This paper presents a style-based inverse kinematics (IK) system that uses a learned probability model of human poses to generate realistic poses satisfying given constraints in real-time. The system learns a probability distribution over all possible poses, allowing it to generate any pose but preferring those similar to the training data. The model is represented using a Scaled Gaussian Process Latent Variable Model (SGPLVM), which automatically learns parameters without manual tuning. The system also supports style interpolation, enabling smooth transitions between different pose styles. The style-based IK system can replace conventional IK in computer animation and computer vision. It is demonstrated in various applications, including interactive character posing, trajectory keyframing, real-time motion capture with missing markers, and pose estimation from 2D images. The system's main advantage is its ability to generate natural poses by learning from training data, even when the data is limited or not highly redundant. However, it requires suitable training data to be available and does not explicitly model dynamics or original motion capture constraints. The paper discusses related work, including traditional IK methods, biomechanical approaches, and example-based IK systems. It also compares the proposed SGPLVM model with other PDF models like mixtures-of-Gaussians (MoG), highlighting the SGPLVM's ability to avoid overfitting and produce smooth, multimodal likelihood functions. The system's learning process involves optimizing an objective function to find latent space coordinates that represent poses, with the ability to interpolate between styles. The paper describes the system's applications, including interactive character posing, where users can define poses by moving constraints, trajectory keyframing, where animations are created by keyframing trajectories, real-time motion capture with missing markers, and pose estimation from 2D images. The system uses a gradient-based optimization method for real-time synthesis and includes an annealing-like procedure to avoid local minima. Style interpolation is achieved by blending between two SGPLVMs, creating new styles that interpolate between existing ones. The paper concludes with a discussion of the system's potential applications in games and animation, as well as future work to improve the system's performance, including better handling of dynamics and constraints. The system is supported by various funding sources and has been licensed by a leading game developer for rapid content development.This paper presents a style-based inverse kinematics (IK) system that uses a learned probability model of human poses to generate realistic poses satisfying given constraints in real-time. The system learns a probability distribution over all possible poses, allowing it to generate any pose but preferring those similar to the training data. The model is represented using a Scaled Gaussian Process Latent Variable Model (SGPLVM), which automatically learns parameters without manual tuning. The system also supports style interpolation, enabling smooth transitions between different pose styles. The style-based IK system can replace conventional IK in computer animation and computer vision. It is demonstrated in various applications, including interactive character posing, trajectory keyframing, real-time motion capture with missing markers, and pose estimation from 2D images. The system's main advantage is its ability to generate natural poses by learning from training data, even when the data is limited or not highly redundant. However, it requires suitable training data to be available and does not explicitly model dynamics or original motion capture constraints. The paper discusses related work, including traditional IK methods, biomechanical approaches, and example-based IK systems. It also compares the proposed SGPLVM model with other PDF models like mixtures-of-Gaussians (MoG), highlighting the SGPLVM's ability to avoid overfitting and produce smooth, multimodal likelihood functions. The system's learning process involves optimizing an objective function to find latent space coordinates that represent poses, with the ability to interpolate between styles. The paper describes the system's applications, including interactive character posing, where users can define poses by moving constraints, trajectory keyframing, where animations are created by keyframing trajectories, real-time motion capture with missing markers, and pose estimation from 2D images. The system uses a gradient-based optimization method for real-time synthesis and includes an annealing-like procedure to avoid local minima. Style interpolation is achieved by blending between two SGPLVMs, creating new styles that interpolate between existing ones. The paper concludes with a discussion of the system's potential applications in games and animation, as well as future work to improve the system's performance, including better handling of dynamics and constraints. The system is supported by various funding sources and has been licensed by a leading game developer for rapid content development.