Bolt: Clothing Virtual Characters at Scale

2025-04-24

AI Summary

Large anatomical variations among virtual avatars hinder efficient, automated clothing adaptation, often requiring time-consuming manual intervention. Method: This paper proposes an end-to-end, three-stage framework: (1) garment geometry transfer, (2) differentiable 2D pattern optimization preserving original seam topology, and (3) physics-based draping coupled with automatic skinning binding. The method integrates 3D mesh deformation mapping, differentiable parametric optimization, and multi-layer cloth co-simulation for entanglement-free unfolding. Contribution/Results: It achieves, for the first time, seamless, high-fidelity dynamic dressing of complex multi-piece garments onto arbitrary target avatars. The framework supports batch processing of thousands of characters and produces production-ready outputs directly applicable to real-time rendering (e.g., games, animation) and synthetic data generation, significantly improving adaptation efficiency and cross-body-type generalization.

Technology Category

• Humans and AI: Game Design — Virtual Humans, NPCs and Autonomous Characters
• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Clothing virtual characters is a time-consuming and often manual process. Outfits can be composed of multiple garments, and each garment must be fitted to the unique shape of a character. Since characters can vary widely in size and shape, fitting outfits to many characters is a combinatorially large problem. We present Bolt, a system designed to take outfits originally authored on a source body and fit them to new body shapes via a three stage transfer, drape, and rig process. First, our new garment transfer method transforms each garment's 3D mesh positions to the new character, then optimizes the garment's 2D sewing pattern while maintaining key features of the original seams and boundaries. Second, our system simulates the transferred garments to progressively drape and untangle each garment in the outfit. Finally, the garments are rigged to the new character. This entire process is automatic, making it feasible to clothe characters at scale with no human intervention. Clothed characters are then ready for immediate use in applications such as gaming, animation, synthetic generation, and more.

Problem

• Automate fitting outfits to diverse virtual character shapes
• Transfer garments from source to target bodies efficiently
• Enable large-scale clothing of characters without manual intervention

Innovation

• 3D mesh transfer and 2D pattern optimization
• Progressive garment draping and untangling simulation
• Automatic rigging for immediate character use