afif (@afifwm) header
afif (@afifwm) profile picture
Joined March 2015
Photo by afif (@afifwm)Photo by afif (@afifwm)Photo by afif (@afifwm)Photo by afif (@afifwm)Photo by afif (@afifwm)Photo by afif (@afifwm)Photo by afif (@afifwm)Photo by afif (@afifwm)Photo by afif (@afifwm)Photo by afif (@afifwm)
スマホの傾きに合わせて、絵が回るようにしてみた。 3D化しているわけではなく、1枚の絵から360°回転の「動画」を作り、傾きに合わせてフレームを前後させて、立体っぽく見せています。 #MiniMaxH3 + ORB360 LoRA
91
1,027
148
6,850
369,042
afif retweeted
Replying to @yzyupdates
Bro This shit was a vibe The crowd was very loud It’s called rock and roll
1,022
9,734
1,025
108,299
3,065,241
3DGS is powerful, but it has one obvious limitation: it’s static. So I pushed the same workflow into 4DGS. For this test, I took an animated lion walk cycle with 251 frames and over 4 million hairs, then rendered it from 20 camera views in @Blender + Cycles using @rendernetwork That produced a 5,020 frame dataset, ran across 4 parallel jobs with dozens of concurrent frames rendering and finished in about 8 hours. On my local 5090 workstation, that same workload would have taken roughly 288 hours!!! From there, I trained 251 splats (one PLY per frame) with Postshot CLI, at a little over 4 minutes per frame. The result is a near real-time 4DGS lion asset that I can duplicate 100x, offset for staggered motion, and even relight in @OTOY OctaneRender. 🔥 What makes this even more exciting is that everything just worked - and now that the sequence has been rendered on Render Network, the full 251-PLY sequence is hashed and persistent through Render Networks Differential Upload feature. That means anytime I want to use this asset again in a new 3D scene, the data is already live on Render Network. Uploading the 4GB sequence is completely bypassed during scene upload and job submission. That’s a huge win - not just for converting a heavy animated asset into something modular and real-time, but for making it reusable, fast to deploy, and practical across future render workflows in both Cycles and Octane for ANYONE who uses this same splat sequence within their scene. 3D to splat conversion keeps getting cooler.
7
16
4
113
9,139
🧍‍♂️3D Gaussian Splattingに物理演算を適用 🔹TeleportのPaul V.氏が自身の3DGSをリギング 🔹コリジョンメッシュと物理演算を追加 🔹「GMOD」のような挙動を実現 情報元はリプ欄。
5
30
1
276
16,052
🎮 GPT-6 Astra + XGRIDS 3DGS + Tripo: real park → playable game in under 2 hours. XGRIDS delivers photoreal 3DGS + a matching-scale mesh for collision and lighting in Unreal. Want the workflow? Webinar coming soon. Join discord to try the demo build👇 discord.gg/b99H8wjRaH
1
14
87
5,683
Volinga Plugin Pro now lets you relight 4D Gaussian Splats inside #UnrealEngine The #UE5 plugin is pitched as a 'major leap' in integrating 3DGS workflows into Unreal pipelines for #VFX and #virtualproduction Check out its key features: cgchannel.com/2026/08/voling… @volinga.ai
14
56
4,278
afif retweeted
Google DeepMind just unveiled GenCeption. This AI turns video into depth, segmentation, 3D keypoints, and searchable 4D worlds. 8 wild examples:
25
70
8
725
72,412
High-fidelity thin-shell 3D in about a second on a consumer GPU — it even runs on a laptop CPU. No server needed. Introducing DiffGI: a fully differentiable geometry-image framework for end-to-end 3D generation. Accepted to #ECCV2026 🇸🇪 🧵👇
8
42
309
20,736
🙌🏼@UnrealEngine 5.8 just leveled up #Character #Animation! #ControlRig #Dynamics brings fast, intuitive physics directly into your rigs! NO heavy setup, NO complexity. You can now: Spawn dynamic solvers in sec, build & control chains! Enjoy! 🤩 youtu.be/tCSY0j60TE0
11
65
4
788
59,030
They modeled everything flat and then turned it into a sphere, this is insane
New on the Blog: “Messenger” — a Case Study by @abeto_co    awwwards.com/messenger.html Dive deep into this interactive experience and discover how abeto captures a unique balance of total tranquility with a subtle sense of “oddness” — all within a game.
28
390
16
4,762
244,099
Clarity Pro Upscaler now supports up to 10K That's 100 Megapixels and a single PNG is already ~130 MB 😅
25
31
1
525
48,950
afif retweeted
Claude now connects to the tools creative professionals already use. With the new Blender connector, you can debug a scene, build new tools, or batch-apply changes across every object, directly from Claude.
1,519
4,145
2,525
45,435
12,937,563
Early R&D using our SP-6M dataset. Exploring image-to-3D reconstruction from single images, including heavily modified inputs (lighting, hair, etc). Still a work in progress.
27
67
8
595
53,623
Excited to present DRoPS! A novel method for dynamic 3D reconstruction from monocular videos, accompanied by a static-prescan of the dynamic object. 🌐 Project: drops-dynamics.github.io 📄 Paper: arxiv.org/abs/2603.24770 🧵 [1/7]
7
32
4
220
35,103
afif retweeted
Every cool animation you see is ‘cheating’ behind the scenes. Credit: 3D artist @moranimates
4
52
4
845
69,443
For an upcoming project at @active_theory I've been experimenting with a file format i'm calling ActiveFrame. Like a video format, leveraging H.264 compression, allows random-access to any frame and it's fully hardware accelerated. No runtime depedencies, works eveywhere.
40
100
6
1,387
93,217
Been experimenting with procedural locomotion on #GaussianSplat environments from @theworldlabs . Built a little multi-legged robot in Unity that raycasts against the splat data to figure out where to place its feet, no meshes or colliders involved... Pretty fun to climb surfaces, walk on walls, and set a weird number of legs on the fly :) Still rough around the edges but pretty fun to watch it figure things out. #GaussianSplatting #Unity3D #WorldLabs #ProceduralAnimation
61
153
19
1,295
146,669
afif retweeted
BULLY ON THE WAY NO AI
10,101
43,292
6,952
277,215
35,317,012
afif retweeted
Tiny streaming: unlocked For streaming 4DGS content, we target 75 Mbps (120K splats per frame). A sweet spot where quality stays high and bandwidth requirements remain within standard internet speeds. But we wanted to see how far we could push it. The result: 17 Mbps (15K splats per frame). For reference, Vimeo streams 1080p video at around 10–20 Mbps. With this compression level, the finer details take a hit, but it opens the door for 4DGS streaming to larger geographies with slower internet connections.
6
31
2
272
15,805
afif retweeted
Studio @Ten_24, the creators of 3D Scan Store, unveiled SP-6M, "the largest database of high-resolution individual human head scans ever created." It can potentially make 177,978,515,625,000,000,000,000 "unique, identity-safe synthetic human heads": 80.lv/articles/3d-scan-store…
9
93
7
1,057
66,700