This one was quite a challenge! I spent almost an entire day working on this procedural cave generator for #codevember
akirodic.com/p/codevember202…
☝️ Use on mobile at your own risk. Press "g" to generate more caves!
The #hollow cave geometry is a bunch of SDF sculpts smooth-blended together and re-meshed into meshlets with LODs. Material textures and light is also baked.
I wanted to take it further but I ran into a wall of diminishing returns. It turned out quite nice.
Here is related material look-dev scene:
akirodic.com/p/codevember202…
64
Feather 🪶
Another #codevember sketch. Interactive link below:
akirodic.com/p/codevember202…
#codevember day 12 - Feather
You ask for a feather, you get the whole birb. Pet it here: grgrdvrt.com/works/codevembe…
Here is some folded sandstone canyonlands for #codevember in cinematic 2.39:1 made with #threejs
Am I too late for it @thespite? At this pace I'll be happy if I make it to moss 😅
akirodic.com/p/codevember202…
3
14
922
Aki Rodic retweeted
since there's quite a few combinations, i've added a gallery to browse the geometries and the finishes here spite.github.io/codevember-2…
you can share your favourites!
Aki Rodic retweeted
写真からオリジナルのMetaHumanの顔を作れる「KeenFace」に注目
開発者(KeenTools氏)が新しいWebツールを使った制作デモを公開しました。参照画像をもとに人物の顔を再現し、UE5で使うキャラクター制作につなげる。実在する人物をモデルにしたキャラづくりの選択肢として、活用が気になるツールです
I’m starting to get some of that flow feel as well. I’m also coming to terms with the idea that I’m giving up my craft and learning something entirely new.
I'm deep in flow state with Opus 5.5 right now.
Turns out it was never about writing code, it was about making something amazing.
Code was just the medium. Now with AI, I can build my own medium around the things I actually want to create.
12
405
Aki Rodic retweeted
BRICKAGEN 3000 is out!
it lets you convert a 3D model into a - theoretically - builbable @LEGO_Group set.
capture beauty shots, perform slight edits and export a complete instructions booklet ✨
try it here:
storage.googleapis.com/nicop…
uses @threejs
Aki Rodic retweeted
OpenDLSS 5 comes to ThreeJS. NVIDIA’s DLSS 5 neural rendering network, as @threejs compute nodes, byte for byte. github.com/bhouston/three-dl…
I've decided to revisit one of my old #threejs projects: a real-time visualization of 15 years of my life (so far). Here are 4 cities that are very dear to my heart.
In my original visualization almost a decade ago, I loaded a few years of high-resolution GPX tracks into vertex array buffers and added attributes for date, time, speed, etc. It allowed me to implement custom shaders for real-time filtering. It was a very naive approach but it worked great.
Now that I have a ton more data, the original approach fell apart. So I turned to Ai for help and here is what we came up with...
The GPS data pipeline packs 6GB of GPS data into a 190MB range-addressable binary archive. The data is addressed as a Hilbert-ordered quadtree with precomputed LODs. Nearby tiles sit near each other in the archive.
The loader then range-fetches tiles based on camera frustum and evicts data that is out of view or exceeds the point budget.
It's pretty amazing I was able to do this in a few sessions. Back in the day it would have taken me months for something like that and it would have been a lot sloppier tbh.
But it gets crazier...
I loved how street maps emerged from my tracking data in the cities that I've covered a lot, so I wanted to exaggerate that effect! The goal was to have the street maps appear as fog-of-war around the frequently visited areas. I already had bloom effect on the track lines, so it could double as the fog-of-war street map reveal.
To get the street data, I created another pipeline that downloads OSM maps for areas that I've visited. Since that's more data than I actually need, I also computed intermediate visibility buffers that were then used to cull out all streets that are not close to my tracks. Finally, I used the same range-addressable archive to pack the data and deliver it to the frontend.
More to come...
Superheroes do exist!
I built a Photoshop-like image editor. It’s called Compositor.
robbietilton.com/compositor
I originally built it for myself to get off my Adobe subscription, but decided to release it for free and make it open source. It has all the essential tools I need for compositing, with none of the BS.
I know this workflow is kind of archaic in 2026, but I’m still using it until AI can actually get pixel-perfect on some of the details.
The entire app is 12MB. Photoshop is 6,455MB on my machine.































