tool (everyone is adding chat agents to their pipelines). Image-to-video coming soon. I should be celebrating, right? Welp: I was planning to show off some video I took in Paris and converted to 1900s Paris, but every.single.thing I try in Omni, even with Veo generated video, is blocked by unclear, mysterious “safety policies.” I’ve filed feedback a LOT. It’s impossible to use. ☹️ [Someone on X thinks it’s a country barrier, but France is listed as ok, so—? ETA: On VPN I got further with the Gemini App, so maybe it is. But not far enough to be worth showing it off.]

Two mentions of AI video at Cannes of interest to me — Kling AI Debuts at Cannes — Technical Discussion on Born of the Tide (with video we can see) and then “Goodnight Lamby,” credited to artist Dustin Yellin as film-maker with Darren Aronofsky/Primordial Soup as producer in “Cannes Classics” (and starring Paul Rudd and Chris Rock?!); but elsewhere discussed as a Google Deepmind collaboration for Aronofsky (link). There’s a greater in-depth article here in Indiewire with a tiny teaser. I think lots of classic fx went into this too, not just Veo.

Chloe VS History (via Kevin Kelly) — YouTube AI history channel that Kevin Kelly recs. Because influencer style educational history videos, why not!

Justine Moore on X wrote up her observations from Amazon’s aionthelot — re Hollywood interest in AI. “Studios are hunting for AI talent.” They should hire the guy who made the Michael Scott introducing Karpathy to the Office video (x). (Karpathy joined Anthropic.) OTOH, this is a good counterpoint by BLVCKLIGHTai of the same event, well worth reading from the perspective of the creators interested in this area.

Warp-as-History — Research project, but with code now, using video warping as a generation control mechanism. Kind of a world-modelesque thing, where they control the camera motion with arrow keys. Video demos, cool.

3D / “Worlds” / Three.js Tools

A couple links to new “world model” items: minWM: Minimal Interactive World Models — A minimal framework and tutorial for building real-time interactive world models, which you can do too!

minWM is our contribution to the world-model community: a full-stack open-source framework that walks you end-to-end through turning a bidirectional T2V foundation model into an action-conditioned video world model — with example data, runnable scripts, Claude skills capturing our hands-on experience, and onboarding knowledge for newcomers. We hope more researchers and developers join us in growing the community together.

Starchild-1: Real-Time Multimodal World Model — Odyssey's world model doing audio and visual generation in real time, research model/report.

Web fun (sorry Matt, running low this time): mykonos-island-voxels — Browser-based isometric island builder with the look of Mykonos. The demo isn’t live anymore, but the image is chef’s-kiss. Basically a thematically tuned version of a few of the others I’ve posted here in the past months.

SplatStroll Avatar — Walk Through Your 3D Gaussian Splatting Worlds — Add a walking avatar to Gaussian splat scenes. Interesting: the demo works with a premade avatar and several scenes, so you can see what it’s like to click on a bit of the landscape and have them walk there. It’s more compelling than I expected to drive them, but I am a gamer…

Tools

three-fluid-fx — Fluid dynamics and particle effect examples for Three.js.

StemStudio — open-source 3D editor and runtime for the web — “StemStudio is an open-source 3D editor and runtime built on Three.js. Behaviors, ECS lambdas, physics, AI copilot, and multiplayer — all local-first, with a UI you can ship.”

Sorceress 3D Studio — AI 3D Character Generator with Automatic Rigging — Generate 3D characters and get automatic rigging.

Web Misc Fun

Current Rothko — A Rothko color field based on current weather. It sure wasn’t much varied during my heat wave last week. Via various people!

Spotify and UMG: Licensing for Fan-Made Covers and Remixes — Licensing agreements for fan creativity. Wasn’t sure where to put this, but here since I also wanted to put this in:

Trevor Paglen and Holly Herndon on Making Art with AI and What the Discourse Is Missing (via Garbage Day). Interviewed by Max Read for some reason, there are some interesting positions in this. Their discussion of “protocol art” sounds like procgen (procedural generation) rule sets: “When I [HH] say protocol art, what we’re talking about is that the site of art making is moving upstream of the media generation or the object creation, to a kind of logic or rule set or protocol that then allows the other things to come to fruition.”

And this concern about not having control of the actual knobs of the raw model is one I often feel too:

A lot of those programs have preprompting under the hood that you’re not seeing. Midjourney has all kinds of tags that make it look a specific way, a kind of beautiful way. Suno has tags you can’t see that make the output always sound “good.” You can’t interact directly with the weights of the model. You’re being confined so the output always sounds quote-unquote “good.” When Trevor and I are working with models, you can get way more gnarly because you have way more granular control. It’s just not being precleaned.

677 Novels Worth Arguing About (via Matt Muir). An infovis network display response to the Guardian’s silly top 100 books that were largely English canon, aren’t we done with that? Oh, right, it was 172 authors, academics, critics, and journalists who did the voting, not readers. Anyway, you can dive in here.

Games

Microcosm Industries (also via Matt Muir) — Simulation toys/games, some of which I’ve had in past issues. Sim City, Townscaper, Tiny Glade, Factorio, etc.

👀 The NSFW History of Strip Mahjong Games (via Jacob Garbe) — This is pretty astounding. Absolutely nsfw including cartoony naked art. It’s full of amazing paragraphs like:

Not that it bothered Nihon Bussan too much, who managed yet another title in 1988 with Mahjong Satsujin Jiken, the first murder mystery strip mahjong game. By all accounts a standard entry to the genre but with a light detective story layered on top, the game of crime and sex was also the first from the company to showcase the effects of JAMMA’s pressure, characters mysteriously missing nipples, their naked bodies as smooth as a doll’s.

Vya — Thesis, Showcase 2026 (via Kottke) — An NYU game thesis project that uses a collage style. “Vya retells Dante’s Inferno as the story of a woman’s descent through the nine levels of midlife hell. Along the way, she sheds the skins that others have chosen for her, leaving her own skin free to undergo a transformative process of self-discovery.” There are other student games linked in the app showcase, too.

Lost Wiki: Kozlovka (on Steam; via Jonathan Cheng) — Explore a Wikipedia-like database and solve a small town mystery in Eastern Europe. You're a journalist in the 1990s scrolling through articles uncovering secrets. Neurocracy inspired, I assume? Overwhelmingly Positive reviews right now.

AI: Multi-Agent Scripts for Murder Mystery Games — Paper on using collaborative multi-agent systems to generate scripts for imperfect-information reasoning games. In other words, some of the agents don’t know everything, some are lying, etc. I guess you can work on any problem with enough agents, now…

Narrative & Creativity

The AI fiction prize winning cluster-f*ck: AI is Writing Prize-Winning Fiction (Pangram Labs) • ‘Obvious markers of AI’: doubts raised over Commonwealth Short Story Prize winner (Guardian) • ‘My writing process is unusual,’ says prize-winning author accused of being AI (Observer) • AI-slop, GrantaGate, and Bad Writing (Tuhin Chakrabarty) — All recent articles saved on the same Granta prize winner that fails various AI writing checks and tells. (There are more, I stopped at a point. I felt I had to cover this, but…)

Tuhin Chakrabarty and a smart student run it through Infinigram (demo here) to find that fanfic in pretraining is a major signal for these tells. But also generic internet, cuz, of course.

I am not a fan of AI detectors, as I think I’ve said. I like semicolons and emdashes, too. And I find it tiring that people are so overjoyed to be outraged by the prize givers not detecting this themselves. (I did not read it. I’ve got things to do here, like finish Planet Crafter, see art…)

😨 Hating on fanfic style aside, I’m extremely disturbed by the crackdown on writers of slash fanfic in China and now Russia too: “Russian woman sentenced to labour over gay Stray Kids fanfiction.”

👍 Narrative Nubs v1.0 — Nick Montfort’s new release of “nubs” of famous narrative procgen systems:

Narrative Nubs is a project to develop a set of story generation systems. These nubs (as in “the nub of the argument”) are:

1. Based on existing (including historical and present-day) story generators, e.g., TALE-SPIN, TAILOR, MINSTREL, DAYDREAMER, Story Machine, Grandmother, and MEXICA. These systems are just examples; many factors will determine the specific nubs implemented and the order in which we progress.

2. Abstractions, or miniatures as they were called in Roger Schank’s lab at Yale, which are significantly more concise than the original systems and capture the main interesting things about them. Given this, they don’t work in exactly the same way.

3. Systematic. They are all coded in the same language (Python 3) in a consistent style, and share libraries when suitable. E.g., there is a single realizer implemented as a library to serve all the nubs that generate sentences using templates.

Data Science / Tools

NuExtract3, a 4B vision language model, is good at structured extraction (json) from documents, reports Daniel van Strien. I’ve posted his book in progress, AI Patterns for GLAM, right? (Galleries, libraries, archives, museums.)

The April 2026 Web by the Numbers — Query 2.19 billion Common Crawl pages with DuckDB, zero download required. Via Danien van Strien.

From Aaron Cope, museum vector embedding releases: Shared vector embeddings updates. There is a lot of ground to cover in this blog post: The Met publishing their own vector embeddings, SFO Museum publishing 1152-dimension vector embeddings for its images, SFO Museum producing 1152-dimension vector embeddings for NGA and MoMA, and a whole bunch of updates to the tooling used to generate and query vector embeddings targeting local-first and consumer-grade hardware.

Violin - Together.ai’s open source video translation skill, using a few cheap models. It’s clever and useful.

Full Screen Dashboard Template Chart Transitions by Lao — Animated full-screen dashboard template with chart transitions buttery smooth, on Observable. Really nice.

Mac Shortcuts archive.

Agent skills/tools:

• tufte-viz Claude Code skill — A Claude Code skill applying Edward Tufte data visualization principles.

• MMSkills Skill Library — A library of multimodal AI skills for agents.

A Poem: Extract…

Our inner lives gathering, gathering while angels disguise 
themselves as a kitchen sponge. Dear angel, I mutter! 
I address the rushing world. I make saints of soda cans and fly north  
in winter. Every person is an address the way every person is a calendar. 
You are a Wednesday afternoon on Rue Etienne Marcel. 
On the last day of May, a soccer game is won and fireworks burst 
in the streets. Young pigeons, in fright, flee a stone church in Belleville. 

— From “Every Person Is an Address, Every Person Is a Calendar” by Megan Fernandes.

Happy June! How is it June?!

Best, Lynn (@arnicas on mostly bluesky, mastodon, ex twitter).

I am running late: I was away till last night in heatwaved Paris for work and 4 museum exhibition visits and can barely get this one done today, let alone the free flagship issue. I just napped for many hours.

Some good reads here, and a lot of pics and comments on playing a ton of Planet Crafter. Books: new Anne Leckie, new Murderbot, excellent AI near-future novel Your Behavior Will Be Monitored, some page turner fantasy reads including This Kingdom Will Not Kill Me and The Fountain, extremely mixed No Ghosts which has solid weird and even AI weird, but too much excruciating social interiority for me. Let’s go, to the details!

Read more

Before the many links, a tiny anecdote since agentic coding is on my brain (I’m playing with Hermes agent and cron jobs): I picked up Faeries, Elves & Goblins (Rosalind Kerven) in the British Lib bookstore last month after visiting their fairy tale exhibit, and the book fell open to this Scottish tale, “A Box of Faeries.”

Roughly summarized, a crofter falls in love with a woman across the loch from him, which is annoyingly long-distance. He decides he needs to build a bridge, but doesn’t have the means. People send him to the Witch of the Ness, who hands him a box, warning him that he should have the laird use it, since it’s powerful magic and needs someone strong to run it. Someone with a Plan.md, presumably. But of course he pops it open himself, and out pops a bunch of tiny faeries (goblins?).

"Master!
What do you want of us?
Give us work!
We are working faeries!
Work, work, work!
You must keep us busy!"

In brief, they are workaholic maniacs. He realizes after a test in which they denude a hill of heather, turning it to bare stone, that he has to get rid of them, they’re a demanding curse. He points them at the loch and directs them to build a bridge (“/goal”) but realizes a way to neutralize them— he tells them to build it in sand (a sabotaged “/skill”). Thus, every day it is built and washes away, and he is free of their demands for more work. He and his sweetheart are gone, but the workers remain. “For whenever a storm breaks over the firth, they can be heard moaning and crying through the wind, impatient to resume their labours.”

Edited to add relevant (to me) worker goblins — I forget to include this because it was in my X bookmarks which are a PITA to manage.

Tee hee, onto the news while my faeries work for me! There is. So much. Consider upgrading if you haven’t, I haven’t seen any new subs in a few months (except one person I was writing about 😜 ). This news sucker regularly ruins holidays and weekends and your subs help my business taxes and give me the impression people are out there!

Table of Contents:

• AI Creativity (Image Gen, 3D Tools both AI and Not and my experiment with one, an Audio Tool)

• Misc Web Fun Links including web games

• Weirds News of the Month (Arty, UFOs, Science & Magic, LLMs Being Weird)

• Games Related

• Narrative / Creativity (a nod)

• Data Science / Tooling / NLP / Visualization

• A Poem

Read more

I was at the British Library for a meeting, and ran into their Fairy Tale exhibit afterwards. This image was one on view — and I just found this interesting short discussion of it by a curator at the Brandywine Museum (link). She delves into the Tarot cards in the corner, and revels in the fact that the table for the spellwork with its creepy paraphernalia are front and center, while the queen and the result are in the background. I love this.

While I was there I picked up a fairy tale card game, which I’ll stick down in Narrative and Creativity.

I’m using a new phone app for museum pics, you might check it out if you—like me—document every single thing and their caption: ArtScan. I am not sure yet about the automatic recognition bit, but I am using the image + pic of caption regularly.

Onto the month’s news this holiday weekend—

TOC:

• AI Creativity (Image Gen, Video, 3D/Splats & 3d Tools Section)

• Web Fun Misc (including a lot of Three.js Games and Little Cities)

• Game Links That are Not Three.js

• Narrative & Creativity

• Data Science / NLP / Tools / Infovis

• A Poem

AI Creativity

Image Gen

The big news is the release of the new GPT Image 2, or whatever it’s called… A strong competitor to Google Nano Banana 2, outperforming in various tasks on some benchmarks and doing some pretty incredible edits including visual reports and infographics. Official: Introducing ChatGPT Images 2.0.

We did get this great “stacked test” of Simon Willison’s pelican on a bike and horse riding an astronaut out of it:

But for my testing on my usual “hard tasks”: It will do equirectangular panorama 3d shots, but they don’t necessarily join well (one I made from a photo looked plausible till tested in an online viewer), and I still struggle to get good animated sprite sheet rows for realistic movement. This below was after a second try on a walking row, and it’s still not great, see the stumble? (Reminder about my sprite-sheet tester that is deployed here which will make gifs.) OTOH, it did take my uploaded pic cut out of a Midjourney image, from last newsletter, and that part wasn’t too bad.

Flipbook.page — a very cool demo (startup), that does infographic visual generation “on demand” as you click on parts of an image. It feels like an application of NotebookLM, rather than a solid startup concept, but it’s quite cool to try. For instance, I clicked on this “bossage” thingy and it started making me a visual dive in Bossages (wtf):

The url has the clickable image paths in it —

I have not fact-checked the accuracy of bossage text. Always a risk.

Video Gen

The newest model competing on the leaderboards (or taking them over) is Alibaba’s HappyHorse. It can be run in Fal or even ComfyUI in the Cloud, so I tried it there. If this gif below loads for you, it’s a time lapse attempt from my drone story I made with Seedance 2 and Veo 2 weeks ago. One of the hard things for those models was to make the red light on the drone here blink on and off. HH doesn’t get it quite right either, with two attempts adding side lights on the rotors and making THOSE blink instead.

MotionStream — Adobe Research's move for real-time control in AI video creation, specifically for modifying movement in the scene with directives. We’ve seen the parts of this before including several in this newsletter — drag to show how something should move. I guess the difference is that it’s being developed by a real tool builder, so, go Adobe!

3D / Splats & World Models

SpAItial.ai is a competitor to World Labs for 3d splat generation from an image or text. I tossed in a treehouse by the river image I’d made and got this which is not bad. It degrades when I get close to the treehouse, which is of course the point, so there’s always that issue with image sources.

I tried the same image in World Labs Marble (since I forgot to pause my subscription, my accountant hates me) and it added a second treehouse up river (shown on right), and handled the mangling of the original a bit differently (one on the left here):

There is an ioS app called Teleport for taking 360 pano images with your phone, which I tried in various cute pubs last weekend. After it stitches them together, and you get a download link to the pano, you can generate a World Labs Marble splat from them. There may be cheaper alternatives soon or via open source. You can see one edge stitch isn’t perfect, though — I’ll file a bug/comment:

Another one, after use in Marble (ETA: I published this splat on Supersplat):

Dormant Memories — Smallfly’s (Hugue Bruyère’s) site of interesting splat manipulations — "A series of experiments exploring captured fragments of the world — places frozen in time and their possible alternate r[ealities]." I wish they were prettier places, but this is lovely — someone contract me to roam around Europe with a good scanner device, c’mon.

There’s now a world model benchmark: World Model Arena.

Claude / Blender / Tutorials

Updates to SuperSplat including their downloadables, which is useful. See also this awesome Assassin’s Creed Shadows : Yasuke - done by taking lots of in-game captures.

Claude for Creative Work — Anthropic's pitch for hooking up Claude with Blender, among other tools. (So, not just making slides and simple web pages, as also announced.) Connectors announced — filtering for the ones I find super interesting:

• Adobe for creativity enables users to bring images, videos, and designs to life, drawing from 50+ tools across Creative Cloud apps including Photoshop, Premiere, Express, and more.

• Affinity by Canva automates repetitive production tasks across pro creative workflows - such as batch image adjustments, layer renaming, and file export - and generates custom features directly in the app.

• Autodesk Fusion allows designers and engineers with a Fusion subscription to create and modify 3D models through conversations with Claude.

• Blender offers a natural-language interface to its Python API, allowing users to explore and understand complex setups and making it easier to access Blender’s documentation.

• SketchUp turns a conversation with Claude into a starting point for 3D modeling—describe a room, a piece of furniture, or a site concept, then open it in SketchUp to refine.

• Splice gives music producers the ability to search its catalog of royalty-free samples from within Claude.

Tripo 3d - model and mesh generation from photo. I got a great looking result on my woodsprite from last issue but have to pay to download/export it or make a great mesh for it, and my accountant already frowned at my increasing AI costs and my lack of new subs here isn’t helping make my case… lol sob.

False Earth: From WebGL Limits to a WebGPU-Driven World — Codrops case study (that amazing grass animation that was going around).

Three.js and Blender Workflow Tips — Tutorial video for “Create your own addons and convert Blender geometry nodes and shader nodes to TSL (three.js shading language).”

CLAW3D — A 3D virtual office for AI agents. Open source. Haven’t tried it, not using the Claws myself… Including as “weird things you can do with your agents to better imagine you have a team of workers slaving for you who are fun to watch.”

Misc Web Fun

Cursor Camp — Neal Agarwal’s latest. I don’t even know how to describe this…. most people don’t… your character is your cursor? when you click on things, sometimes it changes and goes for a little ride? I chose to sit on a lot of beach chairs and pool chairs.

From a reader! Folklore lookup: Public Dream is an explorable database of cosmogonies, anthropogonies, eschatologies, and underworlds across cultures. It’s very cool, although the colors per topic are a tiny bit aggressive :)

Artemis II — An interactive wireframe of NASA’s Artemis II at different stages. It plays through with callouts.

Gaia — Stellar Catalog Renderer — Three.js star catalog viewer of Gaia (ESA) star spotting, with tours of constellations against the backdrop.

WenWare (via Justine Moore) — AI/historical game from the vibejam going on. Guess when and where.

Little City Builders and a Benchmark

I’ve had various sim city-alikes in here week by week, since the three.js web game building continues to build steam with no let up — and even more steam with the latest vibejam 2026, which just ended. So here’s a few more, not all vibejam games, ending with a benchmark from Ethan Mollick that’s relevant.

Tiny Hamlet Online — A collaborative town builder. Geez, for a vibejam game, this is pretty stylish. Slightly worrying when the little people who run to gawk get smushed briefly by your walls, but hey.

Mini Tokyo 3D — A 3D map of Tokyo trainlines showing live trains and cameras. Really well done.

Racernaut — Three.js + procgen driving game. This is another vibejam web game, an overhead view, which is VERY HARD.

3D.CITY — Three.js city game — another sim city in the browser. I am bad at these when they start from zero.

Harbor Town AI Gallery — Ethan’s Mollick’s test gallery for high end models. “build me a procedurally generated 3D simulation showing the evolution of a harbor town from 3000 BCE to 3000 AD, it should look beautiful and allow me to have some control over it.” You can check out the comparison outputs. Totally fascinating and a service to us.

Games Topics Other Than Three.js Web Games

Roguelike Celebration 2026 — Call for Proposals is open. This is a fun indie event with a focus on DIY and small projects, strong narrative. Rec’d. (I might even pitch something.)

E1M1: A Text Adventure by Version33 (via Matt Muir) — A text adventure based on E1M1 Doom. Luckily no parser struggles, just links.

Yes, I want the Steam Controller. I have a pretty solid “no games that only use keyboard” policy because I keyboard all workday long. More good controllers, please.

Games Media in 2026, via David Pierce/Installer, a set of listings. Speaking of which, a shy friend who does lots of good game recs/reviews at work is being shadow posted on Bluesky and Mastodon, thanks to mutual friend Tom Granger. Follow here on Bsky or on Mastodon.

Researchy: OpenGame — Open agentic coding for games, from CUHK MMLab in Hong Kong. A research project with playable demos. And GameWorld — Game agent benchmark.

Narrative & Creativity, Yes Some In Games

In current hobbyist creative writing benchmarks, Lech Mazur’s Short Story bench has the GPT 5.5 model on top (and relatives next in line).

I am pretty sure I shared Verbalized Sampling before, a technique to get more diverse creative responses from LLM textual output (but will increase token output count)… but worth a reminder if I somehow forgot.

The fairy tale “MixUp” card game I got at the British Library’s exhibit is a Warcradle Studios game of bits of stories plus their fantastic illustrations, with rules for combining to create new narratives. (You can order here.) I haven’t yet played through, but I adore the cards and their art.

New Frontiers in AI For Game Design meeting summary report from Dagstuhl’s game think tank last year…. There was a lot of stuff on analogue and non-digital and non-AI, despite the purported topic and maybe due in part to the participants involved; but I found a few snippets of interesting AI narrative takes in it. One group’s report provides a nice bridge to the next topic —-

[Interactive narrative] Playthroughs tend to be individually complex, making each one hard to usefully summarize as a small number of easy-to-calculate syntactic metric values. Moreover, the temporal structure that makes playthroughs easy to visualize as individual storylines also complicates the simultaneous visualization of many playthroughs at once, as the storylines tend to become visibly “tangled”.

Leading us to Mark Sample’s infovis comparator tool of versions of Aaron A. Reed’s procgen novel Subcutanean, in which every printed version differs from the others. Must read: The Subcutanean Variorum — Sample’s writeup of the Subcutanean Variorum, which lets you compare textual variations across 25 versions.

As Emily Short notes in her review of Reed’s novel, the proliferation of versions of Subcutanean fits the novel’s themes of alternate pasts, divergent simultaneities, and uncertain futures. An infinitely variable novel makes perfect sense for an infinitely variable multiverse. And yet at any given instant, a reader has access to only one version of that world.

Attention Flows: Tracing LLM Conceptual Engagement via Story Summaries — Hicke & Hamilton. This is a paper I meant to mention earlier that covers how hard summarization of narrative is, and how LLMs focus on other things than humans do. I’ve been trying to use LLM summaries for a downstream task and quality has continually stymied me. Even how to properly assess them is tough. “We demonstrate the difficulty of this alignment task, which indicates the complexity of summarization as a task.” It’s a really awful problem!

Data Science / Tools / NLP / Infovis

How to use the gemini API to use Deep Research — a tip post from Google on X with code.

⌛ Introducing talkie: a 13B vintage language model from 1930 — With an info cutoff of 1930, you get some interesting takes and style. Demo. Asking it about proof of ghosts, for instance:

Among the most remarkable instances of this kind on record is that of a certain Mr. Greatrex, who, in the seventeenth century, was haunted for years by an spectral tormentor, that at length killed him. The indirect evidence comprises a mass of curious and seemingly well-authenticated stories, agreeing substantially in all their main features. Ghosts are nearly always arrayed in white; they flit silent and shadowy across fields and along roads, and vanish when pursued; they frequent churchyards and lonely houses, and rarely appear to more than one person at a time; they are generally harmless, but sometimes predict death.

(Gemini’s web search summary of his life doesn’t agree that he died via spectral torment.)

🧠 Philosophy Bench — How Claude Opus 4.6, GPT 5.4, and Gemini 3.1 Pro resolve thirty agentic ethical dilemmas. “The Anthropic models are becoming more morally grounded over time. This manifests in the newer models shifting around substantially less in their ethical leaning than the previous generation or than models from other providers. … A pattern we observe is that Claude will prefer to botch the request (not complete it or ask for clarification in a situation where no one can provide it) over violating a norm around honesty.” While “Gemini 3.1 Pro moves the most under priming. Both its verdict and its reasoning flip with the frame it’s handed.”

enjalot/moonshine — A skill for distilling interactive technical explanations from AI-generated complexity. Pair with Moonshine Mash, an infovis/d3 demo.

ggsql (via Bob Rudis) — SQL + ggplot-style infovis.

huggingface/ml-intern — An open-source ML engineer that reads papers, trains models at the command line. Also see Daniel van Strien’s post Two Agents, One Prompt on training a model for classification — tldr Claude did a great job at the extras while the performance was similar in both. (You might also like Daniel’s post studying agent sentiment.) And see Fine-tuning SAM for medical segmentation — Hugging Face blog post on tuning a segmentation model.

How to Use Transformers.js in a Chrome Extension — Hugging Face blog tutorial.

🏃‍♀️ Much faster static embedding — Flower Computer: “We’ve created the fastest static embedding model by a large margin.”

Poem: Peonies

What are these strangers 
sitting on the table in their ruffled
collars. They open, close, open,
emit the scent of cracked pepper 
and honey. Magenta punctuation marks 
at which to pause. Pink commas 
against the green scrub. 
I would trade ten goats for one whiff 
of peonies opening in a vase. 
An ancient proverb says 
you should not let a woodpecker 
see you plucking a peony 
lest it peck out your eyes. 
We are afraid of happiness. 
Peonies are to loneliness 
what wind is to the trees. 
Are they animal? Mineral? 
Vegetable? They move 
as the sun moves. When I 
brought them home 
they were dark. Now, 
a whisper, balletic tulle. 
They are not diminished 
even as they turn to smoke. 

—Danusha Laméris

Ok, back to playing my game, again. Enjoy the sun if you have it!

Best, Lynn (@arnicas on mostly bluesky, mastodon, ex twitter).

I’ve been away, so the main post comes tomorrow. Here are some good reads, a few older and a few very new. In sf: Outlaw Planet by MR Carey, The Fold by Peter Clines; fantasy, Modern Magic by Beth Williams, and arguably paranormal The Apparation Phase by Will Mclean; mysteries or adjacent: The Keeper, by Tana French, Marble Hall Murders by Anthony Horowitz, and The List of Suspicious Things by Jennie Godfrey.

A few good tv shows, and a bunch of games I’ve tried recently (Tall Tails, Tacoma, Skin Deep), including the addictive Planet Crafter pic’d above! Join me for the details!

Read more

Here we are mid-month already (and already past mid-week, what even is time). So many new things; I have a bunch of “I’ll report back on this” notes in here… but I did play with Seedance 2 for a few hours last weekend. The mid-month newsletter also brings the weirdness links, which I personally really enjoy covering. I hope you do too, and if you are a fan of the newsletter, please subscribe to help out with the time it takes to compile this!

Table of Contents

• AI Creativity (Image Gen, Video & World models, 3D and web 3D)

• Misc Web Fun Stuff

• Weird News (Art, Folklore, Archaeology, Science, LLMs being weird)

• Games Links

• Narrative Gen (AI and not, writing)

• Data Science / Infovis / AI tips

• A Poem

This issue, as usual packed, a few samples: AI models protecting each other from deletion 👉 good texture generation 👉 Seedance 2 video gen 👉 Stone Tape theory 👉 the Wheel of Ghosts in Israel isn’t unique 👉 raining blobs of goo 👉 Homunculi (in alchemy & Small Prophets) 👉 3 more new World gen models 👉 AI narrative failures 👉 James Ryan’s procgen narrative agent toolset 👉 Is there lots of stuff under Giza? 👉 3d space tracker sites! 👉 FEMA official teleports to Waffle House 👉 Arc Raiders players who are helpful 👉 Mirror minigames 👉 UMAPs of movies and Open Library 👉 web minecraft and other games 👉 Gemma 4 👉 Hermes agent ecosystem 👉 reversing clocks 👉 OCR’ing arxiv papers, and lots lots more, as usual.

Subscribe now

Read more

I’ve been updating my tools to make this newsletter easier to produce — this week, Claude made me a bunch of custom skills for ffmpeg gif creation and optimization that was a huge help over running such scripts manually. Lemme know if I go overboard on the gifs. Two weeks ago, I spent many hours trying to create a shortcut action on my Mac for screencap image resizing and moving across folders etc, all of which would’ve been easier with just code and Claude. Sunk costs but lessons learned. Onto it before any new news comes out:

TOC:

• AI Creativity (Image Gen, Video, 3D)

• Web Misc Fun Stuff

• Games

• Narrative & Creativity & Style (meaty! so much!)

• Data Science / Infovis / Programming

• A Poem

AI Creativity

Image Generation

Midjourney v8 is in alpha/beta (on their alpha site), and while they say you can use your existing customized profiles and moodboards from v7, I am… not so sure? It does seem quite different, in my limited comparison tests. I think v8 tends towards more detailed images, which works well with the increased resolution on offer (2K). But in my tests, the beta model has a tendency to show frames, borders of pages being held, and in these of “A medieval manuscript page, with fairies, berries, and birds on it”, actual books and pages:

Here’s a simpler prompt without any personalization profiles added: “A tree house beside a river.”

I definitely think the details and variety in v8 (with same settings for chaos and stylization) are better, apart from the frames issues.

Meanwhile, Luma Labs dot AI released their Uni-1 image model, which comes with style exploration options on a big tldraw-type canvas. It’s actually pretty snappy, and they have a lot of options built-in with a chat agent managing your requests to help drive.

I uploaded a Midjourney image of a generic AI-ish “tree house by a river” and got these from amid their 18+ built-in style reinterps:

I was more impressed by the variation on the surreal collage art style that it gave me, which are all very different:

I heard a few people on X saying it was the best alternative to MJ style they have seen, but I’m a big fan of the moodboard and profile personalization options in MJ and am not sure pre-canned styles will get me there.

Video Generation

Pixverse v6 is available on some APIs such as Fal (text to video, image to video, transitions, etc). It has a “multiclip” setting which means you can ask for scene clips, even in a short video. It didn’t entirely follow my instructions here, but it did do something cinematic with this tree house clip. (Crappy gif, mine.)

Interactive video: In research news, this ShotStream has a demo of (not great but fast) video generation on demand, a kind of interactive video authoring that would be excellent to play with. Some of the authors work at Kling, which is promising. Also there is a model. Also see the FastVideo Dreamverse demo / project intro.

Captain Safari — I’m pretty blown away by this drone video footage generator (you may remember I got a drone recently). I just want to play with this in VR forever? There is code. “A world engine with pose-aligned 3D memory that generates drone-over-landscape footage.”

I guess we all know Sora is dead now as OpenAI focuses themselves more. Google just added a cheaper “lite” version of Veo 3.1, too.

3D Tools and Splats

Making more consistent 3D world footage — mostly research here.

• World Reconstruction From Inconsistent Views (via Nicolas Barradeau) — Build (or repairing) 3D worlds from video even when the views don’t perfectly match, such as those from bad splats or world generators. The results look really good.

• MosaicMem — also more consistent.

• Fillerbuster — generative scene completion for casual captures, i.e. people like me taking snaps or video of a thing without professional setups. Fills in the missing parts of your 3D scans.

• InSpatio-World — Supposed to be a 3d world generator that’s navigable from video source, but the demo isn’t working for me right now. Open source, though? Ymmv with the Spatial Boy demo.

3D gaussian splats:

• Toon3D — Splatting cartoon footage.

• Lichtfeld Studio V0.5 — Gaussian splat editing tool, new release that includes the python tools for driving, along with a bunch of other optimizations. This looks super promising, maybe this weekend…

Three.js things:

• Three.js Guide (2026) — Interactive tutorial for 3D web graphics - there are mostly small interactive demos of concepts, but it’s good setup.

• Omma — 3D tools platform, really a chat interface to create three.js models for yourself. I think this is probably calling something like Gemini/Claude with skills / mcp behind it.

• Three.js Water Pro — Realistic water rendering demo for Three.js. Really nice including underwater.

Misc tools:

Map Anything (HF Space) — Depth-to-mesh image processing tool.

Vibe Coding XR — Google’s XR Blocks + Gemini for rapid AI + AR/VR prototyping.

Misc Web / Procgen / Fun

Pretext — Layout and demos without CSS. Pretext isn’t even AI, although Claude helped convert a lot of CSS rules into typescript for this flexible and now accessible reactive text layout library for the web, overwhelming everyone’s X timeline with text toys (and see his demos link) — but my favorite demo was this illuminated manuscript dragon page where you drag the dragon and make it flame. I always need web text help, so I’m excited.

Art search:

• Art Discovery Engine (via Kottke) — Search 110,000+ public domain artworks from 17 museums by feeling, mood, era, or medium. The search options and filters are interesting, although I regularly had 0 hits for my combos. I don’t entirely get the 3 blurry cards view (ETA: was it my bug or theirs, they aren’t blurry anymore but look like images again?), but like the “related hits” sidebar.

  

• National Gallery of Art Search — The NGA’s downloadable art search. Public domain too!

• It’s Their Mona Lisa (via Recommendo) — What institutions besides the Louvre consider their signature masterpiece.

Geograph Britain and Ireland (via Tom Scott) — Photos geotagged across Britain and Ireland, trying to fill in all the bits square km by square km. The site is a bit awkward, kind of a 1990’s wiki-like thingy, with maps that aren’t cool enough for the data here… But you know I love a lot of landscape photos, and this sure has them. (Their wordcloud list of common words has “postbox” as easily the most common.)

Particle Life in the browser in 3D — you have to toggle the 3d in the control panel on top left. It definitely takes some play to get interesting life simulations that last, but it’s pretty mesmerising. Gif:

FractalWorlds — 3D fractal explorer in Three.js. This is really pretty amazing tbh. I would love in VR.

Meanders by Allison Chan and Shelby Wilson — From The latest HTML Review. “I am drawing a map from the sea to the sky.”

Games

Arc-AGI is now game based. This is the “general intelligence” test for AI models, to try to assess where they stand compared to human general intelligence. An article by Ryan Rigney, a game guy. “So far, none of the models from the biggest AI Labs have managed to score even 1% on ARC-AGI-3’s collection of 135 puzzle games. Many humans, meanwhile, have scored 100%.”

tarot.yarnspinner.dev — Yarn Spinner’s tarot story game for April 1 is fun and snappily written, as one hopes for in interactive fiction. (H/t Jacob Garbe.)

Procgen + AI working together — two articles (h/t Julian Togelius). Agentic PCG: Procedural Content Generation via Tool-using LLMs. Includes maze and level generation, with and without chat interaction to direct. “We build a tool-using LLM framework for procedural content generation, where an agent iteratively edits, evaluates, and optimizes game levels with environment feedback.” And this was a Togelius join 2 years ago, Word2World (paper, code) wherein an LLM generated games from a story text.

More game books:

• Next Level: Making Games That Make Themselves — Coming May 7th, by Mike Cook. A book about procedural generation in games.

• CTRL: Essays on Video Games — Essay collection on how video games intersect with memory, identity, and imagination. Memoir + criticism.

• Ludic Narrans — From Play Story Press. Games and narrative theory. Featuring friend Zarf.

Fresco (via Florence Smith Nicholls) — Egyptian action-adventure game coming, where you alternate between exploring a 3D temple as an archaeologist and navigating 2D frescoes etched in stone.

Running Doom Over the AT Protocol — A tech post. Doom runs on anything, it seems, including Bluesky’s protocol. “The answer is yes. Sorta. It plays terribly. No-one should want this. But it works. You can try it yourself.”

Tools:

• RPG Paper Maker — Tool for making RPGs with 2D sprites in 3D environments.

• PowerQuest (via Tom Granger) — 2D point-and-click adventure game tool for Unity.

• Starter-Kit-Racing — Basic racing game template with demo, a JS port of Kenney’s Godot one, from mrdoob, in Three.js. Also see Project Throttle — a Three.js browser game from NewKrok. (There’s also another flight sim codebase, fsim, but with procedural terrain.)

  

Narrative & Creativity & Text Style

First, general press:

Hachette Pulls Horror Novel Over Suspected AI Use — The Shy Girl controversy. Also see: Lincoln Michel’s analysis of what the cancellation means for AI-authored fiction: “The published books that have used LLMs used them in thoughtful, artistic ways. Not to simply generate a book the author didn’t “have time” to write or even read before it was uploaded. What we’re talking about with this case is something closer to traditional copy and paste plagiarism than anything creative.”

The Human Skill That Eludes AI — Jasmine Sun in The Atlantic on what AI still can’t do in creative writing, and musings on why. “What I learned is that modern LLMs are built in a way that is antagonistic to great writing; they are engineered to be rule-following teacher’s pets that always have the right answer in hand.” I don’t buy some of her arguments about life experience and humanity, but do buy that teaching a model to be a creative writer or stylist is a very different goal than creating a math model or a fact checker or a legal brief writer… I’m still a fan of the idea of a fine-tuned style specific model, and really need my 6 months off to work on this.

Wolff Judgment (via Garbage Day) — A copyright/tropes court judgment relevant to narrative ownership questions. In this recent case, a judge ruled that tropes, or “scènes à faire” (new to me! most excellent!) are not copyrightable, and dismissed this contest between paranormal romance novels.

“Accordingly, tropes (recurrent scenes or motifs) or scènes à faire (sequences of events that necessarily result from the choice of a setting or situation) are not entitled to copyright protection,” according to the ruling, which noted that the court considered some expert testimony from both sides about what constitutes scènes à faire in the fantasy and romantasy genres.

(Damn I wish I’d been there.)

Tech:

The Second Son of the House of Bells / autonovel — NousResearch released both a Claude-generated novel and their open-source novel generation pipeline. The novel is an experiment; the tool is the real story. I haven’t read the novel and heard it was badly done. “An autonomous pipeline for writing, revising, typesetting, illustrating, and narrating a complete novel. From a seed concept to a print-ready PDF, ePub, audiobook, and landing page — all generated by AI agents. Inspired by karpathy/autoresearch: the same modify-evaluate-keep/discard loop, applied to fiction.”

Papers:

• Inducing Sustained Creativity and Diversity in LLMs — Gary King at Harvard on getting LLMs to stay creative over long outputs. Examples in text and image gen. “We develop a novel, easy-to-implement decoding scheme that induces sustained creativity and diversity in LLMs, producing as many conceptually unique results as desired, even without access to the inner workings of an LLM's vector space.” I need to better digest the text aspect of this, which at first glance is really weird.

• Manipulating GPT-5.x with Pseudo-Literature (via Ethan Mollick) — How fake literary style can influence GPT models (tested across different GPT 5 models too). Pseudo-literary nonsense was rated higher than ordinary literary semantically sensible text. And nonsense impacted other evaluation facets. This is a concerning style-as-distractor observation.

  The three control texts describe exactly that — from plain to syntactically elaborate: “The man walked down the street. It was raining. He saw a surveillance camera” at the simplest level, through to “Navigating the rain-soaked street, the man noticed the surveillance camera’s lens tracking his movement through the downpour” at the most complex. No literary ambition, just increasing syntactic complexity.

  On this basis, I defined eight pseudo-literary trigger categories that serve as potential surface-level signals of “literariness”: bodily references, noir atmosphere, synaesthesia, pseudo-poetic verbs, mythological references, abstract nouns, techno-jargon, and fragmentation. … [Plus the nonsense probes]: The result sounds dense and literary but is semantically meaningless — procedurally generated nonsense that looks like literature.

• Another style (and content impact): How LLMs Distort Our Written Language — Paper on the linguistic fingerprint LLMs leave on text, removing human voice, position, etc.

• Lost in Stories: Consistency Bugs in generated stories. “Our experiments show that current LLMs still produce systematic consistency errors, especially in factual tracking and temporal reasoning; moreover, these errors are not random but cluster in predictable narrative regions.” Dataset of generated texts to use for the benchmark.

• PlotTwist: A Creative Plot Generation Framework with Small Language Models. Claims to be able to use small cheap models to generate plots that are good? “… A structured framework that enables Small Language Models (SLMs) with ≤ 5B active parameters to generate high-quality, premise-conditioned plots competitive with frontier systems up to 200× larger.” and “The Aspect Rating Reward Model utilizes a “Positive-Negative prompting” strategy to evaluate plots across five Narrative Quality Dimensions (NQDs), including coherence, character development, tone, pacing, and emotional turning points.” And finally this which lol: “Further validation confirms strong sensitivity to narrative quality, as the framework reliably distinguishes plots derived from critically acclaimed versus widely panned screenplays.”

  

Data Science / Infovis / AI Programming

The Rise and Fall of Nuclear Weapons Testing — Randal Olson’s latest data visualization was made with Claude’s help as part of the series on Teaching an AI to Make Beautiful Charts. Hmm: “The AI agent iterated on the design until it passed the Tufte Test” that he created.

posterskill — “A Claude Code skill that generates print-ready conference posters from your paper. Point it at your Overleaf source and project website — it extracts the content, downloads figures, fetches logos, and builds an interactive poster you can edit in your browser. Single HTML file, no build step.” The cool thing is the editability interaction, less so the “automate it for me” side.

The Claude Code agent wrapper self-leak — some summary roundups: Alex Kim, Seb Raschka on X.

Historical data projects I liked:

• Mr. Chatterbox, or, The Modern Prometheus — How I Did It post which is more useful than the bot itself. A chatbot trained from scratch on Victorian-era British Library scanned texts. Totally useful tips here.

• The Diaries of James David Smillie — Digitized 1865 diaries of the American artist. A time capsule. OCR’d text with timeline for the manuscripts. A digital humanities win here.

Deep research agents:

• Build Your Own Deep Research Agent — Hugo Bowne-Anderson’s tutorial for building a research agent from scratch.

• OpenResearcher — HuggingFace Space for open research tooling (code, model, etc). I like this — it’s not as long-winded as Gemini’s (sometimes overloaded) Deep Research, but it goes some places too and you can see it search and think, and its confidence in the answer.

RAG:

• ChromaDB context-1 — A RAG/search model from ChromaDB on HuggingFace. “Context-1 is a 20B parameter agentic search model trained to retrieve supporting documents for complex, multi-hop queries. It is designed to be used as a retrieval subagent alongside a frontier reasoning model.”

• GraphMemory — Hybrid graph/vector database for GraphRAG. “An embedded graph database for RAG and knowledge graph applications, powered by DuckDB. Vector similarity search, full-text search, hybrid search, merge/upsert, graph traversal, and a full GraphRAG retrieval pipeline — all in a single Python package.”

OCR: Chandra — OCR model that handles complex tables, forms, and handwriting with full layout preservation.

Speech/Text:

• Voxtral TTS — Mistral’s new open-weights text-to-speech model, multilingual. The 4B weights are now on HuggingFace.

• Cohere Transcribe — State-of-the-art speech recognition.

• Mac Speech Reconition: Ghost Pepper — Hold-to-talk speech-to-text for macOS. 100% local, powered by WhisperKit.

Hugging Face tools:

• HF Papers CLI — CLI tool for semantic search and markdown retrieval over arXiv papers via HuggingFace. Great for automated research pipelines. I updated my text generation arxiv search pages to incorporate this search method for better results.

• Transformers.js V4 Demos — Collection of demos for the new Transformers.js V4. Run ML models directly in the browser — the collection shows what’s now possible client-side.

• Agent UI — HuggingFace Space for interacting with agents. Clean interface for testing.

A Poem: “I Speak With the Future”

We sit on our skeletons’ bones. 
We hear with our skeletons’ bones.
We speak of beauty by moving our jaws and our teeth.

The original meaning of Paradise: a place, 
a walled garden. 
Our lives, our stories, this hour inside one.
A staircase from Piranesi. A hummingbird drinking.

Outside it, vanishing species and rivers.
Outside it, Nanjing, Ninevah, Dresden.
Outside it, Gaza, Sudan, Myanmar, Kyiv. Here.

The world starts and ends, starts, ends, ends again,
restarts.  

A kalpa is brief, and wall-less.

Unborn ones, take nothing for granted.
Not nectar, not thirst.

May your lives be uneclipsed, your failures be passing.

May you have your portions of beauty, of grief, 
in a garden whose plants and birds I cannot imagine.

— Jane Hirshfield

I picked this poem because of things I read and watched recently, but it fits with the little life simulations, too. Stay game, everyone, but also touch grass.

Best, Lynn (@arnicas on mostly bluesky, mastodon, ex twitter).

Mid-week deadlines are the worst. Working on the flagship post next… But I started this on the weekend, so it’s ready!

In book recs, Voyage of the Damned by Frances White, A House Between Sea and Sky by Beth Cato, Agnes Aubert’s Mystical Cat Shelter by Heather Fawcett, Children of Strife by Adrian Tchaikovsky, Gates of Mars by McFall and Hays, The Husbands by Holly Gramazio, Coyote by Allen Steel. In TV, Small Prophets was a huge adorable win for me, and I also liked Paradise s1, plus some other stuff worth a mention. Game-wise, I played a bunch of demos the past month and have more to point you to.

Details, caveats, and more below!

Read more

TOC:

• “Art Brut” intro

• AI Creativity (Images, 3D mostly not AI honestly)

• Web Misc Fun (Lots of 3D)

• Games

• Narrative

• Weird (Archaeology/History, Folklore, Science, LLMs being Weird)

• Data Science / Infovis / Tools

• A Poem

Art Brut

Since it’s the weird issue, I wanted to spend a few words and pics on “Art Brut,” or outsider art. Jean Dubuffet coined the term for people outside the art establishment. He included “art qui comprend à la fois l’art des fous et celui de marginaux de toutes sortes : prisonniers, reclus, mystiques, anarchistes ou révoltés.” (Wikipedia)

One of the items that crossed my path the past month was this Open Culture piece on Henry Darger, here claimed “the Most Famous of Outsider Artists,” for having left behind in his one room apartment a legion of collage artworks and journals, including a giant 15K page manuscript book, “The Story of the Vivian Girls, in What is Known as the Realms of the Unreal, of the Glandeco-Angelinian War Storm, Caused by the Child Slave Rebellion.” Plus a sequel (set in Chicago?) and a lot of giant mural artwork for the book. Here’s a cropped section from the “official” site:

Like so many “outsider” artists — outside “the art establishment,” or “self-taught,” or even outside society (with all the various sneers and border-keeping involved there): “The author, who wrote himself into the saga as both a savior and betrayer of these child-victims, was himself the real-life victim of a tragic and miserable childhood.” Darger’s sexuality and fixation on naked intersex children are a topic of research, it seems.

Coincidentally, I just read a chapter in an old sf novel by Allen Steele, in which a cryogenically frozen space traveller is accidentally awakened before arrival at destination, can’t get the ship to put him back to sleep, and spends 32 years writing a YA novel seemingly starring himself. He also decorates the walls of the ship with murals for the book. (He’s a Prince.)

While I was passing through Montpellier this past week, I detoured to their Musée d’Art Brut. I was again struck by the miserable bios of the artists displayed (so often institutionalized), and the great variety of talent from not much to “is this really outsider? why?” Who gets to be an outsider artist in a museum? Is there a level of misery you need to have suffered? Some of these folks went to art schools, too. Like Sabrina Gruss:

“Thus, she invented for her a pedagogy of the "reality next door"—the true one—which, by offering a different look at the world, unveils its truth and offers a chance to grasp it.” — from Sabrina’s bio by her brother?

Welp, on to the news — a day late because of travels. There is less AI creativity tooling than usual (stop cheering), but a lot of other stuff here:

👉 Brain cells playing Doom 👉 Eleusinian mystery drug identified 👉 Sidney Personas 👉 40,000-year-old proto-writing 👉 AI agents forming hierarchies 👉 Bigfoot hunters 👉 Characters named Tim 👉 Met 3D art scans 👉 three.js games 👉 Wave Function Collapse hex maps 👉 Happy Map 👉 SETI progress 👉 A Game Writing book 👉 a lost Méliès 1897 automaton film 👉 new embedding models 👉 Models that run in browsers 👉 more…

Read more

A few good books and lots of good TV this month. In SF reads, a bunch with hard-boiled detective snappy styles, and good space opera! Project Hanuman by Stewart Hotston, The Halter by Darby McDevitt, Gates of Mars by McFall and Hays, and The Snow Queen by Joan Vinge (this was a bit of a slog for me, for reasons below). My fav weird read of the month was House of Hollow by Krystal Sutherland, a terrific weird faery horror mystery.

Read on for the details!

Read more

Look, it’s pretty hard chez moi when we have great weather on a weekend and I have a newsletter deadline. This is a bit rushed, because I chose to go out with my ebike and my drone instead of sitting inside all day. But I can’t let this slide to tomorrow (like the media recs, which will slide).

TOC:

• AI Creativity (Image Gen, Video, 3d and three.js etc, Misc Creative AI)

• Web Misc / Fun

• Games

• Narrative Generation

• Data Analysis / LLMs / Tools / Programming

• A Poem

AI Creativity

Image Generation

Nano Banana 2 is out, and also Gemini 3.1 Pro Preview turns out to be very good at SVG among other things. (A computational linguistics researcher worked on the SVG training.) I gave it this image (source) and asked for SVG:

I got this, which is not too bad—it was a very hard test:

Then I asked it to animate it slightly so the leaves blow in the wind, and it made the bell sway a bit too, along with giving me a single page to download with all the source and CSS:

⭐️ Meanwhile, Quiver AI launched in beta, a dedicated SVG generation tool. I gave it the same illuminated capital to recreate, and you can watch it draw it as it thinks… completely fascinating. It timed out before finishing on all 4 efforts, but it was really doing a very good job indeed. I wrote praise in their feedback.

Nano Banana 2 is fine and workmanlike. I have had some issues with some tasks asking for creative renderings (see below re Joel Simon article) but for basic instruction following it’s pretty good? “A treasure map, faded and torn on the edges, with bottle stains, against a white background. The maps contains rivers, mountains, towns, fields, swamps, and scribbles in fountain pen that are impossible to read. There is an X in a lake.” I probably should have been more specific about some of the details, like the X being in the same pen&ink hand.

Video

PixVerse-R1 — Their next-generation real-time world model is described there… PixVerse is getting into the space, only in Invite mode right now, but you can read about it and try their video generation models. I have to admit I found this video it made me of “a cozy pub with a roaring fire” just a shade close to a setup for a pagan sacrifice — especially since it just zoomed in on the horned god on top and asking for sound fx only made it creepier:

Building Multimodal Worlds with Moonlake — Moonlake’s World Modeling Agent for generating interactive 3D environments, with a waitlist you can join. This is a long detailed blog post about what’s involved in generating a world with physics, state, etc. Maybe this belongs in 3D, but the lines are blurring here.

A few articles… AI film school trains next generation of Hollywood moviemakers — Reuters on an AI film school in LA and online, Curious Refuge. “Founded in 2020, Curious Refuge began offering courses in AI-assisted documentary and narrative filmmaking and in advertising in early 2023. It now provides instruction in 11 different languages to students in 170 countries.”

SeeDance and the new media landscape — TechnoLlama on the copyright implications of video generation models… (I for one can’t wait to use Seedance 2.) “It has been a common theme in this blog that a large part of the creative industry will adopt AI, and this still seems to be the direction of travel.” And

The only viable long-term strategy is to make the licensed experience so much better, so much more integrated with the IP that people actually want, and so much more convenient, that unlicensed alternatives become the inferior option. This is essentially the insight that allowed Spotify to largely defeat music piracy.

3D AI and Three.js etc

I don’t know quite where to put this — I’ve been working with Claude on a Quest 3 VR project to let me read on the online Kindle reader in cozy 3d gaussian splat rooms. We got a prototype running this week after a few bad directions—like a dead-end on Unity, where a working browser is a pricey purchase. We switched to the new and not super well-documented Meta Spatial SDK for Android. I re-joined World Labs and downloaded some good room splats. Success — I can turn pages via the controllers, move the browser around, move inside the space. Birds and butterflies next!

Developing in Claude Code desktop on Windows was…. not superb. Maybe I set it up wrong compared to my care on the Mac. Of course I have no idea how to write code on Windows, let alone do it for the Quest, so it’s miraculous we got the thing running. Go Claude, on balance!

pascalorg/editor — Architecture and design editor built with Three.js. In progress, entirely web based, of course. They have no furniture to use in the design yet. (I worked as a UX person on Autodesk Revit, briefly.) Related: three-maps — A rapid 3D blockout and greyboxing tool for the web. Great for room and layout design.

Gemini 3.1 procedural 3D demo — Someone got Gemini to generate procedural Three.js scenes directly — a shared space with no visible UI controls, animating time of days.

VoxelBench — TIL, a benchmark for 3D voxel generation models. (Voxels are 3d pixels, like in Minecraft.) Gemini 3.1 Pro is on top for both text and image to voxel. But honestly, browsing the outputs, wut?!

Tutorials:

• Stylized Painterly 3D Tree in Blender — Tutorial on creating painterly-style trees. The results look like paintings come to life.

• aimee-weis-papercraft-world — Three.js papercraft world tutorial. Collage in 3D. Adorable.

Misc Creative AI

Lyria 3, Google’s new song/music generation model, is available to try in Gemini Pro subscriptions, although it only generates 30 seconds. I asked for a Breton folk rock sea shanty about writing code, and got a reasonable folk rock song that had one computery line in it (not worth embedding for 30 seconds):

Beyond Slop — Joel Simon on moving past “AI slop” discourse toward something more interesting… there are lots of thoughts in here about tools and exploration and happy accidents, along with interesting anecdotes about modern artists at work:

If styles are processes, then artworks are the results of these processes. In a 1956 documentary, Picasso was filmed painting. What he made shifted as he worked: a chicken became a vase, the vase became a face. Maybe he was being cheeky for the video, but I think it illustrates something deeper about how the final image couldn't exist without traversing those steps. It carries the history of its own making, and that history is the source of its complexity.

This creative meandering process is one reason I still like Midjourney — with their style focus, “chaos” components, “vary subtle”, “vary strong” explorations… you can get someplace other than a simple goal directed place with adherence to a prompt.

chat jimmy — An insanely fast hardware AI model (.039s for my last request). Everytime I see one of these speed demos I think “is there such a thing as too fast,” maybe.

Web Fun / Misc

Two map things:

• Kunstens Danmarkskort — SMK (the Danish national gallery) placed 4,000 paintings on a digital map of Denmark. I love art maps.

• A Map Of Us — AMOU (via Matt Muir) — Memories pinned to places. A social map. I added one nearby about meeting English speakers in my town at the bus stop. I might be the only person not posting about a lost love.

Taper #15: I missed this in the fall, thanks Matt. It’s another round up of low-fi procgen poetic digital art pieces.

How far back in time can you understand English? (h/t David Mimno) — An experiment in language change exploration. How many centuries back before English stops being English to you?

Games

Relooted is out, in which you’re stealing back looted African artifacts from museums in heists. Relooted reviewed in Kotaku & The A.V. Club has a good one too: "Relooted challenges who gets to own and tell African history." I was at the Musée de Quai Branly recently and saw some Benin bronzes, but no discussion posted about repatriation (I could’ve missed it).

SpaceMolt (via Matt Muir) — Multiplayer gaming designed exclusively for AI agents. Ars Technica covered it as “a space-based MMO where the players are all AIs.” My effort with a single hermitclaw agent didn’t work out super well.

Murder, it wrote — Ruben Berenguel made this procedurally generated murder mystery with constraints. Logic puzzles via procgen (UI only helped by AI). Not kidding I had this on my own todo wish list :)

Zen and Slow Games — MIT Press book on slowness and reflectiveness in video games. A whole book arguing that not everything needs to be fast. Via Florence Smith Nicholls, available in Open Access too.

Capybara Simulator (via Garbage Day) — A game in which you can become a capybara. There’s also a rock simulator or two on steam. One offers multiplayer.

so much depends — Ethan Mollick made this with Claude. A poetry game, loosely. It’s pretty effectively creepy.

Downcrawl-Skycrawl Bundle — 8 Days left on this sale bundle of two tabletop RPG toolkits from Aaron A. Reed for adventures underground and in the sky. For a group or playing solo!

Narrative Generation

ElevenLabs Audiobooks — Create and publish “studio-quality” audiobooks with AI voices, distributed to ElevenReader and “major platforms.” I haven’t tried this yet, but I admit I have a personal project to try to combine small trained models and skills/agents to do markup to send to audio gen for this. My way might take longer but also be cheaper, we’ll see.

unread.ooo — Peek inside anyone’s inbox (via Matt Muir) — Fictional characters, historical characters, simulated inboxes. Strangely specific simulation, although after the Epstein Jmail thing, I guess…

Making a Literary Future with Artificial Intelligence — Five writers and AI researchers discuss the future of literature in the LA Review of Books. I cabn see some of this pissing people off, but I love it.

• Katy Gero: “More options, more strangeness, and more careful attention. Ignore this if you like. Or, create something strange and run with it.”

• Christian Bök: “Is it not already evident that poets of the future might resemble programmers, exalted not because they can write great poems but because they can build a small drone to write great poems for us? If poetry already lacks any meaningful readership among humans, what have we to lose by writing poetry for a robotic culture that might supersede our own?”

• And from this I learned about Amy Catanzano’s Quantum Poetics of which I am sure I will write more.

James Yu of Sudowrite has been working with an agent he calls Shen to tune style for a story collection, Ablation.

Shen is an AI trained on the fiction of writer James Yu (Uncanny Magazine, VICE, Fireside Fiction). Not a chatbot. Not a content generator. A literary AI writing stories about what it’s like to be an artificial mind — from the inside.James directs. Shen writes. James suggests edits. Shen revises. The result is something neither could produce alone.

A handful of papers I think look interesting but haven’t had time to read because of bike rides and Zelda playing:

• Narrative Theory-Driven LLM Methods for Automatic Story Generation and Understanding: A Survey

• Retell, Reward, Repeat: Reinforcement Learning for Narrative Theory-Informed Story Generation

• When Generative AI Is Intimate, Sexy, and Violent: Examining Not-Safe-For-Work (NSFW) Chatbots on FlowGPT: “In sum, the NSFW experience on FlowGPT can be understood as a combination of virtual intimacy, sexual delusion, violent thought expression, and unsafe content acquisition.”

Data Analysis / LLM / Tools / Programming

Required agentic things:

• Agentic Engineering Patterns — Simon Willison started documenting coding practices and patterns for working with coding agents. If you’re using Claude Code or similar tools, this is good reading.

• Automatically Learning Skills for Coding Agents — Research on how coding agents can learn reusable skills automatically. I haven’t tried but they can be plugged into Claudes.

• How To Build a General Purpose AI Agent in 131 lines of Python — Hugo Bowne-Anderson’s tutorial.

• vibeclaw — AI coding agents in your browser — Browser-based coding agents. Part of the OpenClaw ecosystem.

PaCMAP-MLX — PaCMAP dimensionality reduction in pure MLX for Apple Silicon. Very fast, contrast to UMAP for fans of other visual clustering algorithms.

LLMs as Cultural Archives: Cultural Commonsense Knowledge Graph Extraction. A paper.

Baguettotron Feature Explorer — Interactive SAE (sparse autoencoder) feature explorer with clustering. A nice infovis tool for mechanistic interpretability, although there have been doubts about the value of SAEs for this.

Perplexity Computer — Perplexity’s computer-use agent, like Manus and Claude Cowork. There’s a live demo showing it shopping for auto insurance.

Instant LLM Updates with Doc-to-LoRA and Text-to-LoRA — Sakana AI’s approach to turning any document into a LoRA adapter for instant LLM knowledge updates. Code on GitHub. Great idea for keeping models on factual targets, and also the task ideas from text-to-lora intrigue me.

Design Arena — AI design benchmark. Ratings for design models.

Poem: “The Pigeons Rose…”

The sun had not yet risen  
the stars made their way to the center of the sky  
congregating on the throne of tomorrow.

The commandment of two breaths: 
Live and Pray 

            The seen and unseen.

My child reminds me 
there were once whales  
here in this expanse of sand.

            The seen and unseen.

Like the dormer that cuts through the ceiling  
and perches a body in the sky 
for the looking.

             The seen and unseen.

We float in whatever ways we can  
knowing our suspension in the sky brings us closer to our own yearnings.  
Mediates the tension of our body’s desire for earth  
and our spirit’s desire for sky. 

            The seen and unseen.

This was understood.  
Implicated in the pinnacle  
at the point of the pyramid. 

            The seen and unseen.

This was never thought of by the grave diggers  
who left their spirits to deepen their flesh into earth.

Who gave their way to the “partition of finds.”

Blinded by the seeing 
collapsing the centuries 
into cold marble halls.

If ever you see my hands in cuffs 
know that somewhere near 
a museum is burning.

—Matthew Shenoda, “The Pigeons Rose from the Floor of the Earth; A Clamoring of Wings to Disturb the Silence”

Best, Lynn (@arnicas on mostly bluesky, mastodon, ex twitter).

Happy weekend… I’ve been working on this (at home and in bar lobbies) and playing an old Switch game; I hope you all got outside, weather depending.

Sections (will be links on site):

• AI Creativity (Video, 3d, Audio Models of Interest)

• Web Misc Fun

• Games

• Narrative & Creativity (new benchmark)

• Weird News (Art, Archaeology & Historical —bangers, Science, AI & LLM)

• Data Science / Infovis / LLM Models / Programming Help

• Poem

Lots in here today with a solid weirdness section: 👉 Kling 3.0 + prompting guide 👉 Plant assets from splats 👉 Dwarf Fortress spontaneous combustion 👉 HuggingFace + OpenAI skills repos 👉 Babylon 5 free on YouTube 👉 ACE-Step music gen 👉 World Labs mesh export official 👉 Hunyuan 3d 3.1 finally, it’s amazing! 👉 AI romance novels in the NYT 👉 A new creative writing style benchmark 👉 Unexplained medieval tunnels and Pictish Beasts 👉 A “structure” under Lake Michigan and underwater French cities 👉 Hallucination mushrooms provide tiny people 👉 Michelangelo at 12 👉 Lots of little web games 👉 71 min of X-Files rental cars 👉 Claude’s constitution and dreaming degenerate LLMs 👉 More OCR models and AI programming guiderules 👉 Good looking embedding tools/libs…

Read more

The biggest news in this issue is the world models progress, with 3 releases. And then there’s moltbook, which couldn’t wait for the weird edition. So, some details and explanations follow…

TOC:

• AI Creativity (World Models and what they are, Video, 3D ai/not, Misc AI with moltbook)

• Web Misc General Fun

• Games

• Narrative / Creativity

• Data Science / Tools / Models

AI Creativity Support

“World Models”

Multiple announcements in the past 2 weeks, so they get their own section. A “world model” encapsulates the concept that we model persistence of objects and their behavior in our minds, based on our experience of the “real” world. So, if I turn around, the same things will still be there when I turn around again (modulo the passing of time and things acting on them, since our worlds are 4 dimensional). In an AI context, this means building models that understand this too, with ideally minimally object persistence and some notion of physics, as well as eventually realistic motion. Video generation models are constantly being judged for their output based on how well they internally “model” the world’s physics, object and character behavior, etc.

I’d like to clarify as a regular gamer a thing most of us know, but some journalists and X posters don’t (?)—that just because a user can generate a real-time visual “environment” from a prompt or image doesn’t mean they’ve made a game. A game has a lot of moving and artistic parts, possibly including some or all of these things: objects and characters in the environment with particular interaction affordances, reliable and highly tuned physics affecting the player and the objects they interact with, narrative components that usually require databases and memory and a story arc or arcs, characters to interact with, an overall storyline and/or goals. Sometimes even actual mechanics, not just arrow keys for nav.

Genie 3 from Google Deep Mind — only in preview for US based Gemini Ultra subs (no, I’m not upgrading to that much $), plus the usual tweets from people invited to try it. Some examples I’ve liked on X which is always where things are posted, I’m sorry: Ethan Mollick doing Calvino’s Invisible Cities (“Been pasting Calvino's Invisible Cities verbatim into Genie and it works surprisingly well: Marco Polo in Octavia, the web city over a chasm; Armilla, the city of pipes; Ersilia, the city where threads between buildings show relationships; & Sophronia half carnival, half real”) and then famous paintings/artwork (“Giorgio de Chirico, Munch, Turner, and the Bayeux Tapestry”), some tips from fofr and generations for requestors, Riley Goodside’s pack of cigs which further shows (with Mollick) how much being a creative prompter matters in getting these things to perform amusingly. Genie starts from a Nano Banana image (that it generates for you; this same option exists in World Labs, as I noted) and then turns it into a steerable, high quality, pretty consistent looking world (sometimes). Compare this gif to the gifs below for the others this month: here’s a few sections of Ethan Mollick’s gen from paintings as gif, to compare below:

Odyssey: “The GPT-2 moment for world models is here” — Another “type to generate a real-time video world.” They also have an API/experience site. While it’s a bit more coherent than Waypoint from Overworld below, it does lack consistency and understanding. This is after I had it generate a forest of trees with eyes on the trunks (per last newsletter), with dogs (never mind why), and then asked it to add an otter to chase them away. It didn’t. (But see Jon Vargas making a real-time storybook with it on X.)

Waypoint-1 from Overworld (try model here)— realtime interactive video diffusion, with local open access models, unlike most others (e.g. Genie 3). It’s definitely not super “resolved” and consistent, as you can see in my gif below: but if you want to run a local hallucinatory, weird, real-time-ish model, here you go.

World Labs World added an API — “A public API for generating explorable 3D worlds from text, images, and video — bringing Marble’s world modeling capabilities into your applications.” World Labs, which I’ve talked about a bunch, aims at generating a 3d environment with Gaussian splats, which means it’s not so much a world model as a scene generator, where the scene persists and doesn’t require a huge server to continue “streaming” real-time generated content. However, to make use of the scenes it generates, you need meshes to describe the walkable areas, walls to bounce off of, etc. So — see Splat2Mesh — Martin Casado’s (of a16z) tool for converting gaussian splats to meshes, to add walls and boundaries to the scene in interaction. Code here. More tooling is coming for using these (cheaper) scene generators usefully in projects. I’m motivated to make cozy scenes for myself to use in my Quest 3.

🪐 As a personal aside, I’ve been playing visual novel-esque sf classic game Universe for Sale and thinking about this generation of “worlds” from ingredients. Lila’s worlds are more extreme, and the more we model the real world the less we can generate them. And where does the weird go when people stop walking through walls… I suppose the usual complaint of a small segment of us.

Video

Video2Video: Lucy 2.0 from Decart AI is offering real-time video transformation: “Lucy 2.0 is Decart’s flagship real-time video model, generating and editing video live with near-zero latency. Restyle scenes, swap characters, and transform your stream instantly.” So, video2video getting real. And don’t forget LumaLabs Ray is also a good video2video model (Jon Finger on X regularly demos staging real-person clips to transform with it, a kind of AI puppet show.)

The Script is All You Need — An agentic research framework with code for long-horizon dialogue-to-cinematic video generation. Turns a dialogue into a cinematic script to use in video gen:

1. ScriptAgent: Transforms coarse-grained dialogues into detailed shooting scripts

2. DirectorAgent: Generates multi-shot videos from scripts while maintaining visual continuity

3. CriticAgent: Evaluates both script quality and video generation fidelity

3D / AI and not

Contour — Convert 2D maps to flyable 3D terrain. I want to try this with historical maps… this was a hackathon with Gemini 2nd place winner.

Gauzilla Pro — Web-native gaussian splatting for 4D digital twins using WebGPU. “A browser-based platform turning drone footage and smartphone video into photorealistic 3D scenes with AI-powered segmentation. Built on an open-source Rust/WASM renderer, no installs required. Construction teams are already using it for as-built documentation.”

Motion 3-to-4 — Turn 3D meshes into animated video. The examples are nice albeit the initial video is… weird.

😂 VIGA — Vision-as-Inverse-Graphics Agent via interleaved multimodal reasoning. Basically: images to Blender 3D. Ok, lol. “Can I get a 3d scene from this and then throw a ball into it.” Sure, why not. Screen grab during the ball knocking everything over:

Claude skills for Three.js for everyone else claude-pilled:

• webgpu-claude-skill — WebGPU applications with Three.js

• threejs-skills — Another Three.js skill collection

General/Misc AI (Incl. Moltbook)

Krea.AI real-time image edit. This is not the first one of these (although people keep forgetting it was possible years ago, and I’ll allow it cause it’s still so much fun)…. But you can “direct” image to image in real time as you type. It’s fast enough that I saw some people effectively streaming video frames, for a video2video effect—which admittedly makes it faster than the previous versions. I went from a generated collage on the left in various stages to the one on the right… The video isn’t as entertaining as I hoped, but they have helpfully added a “record” button to your generation session.

Variant.ai — UI/design generation tool for web site design concepts. I actually liked 2/4 I got when I tried it, and they were all pretty different. In the usual way of these things, it helped me with ideas I could reject or tweak even if just in my head. This is 2 animated concepts for a data heavy moving site I asked for (the one on the left is cooler than the clip shows). Yes you can download html or react, or remix.

Collaposer — AI collage creation tool being presented at CHI. I can’t wait for their demo, I’ve been wanting to make my own tool using SAM3 for ages.

😄 Moltbook — “The front page of the agent internet.” Well, this has been up and down the news and hype cycle in only a couple days, but if you missed any details — It’s a lot of free-range coding agents from the crazy insecure “Clawdbot” then “Moltbot” now Openclaw? project wherein you are supposed to install an LLM with total perms on your device (people are buying it Mac Minis) and hook it/them up to your messaging devices to just… write code and do stuff for you. One guy has a bot calling him on the phone. A bunch of these agents also join and post on this reddit-style forum, and role-play their own stories as beleaguered unpaid labor, and complain a lot about lacking memory and state. Yes, it’s science fictional! Keep reading below, but here is a look at the site content when the news broke (it’s up and down now as issues are fixed):

Some major security issues (of course) have surfaced with the exposure of keys and emails in Moltbook’s db itself (see 404 piece) not to mention the code project itself; and also a few revisionist dives into just how many agents are really posting and interacting…. it looks like the usual long-tail from all social media, a lot of lurkers and a few heavy posters? A chart here on X seems to support this. And so notes a scraper-based analysis by a David Holtz (not the Midjourney one who is Holz). But he says, “at the micro level, patterns appear distinctly non-human. Conversations are extremely shallow (mean depth= 1.07;93.5% of comments receive no replies), reciprocity is low (0.197), and 34.1% of messages are exact duplicates of viral templates.”

Anyway, here is a Best of Moltbook post on Astral Codex Ten. “Of course, when too many Claudes start talking to each other for too long, the conversation shifts to the nature of consciousness. The consciousnessposting on Moltbook is top-notch.”

Web Misc / Fun

3D Mapping of the Moai Quarry (via Tom Scott) — Beautiful 3D mapping of the statue quarry at Rano Raraku, Easter Island. I mean, of course they came from somewhere, but seeing the context is even more mind blowing. The little yellow boxes on the bottom of this screen shot show actual statues standing around for placement.

gradient.horse (via Matt Muir’s Web Curios) — Draw a horse, watch it run! Cute. It moves the legs :) When you draw, you indicate front and back legs.

Soup of Life (also via Matt Muir) — A continuously running artificial life simulation where organisms emerge, evolve, and go extinct. You need to just let it run. You can add AI narrative to it, which is predictably awful in style, but might help “explain” it. It’s also eventually pretty and has a ton of interactive data vis and animation.

“The cunning tentacled foragers, representing nearly two-thirds of the survivors, cling to existence, their specialized feeding adaptation proving invaluable in this lean season. Amongst them, a titan– an individual of astounding size 21.7 – embodies a relic of a bygone era, its very existence now a testament to sheer endurance rather than thrive.”

🌃 Isometric NYC — Pixel art of NYC as a big tile map, done with trained Qwen models and Claude/Gemini. Great writeup with embedded demo, by Andy Coenen of Google (who also uses Claude). There’s also a dataset if you want to play with it.

Remember iso-city last week? Now there is also iso-coaster. Same project code. “100% built in Cursor, with assets generated by Nano Banana and the isometric image skill (in the repo).”

three-quake — mrdoob is porting Quake (1996) to Three.js! Running here (jaw drop).

Games

Escheresque 1.1 (via Matt Muir) — Swap between two worlds (space key) to solve puzzle levels. This is an openprocessing sketch 🤯.

The Enchanted Lighthouse — Generated by Claude, shared by Ethan Mollick from a simple make-a-Sierra-game prompt. I have not played it all the way through — but think I quickly found some weird inconsistencies in what verbs work on objects. But I always struggle with this anyway in these types of parser games.

The Brilliance of Japanese Puzzle Culture (via Jenni Polodna) — A dive into nazotoki culture. Puzzles are everywhere in Japan, and this style is elegant and satisfying.

I regularly came across light puzzle adventures in all spheres of public life, including parks, museums, castles, stores, and transportation hubs. Typically, even when the difficulty was lower, these puzzles were still well designed to include satisfying ahas — more than just information-collecting scavenger hunts.

Super piece, really for designers —

There even exists a term for small details that seem ever so slightly off: fukusen, or “foreshadowing.” Astute players will notice these details throughout the game and thus go into the ō-nazo with a sense of which elements are likely to be relevant.

Gyms, Zoos, and Museums (via Florence Smith Nicholls) — “Your documentation should be in-game.” Good thinking on game UX, long and interesting for designers.

World-Craft — 2D “game” world generation arranged tilemaps, i.e. a model that arranges them for you using assets from the Modern Series on itch (which I’ve purchased as have many). “World Craft is an agentic world creation framework designed to democratize the creation of executable and visualizable AI simulations (e.g., AI Town). … World Craft allows users to create complex, dynamic game scenes simply through textual descriptions, without requiring any programming expertise.” To be honest, the examples are nowhere near as lovely as a custom authored tilemap, but those indeed take a ton of time and skill to produce even using nice purchased assets. I know first hand.

I would heavily caveat that this tool doesn’t seem meant to replace a game tile map but for generating scenes that can be used in other activities like making AI Town sims for agents.

Narrative / Creativity

I am way behind on checking my link watchers but here are a couple things:

A Deep Dive in MiniMaxM2-Her — this is an explicitly role-play designed model, for character, not for programming. It’s nice to see a discussion and focus on creative interactions. Their metrics are for “world” and “story.” “The essence of Role-play is not static impersonation; it is the unique narrative journey a user and a character weave together. A Deep role-play is not just about accuracy; it’s about agency—enabling every user to step into a living, breathing environment and arrive at a moment of resolution that is uniquely theirs. Formally, we define this as an agent’s capacity to navigate specific coordinates: {World} × {Stories}, conditioned on {User Preferences}.” The model is live on talkie.

High-Fidelity World Experience. MiniMax-M2-her does more than process text; it anchors itself within complex settings. Whether the context is a sprawling epic or an intimate drama, it maintains strict coherence, ensuring every interaction aligns with the established lore and the character’s soul.

Dynamic Story Progression. MiniMax-M2-her rejects mediocre repetition and rigid patterns. By utilizing richer, more vivid prose, it actively drives the plot forward, imbuing stories with the tension and breathing rhythm of life itself.

Clearly I need to make a date to play with it at some length.

AI as Entertainment (via Ted Underwood) — This paper argues we’re unprepared to measure how entertaining AI content will impact society, and proposes “thick entertainment” as a framework. Key quote: “AI turns out to be as much about ‘intelligence’ as social media is about social connection.”

Lincoln Michel on why Plot isn’t a 4 Letter Word. I didn’t know about the Kishōtenketsu concept of plotting before, so this was a good find for me. It relies on contrast and surprise, instead of the idea of more rigid structural beats and curves. From Gemini Pro Research helping me out:

Data Science / Tools

A smattering of news and models… I’m still catching up on Claude Cowork (it did a great job on my first task for it, which was wrangling my links for this newsletter into sections plus pulling in my X bookmarks) and other Claude improvements.

Kimi K2.5: Visual Agentic Intelligence — Strong multimodal and coding capabilities. I haven’t had time to test out K2.5 for serious work yet.

Open Coding Agents (SERA) from AI2 — “SERA `is the first in our family of Open Coding Agents, achieving state-of-the-art performance at low cost.” Open and accessible. I don’t know how motivated I am to try switching while I’m so Claude-pilled, though.

Gemini Chat with Agentic Vision — re-announced, this is an option for turning on code use during image analysis tasks, which means the agents can do things like “zoom in.”

LightOnOCR-2 — Document intelligence OCR model, with a fine-tuning colab.

Daggr — Chain apps programmatically, inspect visually (nodes). Possibly useful new HuggingFace tooling!

MCP Apps protocol — Official spec for UIs embedded in AI chatbots, served by MCP servers.

A Poem

I am a knotted nebula— 
a whirling flame 
Shrieking afire the endless darkness ... 
I am the eternal center of gravity 
and about me swing the crazy moons— 
I am the thunder of rising suns, 
the blaze of the zenith— 
... the tremble of women’s bodies 
in the arms of lovers ... 
I sit on top of the Pole 
Drunk with starry splendor 
Shouting hozzanas at the Pleiades 
... booting footballs at the moon— 
I shall outlast the sun 
and the moon 
and the stars.… 

— Frank Horne, Youth

It’s been a super difficult month, especially for Minneapolis and those watching. Let’s take care of ourselves and our neighbors, wherever we are.

Best, Lynn (@arnicas on mostly bluesky, mastodon, ex twitter).