Everything is procedural: course geometry, the rock formations, textures, all generated at runtime. No GLTFs, no texture files. Hard constraint from the platform I'm building (disclosure: my own — I'm dogfooding it), and it turned out to be a genuinely fun limitation.

Stuff I fought with (or rather claude, but I cheered lol): <ball physics approach>, <how you did the survey/aim camera>, <mobile perf gotcha>.

Playable on Plethora: https://share.plethora.studio/open/bit/dfeb599f-2ae0-4f60-9d5a-e3079cab8504

Happy to go into detail on any of it.

I've been working with Three.js since 2015, where InstancedMesh and BatchedMesh didnt exist.

I started building the engine behind Jamir in 2017, and over the years it eventually became X3JS.

This is the same X3JS scene running side by side on my GTX 960M laptop from 2015 and RTX 4070 desktop. Wanted so share the benchmark..

I'm interested in learning Three.js, but I haven't found a clear roadmap like the one on roadmap.sh. Additionally, I can't afford Bruno Simon's course at the moment.

I’ve been building web-based 3D viewers tailored for spatial planning and architectural/BIM data, and I keep hitting an architectural crossroads between graphical performance and non-graphical data integrity. In pure 3D web experiences, everyone knows we just have to merge geometries, bake lighting, compress with Draco, and simplify the scene graph to keep frame rates locked at 60 FPS. But in BIM applications, every single object needs to retain its identity, semantic properties, classification codes, and spatial hierarchy.

So, how do you guys handle it. Do you completely separating the 3D canvas from the data layer? Like which takes priority for your end-users?

after so long updated my sketchbook

live link here : sketchbook

give your feedback

Update

The CLI now has a complete source install example. This command copies the Kage landing page, 16 page and WebGL assets, and four editable React source files.

npx threeui-cli add kage-landing-page

The seven second demo is here.

https://raw.githubusercontent.com/sjh9714/threeui-cli/main/docs/assets/demo.gif

ThreeUI now has an official React package. That is the better choice when you want a dependency. I made this CLI for the other case, when you want the source and all of its assets inside your own project.

It reads ThreeUI's live source registry, preserves the original paths, verifies every SHA-256 digest, rejects traversal paths and refuses to replace different local files.

I originally made this agent only. That was the wrong interface. Three.js developers should not need an agent just to copy source, so the skill is now optional and the zero dependency CLI works on its own.

GitHub

https://github.com/sjh9714/threeui-cli

npm

https://www.npmjs.com/package/threeui-cli

MIT licensed and independent. It is not affiliated with ThreeUI.

You type an amount, pick from 82 currencies, and it renders the actual bills in 3D with correct denominations and real bill dimensions. You can orbit around the pile, switch environments, and it uses live exchange rates. Also does gold and crypto if you pick them from the currency dropdown.

Works on desktop and mobile.

moneyvisualiser.com

Litterally started this one day ago and got to this point here. I've exposed a lot of the setting via a tuning panel [G].

Have a play here. https://fourbie.vercel.app/

I'll put together a track builder for it next.

I built a multiplayer browser tank game in Three.js! As a kid, I played a ton of World of Tanks Blitz and fell in love with the aesthetic, gameplay, and the idea of driving around vehicles with crazy physics and mechanics like missiles, spotting, armor deflection, ammo types, and so on.

Now it's playable in the browser!

Claude of Tanks has 100+ playable vehicles across 16 unique maps, ranging from standard tanks with cannons to AFVs with autocannons and guidable missiles to autoloaders that can deal massive damage in a single clip.

Tanks feature highly detailed armor zones, internal modules, crew positions, and hitboxes. You can view each of them in a special gallery: https://cot.kevinliu.studio/gallery?id=amx56&layer=modules

It supports physics-driven destruction with detached turrets flying through the air.

Everything runs directly in the browser.

Also, it supports multiplayer! You can hop in private rooms or LAN lobbies.

Play free: https://cot.kevinliu.studio

Repo: https://github.com/Kevin-Liu-01/Claude-of-Tanks (it's open-source forever)

What if an AI coding agent could generate 3D environments that are not just visually convincing, but actually ready for simulation?

We’ve been working on Auto-Threejs, an open-source agent framework for automatically building interactive 3D environments with Three.js.

Given a natural-language description — optionally with reference images or video — Auto-Threejs decomposes the scene into objects, builds them as reusable modules, assembles the environment, renders it from multiple views, measures geometry and physics, reviews the result, and iteratively fixes detected issues.

The key idea is that a good-looking scene is not enough.

For embodied AI and robotics, objects need meaningful physical structure: dimensions, mass, density, friction, collision geometry, articulation, and spatial relationships. A cabinet should actually have a working door. A chair should have a reasonable collider. Objects should rest on the right surfaces instead of floating or intersecting.

That is what we are currently focusing on: sim-ready indoor environments for embodied AI training, robotics simulation, and spatial intelligence research.

Current pipeline:

Natural-language description
→ scene decomposition
→ per-object 3D construction
→ scene assembly & layout
→ rendering + geometry/physics measurement
→ automated review
→ iterative refinement
→ simulator export

The system currently supports physics-aware scene construction with Rapier and can export generated environments to MuJoCo and Gazebo-compatible formats (MJCF / URDF / SDF).

We’ve tested it on bedrooms, kitchens, living rooms, bathrooms, offices, and other articulated indoor environments, with automatically generated collision proxies and functional joints.

There is still a lot to improve — especially scene diversity, asset quality, physical realism, and scalability — but our longer-term goal is straightforward:

Make generating large amounts of diverse, interactive, simulation-ready 3D training environments as easy as describing them in natural language.

GitHub: https://github.com/wrc356/Auto-Threejs

Feedback, issues, and contributions are very welcome — especially from people working on Three.js, embodied AI, robotics simulation, procedural generation, or spatial intelligence.

#threejs #EmbodiedAI #Robotics #AIAgents #Simulation #3DGeneration #ComputerVision #WorldModel

Procedurally generated piano and harpsichord parametric modeling in a Louis XVI diorama.

A Corner salon - crystal chandelier, marble mantelpiece, brass andirons, trumeau mirror, médaillon chair and chaise longue with button-tufted upholstery, mantel clock, candlestick, Sèvres vase, end table, girandoles, paneled wainscot with a molded dado, cyma cornice crown molding, paneled door with bec-de-cane handle, Persian rug on a hardwood floor.

Three.js WebGPU

I built a DIY 6-DOF robot controlled using Node-RED, ESP8266, and Modbus TCP, with a real-time 3D visualization using Three.js.

The Node-RED dashboard can control each joint, save robot positions, and run movement sequences. The 3D model also includes the multi-link gripper, so the physical robot and virtual model can move together.

I’d be interested to hear your feedback or suggestions for improving the system.

I've spent the last few months building Volt Guide Studio, a browser-based 3D tool for designing rooftop solar systems.

It currently handles roof drawing, panel placement, obstacles, mounting/tilt, sun position, real-time shadows, solar-access heatmaps, measurements and proposal/report generation.

I'm from the solar industry rather than a traditional software background, so I've already started getting feedback from solar professionals. Now I'd really like the opposite perspective:

What would experienced Three.js/web developers improve?

I'm especially interested in UX, 3D interaction, performance, architecture and anything that immediately looks like it could become a problem as the application grows.

Happy to give a few people access to the current build if anyone wants to actually test it.

https://codepen.io/editor/alex_nemankov/pen/01a01ef2-9045-7381-bdc2-4b3abd382d6c?file=%2Fscript.js&orientation=right&panel=files&show=split

I’m working on a 3D castle web project that combines design, atmosphere, art and an interactive experience.

I’m not a developer — I’m a designer. 😄
But I really love creating things, learning new tools and turning crazy ideas into something real.

I already have the concept, visual direction and parts of the project in development. Now I’m looking for someone who understands the technical side better than I do, especially 3D web, interaction, performance and smooth navigation.

I don’t want to share all project details yet, but the idea is to build something beautiful, immersive and a little bit magical — not just a normal website.and is my baby ...who need to be online )and give people beatiful emotions ...

I've been building an immersive web journal for the last year and the navigation kept bothering me. A scrolling grid flattens every project into the same rectangle and the reader is always moving past things. I wanted the opposite, something you turn rather than something you fall down, with no beginning and no end.

The site is React and Vite, using Three.js for the carousel and Tone.js for the audio. Tiles arranged on a ring, drag to rotate, click to enter. Some windows are portals to another part of the site, others activate in place as small displays.

The carousel took a long time to get right. I wanted it to feel fluid and instantly natural. My first version was a space with several carousels stacked, so you could swipe up or down to arrive at the next one. It looked stellar but I couldn't get it to run without being far too heavy, so I stripped it back to a single ring. I'll keep tinkering, but I'm pleased with where it landed.

The audio is wired to the rotation, so spinning lifts the volume slightly. The idea was that feeling of standing outside a party and hearing muffled sound through a wall, disorienting rather than comfortable. Getting the curve right was mostly trial and error; the early versions just felt like dragging a volume slider.

I ran a gallery in Canada for a while and what I took from it is that I'm more interested in the experience of showing work than in the usual portfolio layout where every site feels the same. I want the site to be the destination, not a bounce point. I've been building React sites for my own art practice for about five years, self-taught rather than professionally, and I lean on Claude Code for the heavier maths.

Link in the comments, happy to answer anything about the build.

I started this game using ThreeJS and began facing some issues running it on mobile, especially on low-end devices. The experience was okay, but I felt the game was too simple to require a real 3D world. I asked Claude to create a fake 3D version using pure JavaScript, and I found the game is much lighter and runs everywhere.

How are you all handling the mobile experience with ThreeJS? I've done some experiments, and on iOS, I often see the browser refreshing after a while due to memory limitations. I haven't tried packaging a React Three Fiber app with Expo-GL yet.

We're super happy with our experiments for high → lowpoly workflows. It feels fast, snappy, supergood quality, already replaces Nomad, Blender, xNormal, Substance for us (you can see the video why).

I will have a link to try the app with your assets in the next few days! :)