#ARDY running in-browser with ONNX & #WebGPU, no text embedding, only trying to continue a sequence but it keeps producing "acid rave skibidi" moves... must have missed an important feature/setting, anyway I give up. still a great tool to produce motion sequences offline!

trying out ARDY github.com/nv-tlabs/ardy the viewer port to WebGL went well. I managed to re-target the G1 & Xbot rigs and the precomputed animations run nicely in three.js. now, trying to run in-browser inference with WebGPU is _precisely_ where the shit hits the fan...

"Macros-scans, on a budget." I wanted to create Gaussian Splatting models from tiny objects. It was a bumpy ride I wrote an article to cover the hardware (a toy), the rig (LEGO) and the CV algos used to denoise, calibrate, mask & stack focus. the article barradeau.com/blog/?p=1405

Google just open sourced a new parametric human head mesh! Looks like it covers a good range of identity, expressions and head pose components. I've mostly been leaning on ICT and FLAME, but these definitely have their limitations, so a newer option is really grat. github.com/google/GNM/t...

We are pleased to share our latest research, now published in Nature Communications: “Smart Cellular Bricks: Physical Modules That Recognize Their Own Shape and Repair Themselves.” Blog: sakana.ai/smart-cellul... Paper: www.nature.com/articles/s41... Thread 🧵

finally getting somewhere with focus stacking! top is the input microscope footage during a focus change, bottom is the output stacked focus being rebuilt, the last frame retains the sharp details collected during the focus sweep. not perfect but better than nothing :)

tried to improve the videos quality by using an iPhone to capture the scene and it was a disaster, the calibration freaked out because of the static LEGOs in the foreground & the auto focus consistently missed the subject :/ back to the awful microscope quality + CV preprocess

the recent Fable demos I see feel like cheating & I think of Game of Thrones when Balon Greyjoy asks Theon: "did you pay the iron price for it, or the gold? Did you pull it from the neck of a corpse you made or did you buy it to match your fine clothes?" so, Fable users, Iron or Gold?

given the super brittle setup, the micro scans results are not so bad! I used a turntable (360° around the object, in 2D) and never adjusted the focus. the scarab came out well, it lacks views from above. the marbles were killed by refraction and the calibration / lighting need a lot of love

Over 5,500 mathematical artworks can now be browsed in an attractive new gallery-wall layout on the Bridges website, thanks to the efforts of Nathan Selikoff: gallery.bridgesmathart.org/gallery?seed...

new project: "micro-scans" aka "being Dany Bittel on a budget" @danybittel.bsky.social I bought a toy digital microscope and tried a couple of LEGO rigs over the past week first takes were quite "poetic" but the last one starts being usable :) first takeaway: engineering is hard.

this is a footage captured on an iPhone 12 mini, not sure why expressions have much less influence but it runs pretty well (I deactivated the dynamic lighting though) even the camera works!😄 try it out here barradeau.com/2026/avatar/

for the nerds out there, the Konami code lets you access the debug UI and get deeper insights on how the model behaves :) barradeau.com/2026/avatar/

Nicolas Barradeau@nicoptere.bsky.social · last mo.

I made an online demo of the WebGL Flex Avatars! it took some work to pack everything so that it can run online but I managed to fit the gaussian model, 10 expressions & the mesh proxy in 16Mo :) I also added Gaussian relighting and shadow casting, audio playback and a webcam to drive the avatar 🔗 👇️

I made an online demo of the WebGL Flex Avatars! it took some work to pack everything so that it can run online but I managed to fit the gaussian model, 10 expressions & the mesh proxy in 16Mo :) I also added Gaussian relighting and shadow casting, audio playback and a webcam to drive the avatar 🔗 👇️

cool little mosaic maker kooroshrh.github.io/photo-quilt/

today I tried MeshFlow, my goal was to create a mesh proxy for the gaussian avatars so that they can cast shadows but instead I created a nightmare fuel engine 😅 you can view the results here: barradeau.com/2026/meshflow/ I like the outcome very much though, has a "clay" touch to it 🙂

I managed to get FlexAvatar to run fully in browser with ONNX WebGPU ! 🎉 since inference is now rea-ltime, it's running fairly fast (20-40fps) so I added an audio analysis tool to make the various expressions reactive to sound 📢

now working in glorious WebGL !! added the morph target and wasted a lot of time trying to *ahem* _improve_ the avatar generation ; I noticed the tooth & back were poorly reconstructed so my idea was to synthesize and use multiple views to inject them at fitting time, not my finest idea

the problem with Novel View Synthesis, in a nutshell. ( MV-adapter & SDXL, first image is the input )

managed to drive the Flex Avatar GS models with my webcam, also computed custom avatars (created with FLUX KLEIN) and added a feature to compute a 3D avatar from the webcam and it makes me happy :)

I wanted to run a custom app on a Xiaomi Smart Band 10, what a ride! I wrote an (AI-free) blog post to document the process, it's aptly entitled: "the Xiaomi debacle" barradeau.com/blog/?p=1346 enjoy! P.S. it feels good to spend time on a silly project and blog post :)

This study conducts a Fourier analysis of the wave contours in five paintings by Bridget Riley, in order to estimate the extent and character of their harmonic distortion. github.com/golanlevin/r...

excellent dataviz about fungal networks (mycorrhizal network, mycelium, hyphae) a-hidden-infrastructure.spun.earth/story/a-hidd... quite instructive and well executed

A hidden infrastructure — SPUN

Step through the story: A hidden infrastructure — an interactive exploration of Earth’s underground fungal networks.

a-hidden-infrastructure.spun.earth

pretty cool demo of a Gaussian Splat relighting engine-3vv2rs8cx-playcanvas.vercel.app#/gaussian-sp... not sure how it works but they seem to use a proxy mesh to cast the shadows (which is smart)

this is lovely and very insightful (via James Merrill bsky.app/profile/jame... ) staniks.github.io/articles/cat... that Voronoi decomp is 🤌

Catlantean 3D - Making Graphics Like It's 1993

A deep dive into the asset creation pipeline for Catlantean 3D.

staniks.github.io

I finally made a proper website for Fragment 🎉 Right now, it is mostly a wrapper to the documentation available on GitHub but this is only the beginning! fragment.tools Built on top of @vite.dev, @svelte.dev and Mediabunny by @vanilagy.bsky.social

Fragment | A modern toolkit for creative coding

Fragment lets you sketch, render, and export canvas-based graphics with ease, from quick experiments to production-ready visuals. Built on top of Vite, Svelte and Mediabunny.

fragment.tools

after 6 years repeatedly contracting for Google Arts and Culture, today is my first day working at the intersection of nonsense and fuck-all. can't wait to see what future holds, stay tuned!

need to try this github.com/Eyeline-Labs...

a good writeup / tips / how to use Gaussian splats to make a playable FPS level blog.playcanvas.com/turning-a-ga...

Turning a Gaussian Splat Into a Videogame | PlayCanvas Blog

How I turned a photogrammetric 3D Gaussian Splat scene into a playable browser FPS - with collision, pre-baked lighting, navmesh-driven NPCs and behavior-tree AI, all in PlayCanvas.

blog.playcanvas.com

This tool from NASA uses fractions of the earths surface, taken by satellite, to spell out your name. science.nasa.gov/mission/land... This one says Bluesky