3D Model Generation from 2D video stream

Reimagine the future of the 3D world

The CYAN.AI platform pioneers the use of AI image algorithm technology to achieve 2D video generation of 3D motion data, enabling collaboration between the virtual world and the real world.

3D algorithmic models created based on artificial intelligence, computer vision, and biomechanics

The CYAN.AI platform utilizes CNN (Convolutional Neural Network) and DNN (Deep Neural Network) as its core to extract key features of human body movements from 2D videos for body pose recognition. It tracks over 208 key points and combines dynamic and biomechanical modeling algorithms to achieve precise capture of facial expressions and body movements. The platform also incorporates various algorithms such as motion capture feature smoothing, multi-object collaboration in image recognition, spatial detection and positioning, redirection, and skeleton mapping algorithms

Facial capture with 140 points

Body capture with 30 points

Finger capture with 21 points

208

Key Point Capture

900million

parameters

0.1s

delay

The isolated islands of digital models are being connected by bridges

CYAN.AI has developed a brand-new skeletal system that utilizes synthetic data algorithms to match and map the skeletal systems of popular models such as Metahuman、Unreal Skeleton Asset、Daz、CC4、iClone、Unity avatar、VRM、MMD、Mixamo、MayaAdv、MayaHumanIK、3DmaxBiped. This allows the full-body driving of virtual avatars to no longer be fragmented into isolated digital islands

Deployment with almost no difficulty

The CYAN.AI architecture automatically calibrates the environment, recognizes and segments the characters/background/obstacles. It can achieve stable and high-precision motion capture even in complex, narrow, or heavily cluttered environments. With just one click, it can export FBX data to a local folder

With a consumer-grade graphics card like the GeForce RTX 3060, smooth and stable performance of above 30 FPS can be achieved, enabling lower latency data information throughput and data image synthesis. This results in ultra-low latency visual motion capture, with delays as low as 0.1 seconds

Partial list

News and Events

CYANPUPPETS, a provider of 3D virtual social networking tools, secures $10 million angel round of funding

NVIDIA Empowers Cyanpuppets to Create 3D Skeletal AI Transformation Models for Real-Time 2D Video Generation

Source: China Daily

Source: NVIDIA China

Unity|CYAN.AI Self-developed AI model for real-time 3D action generation from 2D video

Source: Unity China

NVIDIA Empowers Cyanpuppets to Create 3D Skeletal AI Transformation Models for Real-Time 2D Video Generation

Source: NVIDIA China

Unity|CYAN.AI Self-developed AI model for real-time 3D action generation from 2D video

Source: Unity China