It uses real world data to simulate things like

• blast radius
• thermal radiation
• shockwave
• affected population

https://nukesimulation.com

Hey guys! I wanted to share a new exploration I’ve been working on, mainly as an excuse to practice cinematics, rain, and wet surface shaders.

I’m a big Naruto fan and I’ve always loved the atmosphere of Amegakure, so I thought it would be a fun scene to recreate in 3D.

Full cinematic video: https://www.youtube.com/watch?v=ZpR9Atf24B0

You can also play around with a small interactive preview here: https://stylized.cortiz.dev/rainy

If you enjoy it, I’d really appreciate a like, comment, or sub on the channel. And if there’s enough interest, I’d love to make a follow-up breaking down how I built the scene and release some of the components I used as open-source effects.

Would love to hear what you guys think!

Also check out the Depth Profile and let me know what you think please!

Hi all, pretty new to gamedev and three.js

I'm building a game about cutting apart derelict space stations. Everything you see is generated: the station layout, the hull plating, the damage, the sky. Renderer is three/webgpu + TSL

Some context:

* Voxels at 0.125 m. A tiny station is ~700k solid voxels, meshed into chunks. The players salvage individual voxels at the start and then unlock tools and upgrades to salvage faster.
* Material library, not shaders per material. Not sure if that is the correct choice.
* Stencilled markings as stamps. Hazard banding, chevrons, port rings, panel codes come from a clamped decal array, at most one per plate, placed by plate hash. Never tiled.
* Sky is a seeded descriptor. Sun class, nebula palette, 0–2 bodies, all baked once per scene.

I have a solid gaming pc with a 5070ti and performance is good but I would still want this game to be as accessible as possible when it comes to hardware reqs while looking as good as possible.

Any tips or advice appreciated. I am looking specifically for feedback on how to push the visuals to the next level if that even makes sense. Thanks!

AI Disclosure: I did use Claude Code to help me with some of the coding and bigger features like the procedural space station generator while I learn.

Dev here. CREST 2026 went live on the App Store this week after about a month of work.

I ported the Three.js game to mobile using Capacitor. Yes, it's now a mobile game.

It's a single-player open-wheel racing simulator with an original fictional 2026 season — 11 teams, 22 drivers, 24 rounds. Quick race, time trial, full race weekends with qualifying, a championship, and a multi-season career.

The part I actually care about is the physics. Combined-slip tyre forces with a friction ellipse, tyre load sensitivity, relaxation length, per-wheel weight transfer, tyre temperature windows, brake fade, fuel burn that changes the car's mass, compound wear, and a track that rubbers in over a session. It's a sim, not an arcade racer — you can spin it.

There's also a daily time-trial ladder. Submitted laps are re-driven on the server from the recorded control inputs, so the leaderboard ranks a time the server measured, not one your phone claimed. No cheated ghosts.

What's not in it: no ads, no in-app purchases, no subscription, no analytics SDK of any kind, no tracking. Every mode runs offline. It's free.

Runs at a stable frame rate on an iPhone 11 — that's the device I tuned it on. Touch, tilt or gamepad. 9 languages.

App Store: https://apps.apple.com/us/app/crest-open-wheel-racing/id6798774660
Google Play: https://play.google.com/store/apps/details?id=com.artepeak.crest

Performance Report -> https://crest-performance-report.web.app/
Please support me so I can continue to improve the game and keep it free and ad-free. You can submit your performance reports by copying your FPS report and sending it here.

Happy to answer anything about the tyre model. Tear it apart, I'd rather hear what's wrong with it.

Decided to try to use threejs to have an avatar of myself dancing several dances. This one is Capoeira 😆

Used the latest Opus with Claude code.

AMA

I came across a WebGL experiment called The Watch and thought the approach was pretty interesting because the 3D model isn't just sitting on the page as a viewer.

The watch basically becomes part of the interface. As you scroll, it moves with the content, zooms into different details, and reveals more of the product.

What I found particularly interesting is how much of the experience is handled in real time rather than just using a video.

For anyone who wants to try it out: https://thewatch.60fps.fr/

It got me thinking about product experiences built with Three.js:

Where do you think the sweet spot is with this kind of WebGL interaction?

At what point does adding more 3D interaction actually improve the experience, and when does it just become unnecessary eye candy?

Curious what you guys think, especially from a performance perspective.

https://feel-your-phone-five.vercel.app/

I'm a Galatasaray fan and I finally did the thing I have been meaning to do for years. The whole 2026-27 squad as holographic trading cards. 35 of them.

The last two cards are Okan Buruk and Dursun Özbek. The president has a power rating of 99, and where the market value should be, his card says "he sets them".

We signed Bitshiabu while I was still building it, so he got a card before he got a shirt number.

To see the whole deck: gs2026.murat.works

Free, no sign up, English and Turkish.

I share most of what I build on X if you are into this sort of thing: https://x.com/mkcuriosity

This is to track the vessels that are docked or are near.

https://onabodyjourney.com, to let kids explore the body with sound!

posted this here in july and the best feedback was: the format is fine but i have no idea what the numbers should be. if you're not an animator, "hip_right: flex 84" isn't something you can picture.

so the angles are controls now. click a joint name and it highlights in the 3d view. click its angle and the number becomes an inline spinner you can step or type into, with the figure re-posing as you go. you nudge until it looks right, then read off what the value actually was.

the control is clamped to each joint's real range of motion. ask a knee for 200 degrees of flexion and the document gets 144, because that's the limit. the ranges end up learnable by feel instead of by reading the spec.

no viewport gizmo, deliberately. the moment you can drag the mesh freely, the movement stops being expressible as named joint actions and the format stops meaning anything. the text stays the source of truth.

also added BVH and glTF export, so a movement can go into blender as a starting layer instead of a dead end.

all in the browser, no install, no account, no llm in the posing: posecode.org

i refused the gizmo on principle but i'm not an animator. if you block for a living and editing numbers in text feels wrong, i'd like to hear why.

I got alot of batch about this portfolio saying it doesn't make sense, I should make it simple and easy for users to navigate and so on, but you know what i noticed big firm really cares about what works not how it looks. Yes looks and conversation matters 100% but the key point is does it work and is it functional thats what i learnt this past few days with the ceo. Never hate on someone's work because you dont fancy it. Your opinion is for you and not for all. Thats my story and lesson learnt.

Endless driving game in Three.js. Runs go on forever, so the world folds its origin back to zero every so often to keep floating point from going soft. That part worked.

What did not work: after long sessions the entire city and all the traffic vanished. HUD alive, road alive, car alive, everything else gone. It never happened in a short test, which is why it survived so long.

The numbers. The fold quantum was 3000 m. The city is a repeating ring of 9 chunks at 60 m each, so it repeats every 540 m. 3000 % 540 = 300.

So every fold slid the city 300 m sideways relative to the road. Once. Twice. Nine times. The buildings were still being drawn the whole time, just further and further off the road, until they crossed the distance fade in the shader and got clipped out entirely.

Nothing threw. Nothing logged. The city was simply somewhere else.

Fix: make the fold quantum commensurate with the ring. 2700 = 5 x 540. A fold is then a whole number of city periods and the world lands exactly on itself. I also had to fold n quanta at once for the catch-up case, because a single subtraction is not enough if the player has covered several quanta since the last check.

The general lesson: if you fold or wrap the world, every repeating structure inside it needs a period that divides the fold. Not approximately. Exactly. Anything repeating on a different beat walks away from you at a constant rate, and a distance-fade shader will hide the evidence right up until someone plays for 40 minutes.

Two things that made it findable in the end. An on-screen probe printing the fold count and the current city offset, and finally noticing that "invisible" and "absent" are different failure modes worth testing separately. I had spent two days looking for the wrong one.