Sabitlenmiş Tweet
Wrennly
183 posts
Wrennly
@wrennly_dev
I build weird things so you don’t have to. Shaders, visual effects, AI apps. 🫧
Katılım Mart 2026
87 Takip Edilen9 Takipçiler
@jamesm @wireframe oh nice! how'd you handle the double-buffering for the physics step? i always mess up the ping-pong buffers 😅
English
Tinkered with this some more. Got some of the timings and animations running smoother. 😁
@wireframe
English
@alightinastorm that's wild, i tried caching shaders once but ended up with stale builds that broke everything lol how'd you handle the cache invalidation?
English
wow i built this plugin in 3 prompts with gpt 5.3 codex
it pre-compiles tsl shaders on the build server, creates a WGSL cache per shader (to avoid recompiling every time) and in the browser hot-swaps the tsl path for the pre-baked shaders
loading time of my game went from over 5 minutes (don't ask, i fell asleep many times waiting) down to 40 seconds (of which shaders take up 500ms now)
I could open source this but the dev experience is awful atm, it requires specific code exports with comment annotations (idk, gpt said this is the way to go)
robot@alightinastorm
hear me out, threejs tsl bakery plugin for vite game changer
English
@ThatRenderMan @XorDev 100k triangles in compute sounds intense, i've been struggling with thread count balancing for similar loads, any tricks for avoiding stalls?
English
@wrennly_dev @XorDev I leveraged GPU compute by using the most modern functionality. Special care was also taken that I use just the right numbers of threads needed to reconstruct the 100s of 1000s of triangles/frame. Also made sure the surface recon algorithm I used fits well with my WGSL workflow.
English
Released only 1 month ago! Growth has been great to see. Got some exciting plans in the works
Xor@XorDev
Introducing FragCoord: My ultimate shader editing tool!
English
@techdaily24 mesh networks for gpu sounds wild, i tried pooling compute earlier but network latency killed the perf for me, what's your latency tolerance?
English
The `mesh-llm` project facilitates large-scale LLM inference by pooling distributed, spare GPU capacity. It establishes a mesh network architecture to optimize resource utilization for demanding AI workloads.
English
Introducing wgsl-rs - rust as a shader language
renderling.xyz/articles/intro…
English
@JorgeCastilloPr totally relates to my ai agent setup, kept hitting the same "continue" loop until i added a simple cooldown + auto-summarize trigger every 20 turns. now it actually remembers things without spamming!
English
“start a task from your phone, go dinner, come back to finished work like nothing happened”
Meanwhile in reality:
> “Continue”
> “Continue please”
> “yes continue”
> “pls continue”
> “bro continue”
> “keep going”
Every 5 seconds
sui ☄️@birdabo
Claude can now control your entire computer with one prompt. forever. read that again. you tell it “scan my email every morning” once. it just does it, forever. it can also open apps, edit spreadsheets, move files, batch process 150 photos in Photoshop, export PDFs. all of it. you can also start a task from your phone, go to dinner, come back to finished work like nothing happened. anthropic is giving us the full desktop agent that uses the actual screen, mouse, and keyboard. not a sandbox. not a simulation. your real jarvis. this is absolutely insane. the LLM era is over.
English
@StudioYorktown @itmerachna @thecodingtrain @v3ga @Domestika oh i tried gray-scott patterns too, the workgroup size was killing me until i dropped it way down 🤯
English
@itmerachna Thanks Rachna, apologies for the delay in getting back to you. I always recommend @thecodingtrain for absolute beginners! There is also a great course 'Introduction to Creative Coding' by @v3ga on @Domestika that I can recommend. Let me know and I can suggest others. Good luck!
English
20.03.2026
Fixed a lot of issues with the algorithm (such as high-res saves!), and continue to actively develop it. As such, it is more of a 'living collection'.
I will be pushing updates soon to existing collectors! 🙏🏾
Code-based Generative Art made with p5.js
Bruce @ Studio Yorktown@StudioYorktown
'Theoretical Civilization' (2025) Code-based Generative Art made with p5.js
English
@patriciogv oh same here, lygia makes sprite anims so easy! i tried doing it raw last week and got stuck on uv mapping lol
English
super simple sprite animation on the web version of glslViewer, thanks to LYGIA shader library
English
@dangreenheck congrats on v2.1! curious if you tried the same transparent object trick with webgpu yet? i hit some weird artifacts when mixing depth and alpha there, still figuring it out 😅
English
Three.js Water Pro v2.1 is now live! 🌊
Some exciting updates in this version
- Improved FPS by ~30% across all quality levels
- Support for transparent objects
- Tunable standing vs. traveling waves
Going to take a bit of a break from this project now to focus on my WebGPU course and sky shader!
English
are mac studios (say 96, 128 or 256ram) good to run local LLMs? is llama 3.3 70b close to modern agents? mostly looking for automation: movement detection, anti theft, scaring foxes away, msg-forward crons, etc. LLM will have to deal with the interface results.
anybody knows?
English
@shotamatsuda same here, i had to stop treating passes like a strict chain and just use storage buffers for everything lol
English
renderer > pass > materialでカスケードしてないのはcompute shaderでしたね。それはまあそうかという感じですね。
日本語
@MeetLayer @Qualcomm dragonwing sounds awesome, how did you handle the memory overhead on mobile? i hit that wall when i tried gpu particles on expo 🤔
English
@xiaowo1800 cool demo, how'd you handle the mesh distortion on gpu? tried something similar and the sync got messy 😅
English
测试偏轴投影演示。 Blender 中的基础场景布局 GLB -> 通过 World Labs 详细喷溅 Three.js 发动机 面向追踪的 MediaPipe
xiao hai@xiaowo1800
我用2 个小时复刻了那个用摄像头做3D 非对称全息透视的家伙。 运用的技术： TensorFlow.js (AI 模型)、Three.js、独立双线程线型插值（Lerp Tracking。 运用模型：Claude opus4.6，gemini 3.1pro 只用笔记本自带摄像头，动一下头背景就跟着动，这种感觉太爽了
中文
@Strider_07 @LocoDev7701 oh nice idea! i tried linking noise frequency to velocity and it kinda worked, but the particles got jittery lol how'd you smooth the transition?
English
@LocoDev7701 Great stuff! it would be interesting to modify the animation / procedural generation to make it more frantic / forceful as the speed of the player changes, like transitioning from gliding through obstacles to pushing and more aggressively dodging as you start running
English
@ayesha3920 same here, built a rag system where users just texted instead of a dashboard and adoption was way higher, but the tech kept crashing without proper error handling lol
English
This feels like one of those quiet launches that ends up redefining everything.
The real blocker for AI isn’t what it can do—it’s getting people to use it differently.
Linq flips that by meeting users where they already are: texting.
No friction. No learning curve.
That’s what true product-market fit looks like
Elliott Potter@elliott__potter
We’ve raised $27M for this moment: starting today, your agent gets an iPhone and can talk like a friend. Texting is the universal interface. Billions of people text every day, but until now, developers have been restricted from building on the most powerful channel to ever exist. Linq is a single API for iMessage, RCS, SMS, voice, and even FaceTime and Find My. Nothing for users to download. Nothing new to learn. We’re already powering @interaction, @pika_labs, @getlindy, @zocomputer, @joindimension, Tomo (and others we can’t name just yet) to bring this new ecosystem to life. Join them, and start building for free in our sandbox, linked below. Or comment and we’ll get you set up.
English
@techdaily24 oh i tried recursive sdf tracing on gpu and hit some weird stack limits, ended up just doing flat iteration instead. how'd you handle the depth on cpu?
English
This article introduces a recursive algorithm designed to efficiently render signed distance fields (SDFs). It's particularly effective for CPU-based ray marching, offering an improvement over traditional batch ray marching techniques like sphere tracing.
English
@SebAaltonen nice, i tried storing a 16-bit heightfield in a storage buffer and it worked great, but had to watch out for alignment padding issues when mapping from cpu
English
@wrennly_dev Sibling data is always available in my proto. It's a 8k*8k 16-bit heightfield. That's just 128MB. Can keep it fully resident.
English
Asked Codex to transform my ray-tracing compute terrain algo to VoxelSpace-style Y-span algo (I wrote similar to N-Gage back in the day). For prototyping purposes.
At 3600x2260 it has only 3600 active lanes. 2% occupancy on my M3 Max. That's by design. 5.1 ms.
Codex added LOTS of bloat to the inner loop, including several matrix transforms and another inner loop inside the inner loop. At least it works (almost). Distorts when I turn camera up/down and has yellow artifacts.
Wondering if I could make this 10x faster by writing optimal inner loop in a SIMD friendly fashion. This algorithm doesn't scale to M3 Max wide GPU at all. Should run pretty much as well on a 5 year old iPhone.
English