iiinova — where your digital self keeps evolving

Nicolas Barradeau @nicoptere.bsky.social · 11m

my last contract for @googlearts was released today 🎉 it was a collab with artist @xin_liu_studio she wanted to use Neural Cellular Automata to create a world simulation, it was fun (& somewhat tricky too). it's called "nom nom" g.co/arts/cSRW3yq...

g.co

Xin Liu: nom nom - Google Arts & Culture

Enter artist Xin Liu’s nom nom, a simulation driven by Neural Cellular Automata and a Gemini weather system. Millions of interacting particles form an evol...

000

Reposted by Nicolas Barradeau

Theo Priestley @theopriestley.net · 29/09/2026

Every AI CEO sounds like this.

61024291

Nicolas Barradeau @nicoptere.bsky.social · 27/09/2026

attended a lightning talk by Isabel Fang about a project she did with David Ronai (@artificialisabel & @makio64) "the battle of AI" www.youtube.com/watch?v=zVJq... found it amazing & started a RL model to micro-manage units so that the LLMs can take high level decisions but then, factions happened!

000

Nicolas Barradeau @nicoptere.bsky.social · 24/09/2026

I may have taken Geometry-Native Autoencoder too far but the result is glorious! the model has a 81 frames attention window where it knows the reference image, here I tried to chain 4 * windows and the poor fox has to shapeshift to remain in the frame 😂 jiah-cloud.github.io/GAE.github.io/

030

Nicolas Barradeau @nicoptere.bsky.social · 23/09/2026

I released the code to test the BRICKATOR 3000 locally & retrain if you're feeling adventurous (untested) github.com/nicoptere/br...

021

Nicolas Barradeau @nicoptere.bsky.social · 22/09/2026

recently, I've been thinking more and more about a hidden web, untouchable by machines. a kind of fortress really.

220

Nicolas Barradeau @nicoptere.bsky.social · 22/09/2026

generating models from the sets detected in image. I'm using the 2026 BrickNet (kulits.github.io/BrickNet/ ) with the 3 different assembly modes, I must be doing something wrong 🤔 it's not 'bad' but it's very random...

120

Nicolas Barradeau @nicoptere.bsky.social · 21/09/2026

I managed to get my Lego brick detector to be "less dumb". it works fully in-browser with 2 models: YOLO-seg & a finetuned BrickNet. I had to create synthetic datasets, I learnt a lot... anyways, without futher ado, behold the "Brickator 3000" ! live demo => barradeau.com/2026/lego/

050

Nicolas Barradeau @nicoptere.bsky.social · 18/09/2026

I'll be part of the Jury of #TripoAI 1st world building hackathon. it's a worldwide contest co-organized with other cool companies and if you're into VFX, games, XR or 3D printing you should give it a shot :) more info below

121

Nicolas Barradeau @nicoptere.bsky.social · 17/09/2026

ok, I'm not saying it's good, just saying it's "encouraging"

063

Nicolas Barradeau @nicoptere.bsky.social · 17/09/2026

trying to make sense of the LDRAW nomenclature from LEGO models ; it's a taxonomy nightmare. special bricks should have been made illegal a long time ago (yeah I'm looking at YOU "weird shit cluster"!! )

260

Reposted by Nicolas Barradeau

Laetitia Nanni @letinani.bsky.social · 08/09/2025

Finally found a short moment to put together a little webpage! mostly just an excuse to play around with some WebGL GPGPU + render target combos in Three.js. By the way, I’m a web creative developer and sometimes Arduino tinkerer working with artists and installations ✨ www.laetitiananni.fr

1113

Nicolas Barradeau @nicoptere.bsky.social · 13/09/2026

By *ahem* « popular » demand, the pics I took on stage. Such an incredible crowd (speakers and attendees alike) such a lovely event ❤️ @threejsconf team gets all my love and respect, see you next year 😘

0120

Nicolas Barradeau @nicoptere.bsky.social · 11/09/2026

I gave a 2 minutes « talk » at the threejs conf, it was fun youtu.be/01hFLAxtA6k

1111

Nicolas Barradeau @nicoptere.bsky.social · 08/09/2026

I made a volumetric smoke demo in #threejs earlier this year, it's essentially a raymarched 2D fluid simulation. the upcoming @threejsconf is the perfect occasion to release it :) it's called "Ashtray Requiem" and should work on mobile (somehow). barradeau.com/2026/ashtray... #threejsconf #webgl

barradeau.com

Ashtray Requiem

Interactive 3D Smoke Simulation Experiment

151

Nicolas Barradeau @nicoptere.bsky.social · 07/09/2026

drumroll ! who's going to threejs.paris ? 🤚 I'll arrive on wednesday evening, I'll give a "lightning talk" on thursday or friday, I may even take part to a panel. anyways, hope to see you there! ping me if you're around 😀

threejs.paris

Three.js Conf Paris 2026 | September 10-11

Join the first conference dedicated to Three.js in Paris on September 10-11, 2026. Two days of talks, panels, networking, and a Seine after party.

1100

Nicolas Barradeau @nicoptere.bsky.social · 04/09/2026

reminder that, under certain conditions, a video model (here H3) can be used to create stable and qualitative 360° videos, much like a world model. these videos can later be used "as is" or to create 3D models (Gaussian Splats). below: input image => atlas of 16 video frames.

010

Reposted by Nicolas Barradeau

Sakana AI @sakanaai.bsky.social · 03/09/2026

Can a general-purpose vision model tell real from AI-generated images with just a simple decision rule on frozen features? Our #ECCV2026 paper says yes. Blog: pub.sakana.ai/percept-lens/ Paper: arxiv.org/abs/2608.18523

1104

Nicolas Barradeau @nicoptere.bsky.social · 02/09/2026

creating people is not trivial with text to images models (here @nvidiaAI 's SANA), one has to actively inject imperfections: asymmetries, wrinkles, eye bags, scars, freckles ... to get actual human-looking humans. same prompt w/wo injection

020

Nicolas Barradeau @nicoptere.bsky.social · 01/09/2026

progress on the Non Photorealistic Render, now with more various objects, a visual debugger to check source images and meshes, improved curve placement etc. I really like where this is going 😄 each models ~ 280Ko for 8K splats and 150 splines

141

Nicolas Barradeau @nicoptere.bsky.social · 31/08/2026

now THAT's more like it ! decimation was intense ( 1M -> 50K ) but the result is very close to the intended behaviour. it makes use of Hunyuan3D-paint-v2-0 as a middleware to synthesize 6 consistent orthographic views then unwraps uvs and bakes images into a single texture✨️

010

Nicolas Barradeau @nicoptere.bsky.social · 31/08/2026

not saying it's bad, just saying it's funny 😄 trying to transfer the colors of a gaussian splat onto a mesh, the source GS and mesh were extracted with @tripoai models (SG and Splat), both super qualitative, just that the meshes are untextured (see last pic) hence my attempt 😀

010

Nicolas Barradeau @nicoptere.bsky.social · 30/08/2026

Non Photorealistic Rendering progress, it starts looking good 😀 at least that's what I had in mind ! ~158 Kb geometry, need to optimize the splines. rendered in #threejs

4545

Nicolas Barradeau @nicoptere.bsky.social · 29/08/2026

r&d : image -> ⋆˖⁺‧₊ magic process ₊‧⁺˖⋆ -> lightweight Non Photorealistic model

020

Nicolas Barradeau @nicoptere.bsky.social · 28/08/2026

Minimax H3 video & audio, FLUX images, automated process, raw result. (models hasn't seen enough spider-crabs 😅 )

010

Nicolas Barradeau @nicoptere.bsky.social · 27/08/2026

still no idea where this is going, just throwing all the models I can at a video, compositing the result into a single MP4 + VTT to display data in realtime in #threeJS and now I have way too many parameters to play with ... 😅

080

Nicolas Barradeau @nicoptere.bsky.social · 27/08/2026

Minimax H3 video & audio, FLUX images, automated process, raw result.

010

Nicolas Barradeau @nicoptere.bsky.social · 27/08/2026

Minimax H3 video & audio, FLUX images, automated process, raw result.

000

Nicolas Barradeau @nicoptere.bsky.social · 26/08/2026

not *entirely* sure what I'm doing right now ... let's call it r&d

240

Nicolas Barradeau @nicoptere.bsky.social · 26/08/2026

Minimax H3 video & audio, FLUX images, automated process, raw result.

000

Nicolas Barradeau @nicoptere.bsky.social · 26/08/2026

Minimax H3 video & audio, FLUX images, automated process, raw result.

000

Nicolas Barradeau @nicoptere.bsky.social · 25/08/2026

Minimax H3 video & audio, FLUX images, automated process, raw result.

010

Nicolas Barradeau @nicoptere.bsky.social · 25/08/2026

Minimax H3 video & audio, FLUX images, automated process, raw result.

000

Nicolas Barradeau @nicoptere.bsky.social · 25/08/2026

a longer ride through 10 environments (Zelda BOTW, Frank LLoyd Wright, a storm, mars colony, etc..) with an "auto pilot" that steers the camera randomly. the quality is questionable & the transition are not always smooth (I'm looking for ways to improve) but the result is rather interesting.

030

Nicolas Barradeau @nicoptere.bsky.social · 25/08/2026

trying out Raven at the moment (MiniMax-H3-RAVEN-Streaming-LoRA). a good reminder that video models are basically world models. the video was produced by interactively steering the camera with prompts, in real-time.

000

Nicolas Barradeau @nicoptere.bsky.social · 20/08/2026

Flamingo: improved visuals, timing, audio etc.

020

Nicolas Barradeau @nicoptere.bsky.social · 20/08/2026

Axolotl: improved sequence, timing, audio etc. done with Minimax H3 video & audio models, FLUX Schnell and some LoRAs

110

Nicolas Barradeau @nicoptere.bsky.social · 18/08/2026

Exploring the MiniMax H3 video & music models. FLUX for image genreation and Gemini 3.6 for prompts but planning on using the Qwen 3.8 to be fully local. produced locally on a single RTX 4090 (24Go VRAM) in ~30 minutes, fully automated, one-shot, subject: "wholesome" 🤗

120

Nicolas Barradeau @nicoptere.bsky.social · 17/07/2026

#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!

150

Nicolas Barradeau @nicoptere.bsky.social · 16/07/2026

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...

010

Nicolas Barradeau @nicoptere.bsky.social · 15/07/2026

"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

0104

Reposted by Nicolas Barradeau

Nick Fisher @hydroxide.dev · 15/07/2026

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...

1282

Reposted by Nicolas Barradeau

Sakana AI @sakanaai.bsky.social · 13/07/2026

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 🧵

2516