More paint tests with the projector and touch frame

Painting with an IR touch frame, projector, and image-paint.constraint.systems

poems are generated via the Savelost algorithm: given an input text, we repeatedly "compress" it by erasing a single letter at a time a text embedding model is used to evaluate deletion candidates & choose the one that least disrupts the "meaning" of the original input bsky.app/profile/maxk...

🌱 Making: More IR frame experiments I got the projector sized to the IR touch frame. I made a desk inset for it and a wood overlay to sit on top. feed.grantcuster.com/post/2025-11...

🌱 Writing: AI Coding checkin Thinking about a couple posts in relation to AI-use and my own recent experiences with it. feed.grantcuster.com/post/2025-11...

I’d like to figure out how to post workspace shots like this of programming projects I’m working on. Screen recordings are a piece of it but it would be fun to figure out a bigger overview.

I ordered this IR touch screen frame not knowing what to expect and am so far pleasantly surprised. I need to map the projector to it but touch detection seems pretty good.

3D blog

Nice! Have you seen Primitive? youtu.be/KVZ9u3BIUBg?...

Codecity

Microsoft Bob

Blogworld

cosine.constraint.systems/text/293929f... Barthes on Cosine

New from me on constraint.systems: Cosine lets you embed and hash any text and view its nearest neighbors. cosine.constraint.systems I wanted to explore what it would be like to use the similarity capability of embeddings as the central organizing feature of a site.

Playing more with embeddings again and getting increasingly tempted to try something with an RSS reader.

Testing/previewing a thing: cosine.constraint.systems/text/293929f... If you want to explore and/or give feedback I'll take it.

This looks like a great pairing with the recent string of Wookash podcasts on C/game UI. Including a dearImGui one podcasts.apple.com/us/podcast/w...

Visualize all the pages (of Byte magazine in this case)

🌱 Making: Exploring an "everything is embedded" mini/fake social network. https://feed.grantcuster.com/post/2025-11-11-08-53-36

tools for directing your own attention

idea: Turbo-C-style initials

🌱 Making: Cosine WIP Working on a stripped down "embed anything and see its neighbors" prototype. https://feed.grantcuster.com/post/2025-11-06-20-16-26-cosine-wip

I've been listening to a lot of playlists lately and wanting a way to learn more about the currently playing artist without pulling out my phone. I put together a little raspberry pi setup to sit on our shelves and display info about the current playing artist.

I made a (Spotify) web player for the excellent Herb Sundays (herbsundays.substack.com) playlist series. It's focused on surfacing a random playlist from the series. I've been using it to put something on in the morning. It's nice to have hand-curated playlists with some intention to them.

🌱 Tabletop: Problem solving https://feed.grantcuster.com/post/2025-10-30-13-50-51-problem-solving

This platform looks fascinating.

Stitched together some Neovim autocmds with Niri window commands so that the app window is bound to the special "app" buffer position and size. One step closer to NeovimOS.

Thinking about reconfiguring my window manager setup to have an always-open terminal on the left. To serve as doc and navigation maybe.

Lately been really wanting to put together a dedicated device (cyber/writer deck-like) and have the entire interface be one document -- lots of patterns here that could be interesting.

🌱 DIY Writer Decks From https://www.writerdeck.org/list-of-diy-writerdecks.html https://feed.grantcuster.com/post/2025-10-14-11-42-27-diy-writer-decks

By changing only a single byte in the code of Mario Paint, the music composer will not stop at the end of the song and instead continue to play the entire contents of the console's memory as music.

@maxbittker.bsky.social I remembered/looked up shaderbooth.com and it is good!

More experiments with Apriltags: a color-picker where x=red y=green rotation=blue freeform physical slider feels interesting

🌱 Tabletop: Levels of perception https://feed.grantcuster.com/post/2025-09-26-20-36-35-levels-of-perception

Book-to-soundtrack experiment - a projector setup where I send the book image off to an LLM for recommendations and then play with Spotify. Meant to be kind of like putting on a record but for objects. (I clipped out the loading times.)

I got a mirror to experiment with the projector

Now I'm thinking about SandScape tangible.media.mit.edu/project/sand...

Trying out april tags. Thinking about how it could complement hand tracking and vision models. This is using github.com/arenaxr/apri... and I know what april tags are because of folk.computer.

Getting a feel for projector and hand-tracking possibilities by experimenting with a color picker. I can use pointer finger without pinch if I give it a path where it doesn't have to go through other things - kind of like a maze.

Trying out hand-tracking triggered buttons and some simple LLM calls. There are definite limits imposed by hand-tracking and projector display - but they're kind of interesting ones.

Table buttons for the projector - hand-tracking and triggering on overlap seems generally promising.

🌱 Tabletop: Robert Irwin https://feed.grantcuster.com/post/2025-09-14-10-04-25-robert-irwin

Playing more with different layouts for the books. Now with a force-directed graph where the nearest neighbor gets linked on click. Think I actually need to have mixed layouts where groups act like shelves?

✏️ I was looking at a short school bus with three windows with three decals applied beneath. I think it was "Quality", "Bus", "Corp" or something like that. And I was looking to see how they had aligned the decals under the windows. Left-aligned on each one? Left, center, middle?

Big limitation versus touch is you don't have touchDown and touchUp. touchDown is doable with fingertip distance check and a activation timer. But touchUp is tough - you don't want to pan when you're moving back to fist but you can't do activation timer without adding latency to all panning.

First step towards trying hand tracking + projector

Book images -> generated summaries (that I checked) -> embeddings -> force-directed graph with nearest neighbor. Results are pretty good -- weaker where there's not a good match available (I should put more books in).

Got the books projecting on to the table:

Taking another run at building a spatial canvas app that is mostly conventional react state but with a render loop that sets camera and block positions directly on the elements in a game-engine style render loop -- with autoTweening when target and position don't match.

🌱 Tabletop: Flattening https://feed.grantcuster.com/post/2025-08-30-16-10-30-flattening