NVIDIA가 30년 역사의 애니메이션 기술을 구식으로 만들다.

NVIDIA JUST MADE 30 YEARS OF ANIMATION TECHNOLOGY OBSOLETE.

One neural network.

350,000 motion clips.

15,000 frames per second.

It is ca…

⚠️ title 필드는 반드시 한국어로 번역하세요. 고유명사(OpenAI, Unity 등)만 영문 유지.
⚠️ [IMG:N] 형식 토큰은 이미지 placeholder 입니다. 번역하지 말고 원래 위치에 그대로 유지하세요.

It is called MotionBricks and it just changed everything about how movement gets created in real time.

Here is what that actually means.

Before MotionBricks a game studio needed a motion capture studio, a team of animators, thousands of hand-keyed transitions, and months of polish to make a character pick up a sword and make it look believable.

Every step. Every weight shift. Every follow-through. Hand crafted. Frame by frame. By a human who spent years learning how bodies actually move.

MotionBricks does all of that in real time.

You tell it: pick up a sword in zombie style.

It generates the foot placement.

The balance adjustment.

The reach.

The grip.

The follow-through.

Every transition between every movement state.

Automatically.

At 15,000 frames per second.

Not a pre-recorded animation played back.

A neural network making real-time decisions about how a body moves through space based on 350,000 examples of how bodies actually move through space.

THE PART THAT CHANGES EVERYTHING ELSE

This is not just a game engine feature.

NVIDIA built MotionBricks as the core motion engine inside their GROOT stack.

The same system generating zombie sword pickups in real time is now powering humanoid robot movement in research labs around the world.

The same neural network.

The same real-time decision making.

Applied to physical robots that need to navigate the real world without falling over.

This is the moment where animation technology and robotics stopped being separate fields.

One neural network trained on human movement is now the bridge between digital characters and physical machines.

WHAT THIS MEANS FOR BUILDERS

Every game being built right now that uses traditional animation pipelines is building on infrastructure that just became the expensive way to do things.

수동으로 작성된 움직임 로직 (hand-coded movement logic)을 작성하는 모든 로보틱스 팀은 이제 움직임이 실제로 어떻게 작동하는지에 대한 350,000개의 사례를 학습한 신경망 (neural network)과 경쟁하고 있습니다.

이 기술을 가장 빠르게 채택하는 스튜디오와 연구소는 이전에 몇 달이 걸렸던 일을 단 며칠 만에 해낼 것입니다.

기다리는 이들은 앞으로 2주 안에 움직인 팀들을 따라잡기 위해 향후 2년을 허비하게 될 것입니다.

초당 15,000 프레임 (15,000 frames per second).

단 하나의 신경망 (neural network).

30년의 애니메이션 기술.

화요일 하루 만에 대체되었습니다.

다른 피드들이 상황을 파악하기도 전에 빌더들이 할 수 있는 일을 변화시키는 모든 AI 혁신 소식을 알고 싶다면 @cyrilXBT를 팔로우하세요.