Software Engineer working on #opensource in my free time. See the full project list at https://xitter.cf/t.co/zem1bkyjHJ
Germany
Joined December 2018
- Tweets594
- Following1.1K
- Followers138
- Likes2.2K
CsUtil.com retweeted
I believe building design systems and libraries is the way to think about and scale design with AI tools.
Not enough people are using product + brand kits to build products, websites, and brands at scale with very lean teams—with the aesthetic/vibe they want, without compromising their code stack or their company’s growth needs.
You want the experience you deliver to your users to feel seamless and coherent across every touchpoint. And this is such a perfect moment to build things that way!
The key is getting that operational layer right. Neither underbuilt nor overbuilt—just enough structure to help you and your team build for scale, without making the process frustrating.
CsUtil.com retweeted
This worked better than I expected! Replaced rasterised light probes with ray tracing, and it was fast enough to run realtime. Running in WebGPU + @playcanvas
CsUtil.com retweeted
This is Pele, my new rendering library for diagrams and charts, compatible with the latest Mermaid syntax.
- 100-500× faster than mermaid.js
- 6× smaller bundle
- No dependencies
- Easy to theme
- Responsive to device width
In pursuit of making Obsidian insanely fast and cohesively designed, I always found the Mermaid library to be too big, too slow, and too hard to theme. Pele will be the new renderer in the next version of Obsidian, but you can start using it now.
Pele [ˈpɛlɛ] is named after the goddess of volcanoes and fire. It's open source under the MIT license. Try the playground!
CsUtil.com retweeted
published some updates to stylized maps work:
- google 3d tiles
- @Takram's cloud and atomsphere work, ported the clouds to webgpu/tsl
- webgpu-brush
lyleklyne.com/builds/2026-05…
CsUtil.com retweeted
I found Argon could sketch pretty quickly when using a custom dsl tool it had built for the job.
So I asked it to try animating a scene by sketching the next few frames from scratch each time, rocking the motion back and forth a bit, and iterating to make a short clip.
Then I asked it to add cel shading and transparency and to put it all together onto a background.
CsUtil.com retweeted
Argon has this amazing natural understanding of the world. It can do so many things from first principles.
Here's a pelican riding a bike, using fourier epicycles.
Video made by Argon.
Time-synced Lyria 3.5 music prompted by Argon.
Gemini 4 Argon is live.
blog.google/innovation-and-a…
There's a lot to show you.
CsUtil.com retweeted
Someone recently suggested to me that the reason OpenClaw moment was so big is because it's the first time a large group of non-technical people (who otherwise only knew AI as synonymous with ChatGPT as a website) experienced the latest agentic models.
CsUtil.com retweeted
Have you guys seen the at guy who uses an Apple Vision Pro to do all his installs?
Feels like looking at tech from the future
CsUtil.com retweeted
We gave four frontier image editing models the same photo and 30 edits in a row: Ideogram 4.5 keeps most of the room intact, while the others drift, GPT Image 2.5 Sunburst most visibly.
We've seen some interesting demos of multi-turn editing consistency from the latest image editing models, so we ran our own test: GPT Image 2.5 Sunburst, #1 on our Image Editing leaderboard, against Ideogram 4.5, FLUX 3 and Nano Banana 2.1. Our leaderboard scores single edits; this tests what happens when 30 consecutive changes stack up, with each model editing its own previous output through a 30-step real estate staging sequence: light the fire, add a sofa, repaint the walls, swap day for twilight, and more.
Why are the final results so different?
@ideogram_ai's Ideogram 4.5 and @bfl_ml's FLUX 3 edit locally. On small edits like adding a vase of tulips, we measured that they left 95% or more of the image essentially untouched. GPT Image 2.5 (Sunburst) re-renders most of the entire image on every edit, leaving only about a fifth of the image unchanged, so small shifts in colour and detail compound over turns. Nano Banana 2.1 sits in between: its edits stay local, but the rest of the image shifts slightly and gradually darkens.
CsUtil.com retweeted
Proud to present our latest work on signed distance field collision detection. Interactive demo and code available here: savant117.github.io/sdf-dcd-…
CsUtil.com retweeted
Ever wondered what #GaussianSplatting speed-of-light rendering might look like?
This comes close: 7.5 ms a frame on a Pixel 7 Pro in Chrome/WebGPU, where sorting takes 19.5. The flickery blobs are a per-frame thing; over time they resolve to the correct image.
Too flickery to ship as-is, but a useful stepping-stone.
(I already have a version that resolves much better per frame — not quite as fast, of course.)
It turns out that stochastic rendering of #GaussianSplatting scenes isn't just like, crazy super-duper fast, it also gives us new avenues for doing cool stuff.
For example, how about dynamic point lights with shadows directly on the gaussian splat scene?!
Scene by @DuckbillStudio
superspl.at/scene/23ebe85c
CsUtil.com retweeted
Cool eval. Simply ask an LLM “Land or Water?” and give it a latitude and longitude coordinate as text. Ask 16,200 times, plot as image. The models know. From compressing the internet.
CsUtil.com retweeted
We release Whistle: speech to text in a 16.9MB file, running on the CPU in the same engine as Needle. It mostly beats Whisper base with 9x less file size and 6x speed. Whistle supports English, German, French, Spanish, Italian, Dutch and Polish.
cactuscompute.com/blog/whist…
CsUtil.com retweeted
Quick update on the meshlet side of the render.... still sorting out some animation and caching issues on the @threejs pathtracer but still looks good on medium settings and running fast. Totally unplanned capture so forgive the random orbit controls.
