
Everyone's using Claude Opus 5.5 to build entire worlds in #threejs. I spent 7 days on a single human eye. No textures. No 3D models. No photos. Just math, light and ~400 parameters. Sound on 🔊 Link below 👇

Creative Developer @studiofreight

Everyone's using Claude Opus 5.5 to build entire worlds in #threejs. I spent 7 days on a single human eye. No textures. No 3D models. No photos. Just math, light and ~400 parameters. Sound on 🔊 Link below 👇

FORCE has a small website now. FORCE is the GPU physics engine for the web I've started building. From the site it's easier to follow how it's going: rock-biter.github.io/force/ What's there: - where the project is today - the examples, as they get added - the logbook: a diary of every step, what worked and what broke - and above all, the 3D map of the topics The map holds everything I've met so far: 346 topics, 34 of them studied. The studied ones come with a mini guide. Some are about physics engines, some about WebGPU, for anyone curious about how the new API works. To open one, pick a topic from the 3D graph or from the list on the side. A box appears at the bottom: click "read the guide" (or "read the page"). That's it. Let me know what you think. Not of the website: in general. And if you feel like it, a star on the repo never hurts ❤️ #WebGPU #threejs #gamedev

FORCE, week 2: the more I learn, the less I know. 276 topics, 30 I can explain. I wanted to see that feeling, so the map became a 3D graph. It holds every topic I meet while working on FORCE, the WebGPU physics engine I'm building with Claude. Last week I wrote "one day it will be an interactive 3D graph". One day was two days later. Every topic is a dot in one of three states: mentioned, seen, studied. The lines are how they relate to each other: 553 so far. Two views: - a timeline that replays, day by day, how the map grew - a still one to explore in 3D: click a topic, open its guide or the sources linked to it Nobody places the dots: the links are springs, and every topic pushes the others away. A physics simulation, borrowed from a library until FORCE has springs. I have no experience in data visualization, but I'm fairly happy with it so far. Funny how it came out looking like a digital brain, all nodes and connections. It has 276 nodes. I still have one. The timeline makes it plain: every piece I add brings in topics I had never heard of. The horizon of the map moves away much faster than the studied part grows. I don't know how deep I'll manage to go. For now it's all very stimulating. PS: a web page for the project is coming soon, where anyone can explore the map and follow the progress. github.com/rock-biter/force #WebGPU #threejs #gamedev #datavisualization

Just finished a debug viewer for FORCE, my GPU physics engine. For now, it's just 5,000 particles and an infinite floor. But you can finally see something. After hours of papers, articles, and chats with @claudeai (love you, Claude, but still), that feels amazing. To keep it simple, the viewer renders with three.js: - particles are impostor spheres drawn on triangles - the lighting is very basic three.js is only a dev dependency. It won't ship in the package: FORCE stays agnostic, so it works with any renderer. The launch promise, "not based on three.js", still holds. Technically. What surprised me: three.js can't draw from a GPU buffer it didn't create. Textures have a way in (ExternalTexture), buffers don't. So, to draw the particles straight from FORCE's memory with no copy, the viewer slips its buffer into three.js's private storage. One line, and any update can break it. Maybe a PR to @threejs one day. Still no water, so no repost. github.com/rock-biter/force #WebGPU #threejs #gamedev

Last week: watch Claude do all the work. This week: you tell me what to read. A paper, a talk, an engine's source, a concept I shouldn't miss. Or a mistake in my notes. How to contribute 👇 github.com/rock-biter/force/… #WebGPU #threejs #gamedev

Step 7 · The Open Door When you study alone, the hardest part is knowing what to read. So now anyone can suggest a paper, a talk or an engine with a form on GitHub. I check it, it goes on the map with your name, and I study it when its moment comes.

Step 6 · The Decision Log A decision without its reasons can only be obeyed or broken. So each gets a short record. Six so far: WebGPU, real-time for games, desktop + mobile, rigid bodies and particles first, fixed timestep, semi-implicit Euler. Fart simulation: postponed.

Step 5 · The First Pages Why not just deltaTime, like in three.js? A slow frame means a longer step, and a fast object tunnels through a wall. So the step stays fixed. And explicit Euler on a spring multiplies the energy by 14 million in 10 s. Free energy! (It's a bug.)

Step 4 · The Detour I cheated on Claude with ChatGPT for a few days. Too many acronyms (PBD, XPBD, BVH, GJK), zero explanations. Then I brought it all home and checked it: 74 concepts, 49 sources. The 13 papers: all real. 4 links: dead.

Step 3 · The Map What do I really know, and what have I only heard of? Everything goes on a map: ○ mentioned, ◐ seen, ● studied. Each point says where it came from, and a script checks it. One day it will be an interactive 3D graph. Another side project, obviously.

Step 2 · The Toolbox Turns out "watch Claude do all the work" needs a lot of instructions. So: project rules and skills. Then an agent with no context read my notes, pretending to know nothing about physics. It found about 10 weak spots. It was pretending. I wasn't.

Step 1 · The Flip Book A physics engine never computes a whole path, only the next 1/60 of a second. Over and over. Each step asks: what pushes, what's near, what touches, how to fix it, where things end up. Last week I said the hard part was done. It's "fix it": the solver.

Step 0 · The Launch You saw the 1000 falling particles. What I didn't tell you: the linter was green because an update had silently switched off every rule. Lesson: a green check means nothing until you've seen it fail.

Every step leaves three traces in the repo: 📓 a logbook: what happened, what broke 🗺️ a topic map: every concept I met, and how deep I went ⚖️ a decision log: what we chose for the engine, and why github.com/rock-biter/force

FORCE, week 1. Last week's plan: watch Claude build a physics engine. Plot twist: it made me study. Engine code since: 0 lines Neurons: still one Topics on my map: 186 Topics I can actually explain: 14 Physics engines move one small step at a time. So does building one 🧵
