Top Tweets for #motionbricks

离岸工具包🎒

@Stanllll6o

1 day ago

NVIDIA 的 MotionBricks 真是厉害，未来游戏开发方式可能要变。日本开发者用 WebGPU 实现了实时 3D 人物动作，太自然了。 #NVIDIA #MotionBricks

黑箱之外

@BeyondTheBoxCN

3 days ago

我越来越觉得，NVIDIA 的 MotionBricks 可能会改变未来游戏的开发方式，尤其是现在有人把它和 Gemini 组合起来了！前几天我连续发了几条 MotionBricks 的帖子，其中一位日本开发者通过 WebGPU，在浏览器里实现了实时生成 3D 人物动作。走路、转身、停下、调整身体姿势，动作衔接非常自然。现在，开发者 Kingston Kuan 又做了一件很有意思的事：让 Gemini 指挥 MotionBricks 生成动作。整个过程其实很好理解：你给 Gemini 一个命令，比如“跑到桌子旁边，绕过箱子，再坐到椅子上”。 Gemini 观察 3D 场景，理解周围的环境，规划角色应该走哪条路线、身体朝向哪里、手脚应该放在什么位置。 MotionBricks 则负责把这些目标变成连续的身体动作。开发者已经演示了行走、跳跃、越过障碍物、坐下，以及根据指令生成新的身体姿势。其中部分复杂动��仍然使用预制动画，但整个控制思路已经很有意思了。看到这里，我��然想到一个更大的可能性。如果我们把这套系统直接集成到游戏引擎里，同时提供给玩家和 NPC 使用，会发生什么？玩家可以直接告诉角色要完成什么任务，Gemini 负责规划，MotionBricks 负责生成动作。游戏里的 NPC 也可以使用同一套系统，只不过它们的任务来自自身的 AI 决策。比如我们开发一款 AI 小镇游戏。开发者创建街道、房屋、商店、汽车和居民，设定这个世界的物理规则、社会规则，以及每个 NPC 的身份和目标。接下来，居民可以自己决定起床、做饭、出门、上班、购物、回家。 Gemini 根据任务和环境规划行动，MotionBricks 负责生成身体运动。再往前想一步，如果未来还能把物品抓握、交互、物理碰撞、身体平衡等能力整合进来，那就更有意思了。以前开发者要制作大量走路、开门、搬东西、坐下、攀爬的动画，还要编写复杂的动作切换逻辑。以后也许只需要定义好角色的身体能力、物品属性和世界规则，AI 就能根据实际情况规划并生成动作。角色甚至可以完成开发者从来没有提前编排过的行为。我觉得这才是 Gemini + MotionBricks 最值得期待的方向。现在的技术还远远没有实现这种完整的自主行为系统，尤其是复杂物品交互和物理模拟，还有不少问题需要解决。但如果这条路线走通，未来制作游戏时，我们可能把更多精力放在创造世界、设计玩法和规则上，让 AI 自己解决角色如何行动。到那时候，独立开发者制作一款拥有大量自主 NPC 的开放世界游戏，门槛可能会比今天低得多。开发者：Kingston Kuan 技术组合：Gemini + NVIDIA MotionBricks 开源项目：https://t.co/xoYb28ffjk

137

140

100

TTT @tac_tic_toe

6 days ago

出来たものがこれ 5年以上前のApple Silicon M1のCPU + Neural Engineでモーションをリアルタイム生成して動いているのは驚異的 #MotionBricks #生成AI https://t.co/IRhDscgTso

185

おすぎ｜AR開発 × 生成AI

@osugi_tools

7 days ago

NVIDIAのMotionBricksのデモが、ブラウザで触れるようになってる。矢印キーで操作すると、その場でAIが動きを生成。要所の姿勢だけ用意して、間を補完してるらしい・・・自分のVRMでも試せる。Webのキャラ演出が「再生」から「生成」に変わりそう。 #MotionBricks #VRM

monoradio

@monoradio102

13 days ago

前に動画を上げた、NVIDIAのMotionBricksのデモをブラウザで触れるようにしました (PCのChrome推奨。モデルのDLが300MBほど発生します) 矢印キーで操作するとその場で動きをAIが生成します。(要所の姿勢だけ事前用意で、間を補完してます) 自分のVRMもドラッグで試せます。 https://t.co/DB720sZDh2

234

186

13K

中西崇文@ChatGPTはどのように動いているのか？2026年1月20日発売

@piano_lullaby

12 days ago

少し遅れましたがMotionBricks試しました。これはすごい。動きが本当に自然。不気味の谷を超えましたね。下記に私の実装メモを入れました。 https://t.co/5KhTKZAIMA #MotionBricks #GoogleColaboratory

372

monoradio

@monoradio102

14 days ago

モーションを生成できるNVIDIAのMotionBricksを、ブラウザのWebGPUで動かしてみた。矢印キーの操作に合わせて、その場で動きを生成してます。同じ動きのループじゃないので、妙に生々しい。 #MotionBricks #VRM

247

128K

テックトークナビゲーター@AIコンテンツクリエイター

@TechTalkNAVI

14 days ago

NVIDIAのモーション生成AI「MotionBricks」を用いて作った動きを元にアニメーション化してみたこれ本当に便利だなぁ✨ NVIDIAのグラボ積んであるノートPCで良かった #NVIDIA #MotionBricks @dreamina_ai #dreminaCPP

tingwu.wang @TingwuWang

16 days ago

Motionbricks v0.2 is here! ✅ New SOMA model + Unitree G1, one runtime ✅ Inbetweener & Scene Preview ✅ Author interactions without retraining ✅ ONNX, HF checkpoints, tutorials https://t.co/q7bXXrgGHG #MotionBricks #CharacterAnimation #Robotics #GenerativeAI #SOMA #NVIDIA

780

Simon Yuen @SimonYuenNV

2 months ago

Nice work, great to see #MotionBricks adoption on different engines and platforms. #realtimeanimation #realtime #robotics #humanoid #humanoidrobotics #digitalhuman #gamedevelopers #nvidia #aianimation #genai

FrostPunchGames

@FrostPunchDev

2 months ago

I implemented Nvidia MotionBricks in Unity as a third person controller! It generates animations from key-frames in realtime using a tiny model on your PC! Planning to release it as an asset soon. #Nvidia #Unity3D #IndieDev #GameDev

237

138

11K

421

Junfan Zhu 朱俊帆 ✈️ CoRL

@junfanzhu98

4 months ago

🚀 @siggraph 2026 Day 2: The future of motion generation isn't bigger diffusion models—it's modular latent intelligence + sparse semantic primitives. #MotionBricks replaces brittle animation graphs with factorized representations for scalable, controllable real-time motion across games, digital twins, and humanoid robotics. Presented by @TingwuWang (@NVIDIAAI GEAR). 🔥 Key Ideas 🧱 Smart Primitives > Animation Graphs Instead of hundreds of hand-authored state-machine nodes, MotionBricks represents high-level intentions—locomotion, parkour, pickup, door-open, crouch, vault, etc.—as reusable semantic primitives encoding timing, pose tokens, root trajectory and interaction context. 💡 "Complexity is a weakness in design; simplicity shows the power of the model." ⚙️ Progressive Glassbox Architecture 1️⃣ Root Module → predicts timing + global trajectory. 2️⃣ Pose Module → refines poses conditioned on the root. 3️⃣ Multi-head Decoder → reconstructs poses, quaternions, velocities and contacts. The staged Progressive Inbetweening pipeline improves reconstruction while exposing interpretable intermediate representations for debugging, editing and controllable production deployment. 🧠 Root-Disentangled Multi-Latent Tokenizer Instead of one VQ codebook, MotionBricks: ✅ disentangles root motion from local pose; ✅ explicitly reinjects root trajectory during decoding; ✅ learns multiple independent codebooks. Because upper-body articulation is structurally more complex, specialized codebooks better capture different motion attributes, yielding lower reconstruction loss, improved FID/L2 trends and a more efficient latent space. 📊 Training Scale • 700 hours • 350K motions • 315K training clips • 27-joint representation SOMA retargeting generalizes across SMPL-X, MHR, GEAR-SONIC, ProtoMotion, Kimodo and diverse body morphologies. 🤖 Applications Omniverse digital twins • GEAR-SONIC • CHORD (91% whole-body manipulation success) • CHIP humanoid control • scene-aware interactions including climbing, vaulting, doors, stairs, object manipulation and multi-agent coordination. The learned representation transfers naturally to both animation and physics-based humanoid control. 🎯 Hot Takes ✅ Sparse semantic primitives replace exploding animation graphs. ✅ Factorized latent representations scale more effectively than monolithic latent spaces. ✅ Progressive glassbox generation offers better controllability than end-to-end black-box pipelines. ✅ Better interfaces—not just larger models—may become the next scaling law for production motion generation and embodied AI. One of the clearest examples yet of graphics, robotics, world models and production AI converging into a unified motion generation architecture. #SIGGRAPH2026 #NVIDIA #Robotics #PhysicalAI #Humanoid #WorldModels #Animation #ComputerGraphics #GenerativeAI #DigitalHumans #Omniverse #EmbodiedAI @NVIDIAworkstatn, @NVIDIARobotics, @nvidia

junfanzhu98's tweet photo. 🚀 @siggraph 2026 Day 2: The future of motion generation isn't bigger diffusion models—it's modular latent intelligence + sparse semantic primitives. #MotionBricks replaces brittle animation graphs with factorized representations for scalable, controllable real-time motion across games, digital twins, and humanoid robotics.
Presented by @TingwuWang (@NVIDIAAI GEAR).

🔥 Key Ideas
🧱 Smart Primitives > Animation Graphs
Instead of hundreds of hand-authored state-machine nodes, MotionBricks represents high-level intentions—locomotion, parkour, pickup, door-open, crouch, vault, etc.—as reusable semantic primitives encoding timing, pose tokens, root trajectory and interaction context.
💡 "Complexity is a weakness in design; simplicity shows the power of the model."

⚙️ Progressive Glassbox Architecture
1️⃣ Root Module → predicts timing + global trajectory.
2️⃣ Pose Module → refines poses conditioned on the root.
3️⃣ Multi-head Decoder → reconstructs poses, quaternions, velocities and contacts.
The staged Progressive Inbetweening pipeline improves reconstruction while exposing interpretable intermediate representations for debugging, editing and controllable production deployment.

🧠 Root-Disentangled Multi-Latent Tokenizer
Instead of one VQ codebook, MotionBricks:
✅ disentangles root motion from local pose;
✅ explicitly reinjects root trajectory during decoding;
✅ learns multiple independent codebooks.
Because upper-body articulation is structurally more complex, specialized codebooks better capture different motion attributes, yielding lower reconstruction loss, improved FID/L2 trends and a more efficient latent space.

📊 Training Scale
• 700 hours • 350K motions • 315K training clips • 27-joint representation
SOMA retargeting generalizes across SMPL-X, MHR, GEAR-SONIC, ProtoMotion, Kimodo and diverse body morphologies.

🤖 Applications
Omniverse digital twins • GEAR-SONIC • CHORD (91% whole-body manipulation success) • CHIP humanoid control • scene-aware interactions including climbing, vaulting, doors, stairs, object manipulation and multi-agent coordination.
The learned representation transfers naturally to both animation and physics-based humanoid control.

🎯 Hot Takes
✅ Sparse semantic primitives replace exploding animation graphs.
✅ Factorized latent representations scale more effectively than monolithic latent spaces.
✅ Progressive glassbox generation offers better controllability than end-to-end black-box pipelines.
✅ Better interfaces—not just larger models—may become the next scaling law for production motion generation and embodied AI.
One of the clearest examples yet of graphics, robotics, world models and production AI converging into a unified motion generation architecture.

#SIGGRAPH2026 #NVIDIA #Robotics #PhysicalAI #Humanoid #WorldModels #Animation #ComputerGraphics #GenerativeAI #DigitalHumans #Omniverse #EmbodiedAI @NVIDIAworkstatn, @NVIDIARobotics, @nvidia

Junfan Zhu 朱俊帆 ✈️ CoRL

@junfanzhu98

4 months ago · Los Angeles

Tomorrow I'll be hosting official @siggraph Birds of a Feather @saturdayrobotic : World Models for Robotics: Bridging Graphics, Simulation & Physical Intelligence ☕🤖 📍 Wed @ Los Angeles Convention Center, Room 510, July 22 3:30pm 👉 https://t.co/TmF9srvp6Q #SIGGRAPH2026 #Robotics #WorldModels #LA

junfanzhu98's tweet photo. Tomorrow I'll be hosting official @siggraph Birds of a Feather @saturdayrobotic :
World Models for Robotics: Bridging Graphics, Simulation & Physical Intelligence ☕🤖

📍 Wed @ Los Angeles Convention Center, Room 510, July 22 3:30pm
👉 https://t.co/TmF9srvp6Q

#SIGGRAPH2026 #Robotics #WorldModels #LA

Amitav Bhattacharjee

@bamitav

5 months ago

NVIDIA #MotionBricks brings humanoid robots closer to natural, fluid movement! https://t.co/gp2viA23V9 #NVIDIA #humanoidtech #humanoid #robot #Robotics #AI #TechRevolution #TechInnovation #ArtificialInteligence #PhysicalAI @PawlowskiMario @chidambara09 @Ym78200 @CurieuxExplorer @efipm @bigfundu @sayedflah @Ronald_vanLoon @cyngn @belindabeibi @odisseiaalfa @DigitalColmer @MyCompanionsAI @KirkDBorne @patricegorissen @jeffkagan @EdwardKens50830 @enilev @insom_ai333 @ChrisCCrowley @sallyeaves @gurmeet_judge @VairagyaSadhana @andresvilarino @amomsimpression @amit6060 @0xAmol @OfficialDabier

672

Top Tweets for #motionbricks

Last Seen Hashtags on Sotwe

Trends for you

Most Popular Users