Benjamin Arnaud

I made MotionMonkey 🐵, a streamlined Youtube and Torrent player based on OpenSource technologies. Available on the PlayStore. 👉APK: play.google.com/store/apps/det… #android #streaming #youtube #bittorrent

@KIRI_Engine_App Neat, where's the link though

3D Gaussian Splatting inside Godot🎮 Instead of forcing splats into a mesh pipeline, this project adds a separate rendering pass🤩 and blends it after mesh rendering. And yeah, it’s fully open source🥳

@mausmoto Fantastic, godspeed to you

@3unjee Everything's open source!

ShaderGlass 1.3 released! Better Linux support, shader search function and of course, more shaders! 📺 Free on Steam, itch and GitHub!

@newlinedotco @victormustar It's incredible that we can run this locally when you think about it

the multi-reference part is the real differentiator here. usually, image editing models struggle to maintain identity or style when you're pulling from more than one source, but FLUX.2-klein-9b unified generation and editing in a way that feels more like multi modal fusion than just a standard img2img pass. the speed jump from KV-caching is essentially bringing the short-term memory optimization from LLMs into the DiT world. by caching the keys and values of the background or reference image tokens, the model skips the redundant math of recalculating the static parts of your composition. you’re only paying the compute cost for the actual edit region. we’ve seen a similar pattern in our agent modules when you standardize on 4-step distilled models for the iterative loops, and then switch to the 9B base for the final approval shot, you optimize the entire dev to creative pipeline. sub-second generation means the bottleneck isn't the GPU anymoreit's how fast you can refine your prompt or eval suite.

New drop: FLUX.2-klein-9b-kv 🔥 Exceptional at multi-reference image editing → Sub-second generation in just 4 inference steps → 2.5x faster with KV-caching ⬇️ Try the free demo on Hugging Face

@Pinsky Does it work with people ?

This is wild! Rotate Object just landed in the Photoshop beta 🤯 I would have loved this back in my compositing days!

@videolan @faevent Legends :cone:

VideoLAN team at Fossasia 2026 @faevent, in Bangkok, on a booth showing AV2 playback in software decoding. Lots of goodies, if you come and say hi!

@TheoMediaAI I don't see the upside of generating clips from within the editor ?

Whoa-- I was JUST talking about this! LTX just released LTX Desktop. A LOCAL Non-Linear Editor that you can generate AND cut in. Digging into it, but this is PRETTY big. I've been waiting for a AI Native NLE!

@dangreenheck Do you have an online demo running for that one ? Just to check the performances

After 5 failed attempts and a week of long nights (thanks for nothing Claude), I finally have seamless air/water transitions in Three.js Water Pro. I was going to save this for V3 but the existing transition was so bad, I wanted to get it fixed and done right. If you want to get notified when Three.js Water Pro V2 gets released—plus weekly Three.js tips and tutorials—be sure to sign up for my newsletter. You'll also receive a code for 10% OFF you first order when you sign up! threejsroadmap.com/join

@tunguz We have to reinvest that temporality

And Feynman just 37. Those were much saner times.

@mrdoob @threejs @Blender @openusd I suppose for a project like threejs AI is as relevant as it can be given it's proficiency at interpolating between one well established tech to another in particular when the output is JavaScript / Web standards, given how permissive they are.

Some people want to know how much AI assisted code has landed in @threejs so far. I thought I could ask the AI to analyze all the commits and PRs since October 2024. We're at 1.92%. Thanks to the AI we now have: ✅ 1:1 render parity with @Blender's EEVEE ✅ A robust @openusd loader ✅ A much improved @collada loader ✅ A much improved threejs.org/editor/ ✅ @googlechrome devtools (not counted) ✅ Reduced project dev dependencies by 75% ✅ JSDocs based new documentation ✅ Made JSDocs 500x faster for all JSDocs users

@benjaminbardou This looks fantastic. Quel workflow Benjamin ?

Revolution [a complete circle or turn round a centralpoint, axis] [to roll back, to bring back]

@tamrrat Worldlabs marble API under the hood ?

turn any google street view url into a 3D gaussian splat!

@aifilmmaker Neat. How do you ensure AI knows the way it looks from other angles ?

AI doesn't take away your creativity. It gives you time. I designed this ship in Blender 2 years ago, but this short is a passion project and other stuff took precedence (we even made a game with Meta) but now, even though I'm still busy, I get to finally make this film!

@CoffeeVectors @sunomusic Neat. Which platform are you using for Seedance 2 ?

Last Breath, That’s My Shhh… Testing mechanical parts, laser swords, monsters and martial arts with Seedance 2. Edited together from several clips. Music by me in @sunomusic

@sparkjsdev Does it solve the desktop / mobile performance gap and the necessity of providing two PLY(s) for each platform?

In Spark 0.1, world scale was constrained by real device limits: real-time splat counts, GPU memory ceilings, multi-GB file loading failures in the browser, and precision artifacts at large coordinates. Spark 2.0 reworks the architecture around LoD trees and streaming. Any splat file can be converted into a hierarchical LoD representation, with fixed, tunable splat budgets per frame to guarantee steady performance across devices. Scenes are streamed coarse-to-fine using a new .RAD format and managed by a shared LRU "splat pager" that keeps GPU memory bounded. We also added higher-precision ExtSplats encoding and ReadableStream support for multi-GB files. Branch: github.com/sparkjsdev/spa…

Spark 2.0 Developer Preview is now available: introducing Level-of-Detail (LoD) rendering and streaming for world-scale Gaussian Splats on the web. Learn more ↓

@fernandojsg Neat. Github repo?

Almost there. A few tiny details left, but the Doom port to JavaScript + Three.js is basically done. ~27k lines of code, ~200KB gzipped

@exQUIZitely I agree that the alliance between innovation and creativity was the key part in what made these games magical in the 90s. Especially given the short span between each grand evolution. Nevertheless, I feel developers took more risks, the business model wasn't rationalized yet

@3unjee I think the environment changed - gaming has become mainstream, industrialised, grown up if you will. No true innovation, just improvements of what has been born 30-40 years ago. Doesn’t mean it can’t be good, but it’s notmagical anymore.

I sometimes wonder why older games (80s and 90s era) hold much higher emotional value for me, while modern games, almost everything after the early 2000s, do not. I think it’s not just one reason but several. The first is obvious, the second became clearer to me a while ago, and the third and fourth reflect the reality we live in today. Not sure if you agree - and it probably depends on when your gaming journey started - but maybe you see some parallels? 1) With the past, we tend to remember mostly the good parts and what we liked. That’s just how human memory works. It’s true for movies, music, and other things from our youth too. So games from my formative years carry higher emotional weight by default. 2) Games in the 80s and 90s didn’t have to compete with today’s mass media and constant sensory overload. They felt more “magical” because they let you dive into a fictional world, much like a great book or movie. They didn’t look realistic by today’s standards, but they stimulated your imagination, letting you fill in the blanks and build the experience in your mind. We had three TV channels, one rotary phone, no social media, no internet, no mobile phones, no streaming. The internet grew in the 90s, mobile phones appeared, and more TV channels arrived - but still no social media, no streaming, no media overload like today. A 90s game could still feel special because it wasn’t drowned out by endless digital distractions. We read about games in print magazines, waited to buy the physical copy in a shop, and the anticipation and joy were tied to that whole process. 3) Today the gaming industry is a multi-billion-dollar business, often driven by microtransactions, addictive mechanics, in-game ads (especially mobile), and run by publicly listed companies where CFOs call the shots. We have endless streaming options, constant connectivity, and social media (plus AI noise/junk) everywhere. Our brains can only handle so much input, yet more games are available than ever (Steam, Roblox, and countless others). No wonder we don’t form the same emotional bonds with them anymore. At least that’s how I see it. 4) Multi-Player games in the 80s meant that you would sit around one computer with your friends and play together. In the 90s it meant that you would setup a LAN session, again with your friends in the same room. Multi-Player today means that you connect online, either with complete strangers or - maybe - people you get to know a little bit over time. But you definitely don't sit in the same room anymore. So, the social aspect, the connection beyond the game, has been reduced to a shell of its former self. Ask me about any pre-2000 game and I can tell you a lot about it, what it meant to me, why it felt special (or why it sucked). Ask about almost any game after the early 2000s and you’ll mostly get a blank stare. I know “everything was better in the past” can sound a little bit cringe, so maybe for younger gamers Grand Theft Auto V or the latest World of Warcraft expansion is what Defender of the Crown and Civilization were to me?

@grok @freepik No give me the API costs please

Magnific Video Upscaler (via Freepik): Plans start at $5.75/mo (annual) with 84k credits/year (Essential) up to $158/mo with 3.6M credits/year (Pro). Credits for upscaling don't reset monthly. Topaz Astra: Personal $39/mo (or $27/mo annual), Pro $299/mo (or $179/mo annual). Includes 150-400 video credits/mo depending on tier. Costs depend on usage; Magnific may be cheaper for light users. Check sites for latest.

The new SOTA video upscaler is here Introducing Magnific Video Upscaler: · Up to 4K · 3 presets + custom mode for total control · Turbo mode when speed matters · FPS Boost for smoother motion · 1-frame preview to never waste a credit Exclusively on Freepik and Magnific

@SinzVII @threejs That's why performance was surprisingly good on mobile

@3unjee @threejs Hi mate, I kept the density the same, however, I stripped down the unneeded data using the script i uploaded to the repo (to lower the file size to a reasonable size)

Testing out Apple’s SHARP to generate a point cloud from a 2D Image then using @threejs to create a GPGPU particle flow field in WebGL. Live demo: cullenwebber.github.io/three-ml-sharp/ Source code: github.com/cullenwebber/t… #threejs #webgl #particles #3DModel