Research map / 3D vision / Face and shape
SimAvatar: Simulation-Ready Avatars with Layered Hair and Clothing
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Generates simulation-ready layered text-to-3D avatars (body SMPL, garment mesh, hair strands) with 3D Gaussians optimized by SDS and driven by HOOD garment and hair simulators.
Problem
Text-to-3D human avatar generation typically uses a single entangled geometry or implicit representations (NeRF/SDF) that cannot be easily simulated for realistic hair and garment motion.
Approach
Use simulation-ready layered geometry (SMPL body, watertight-free garment mesh from a text-conditioned latent diffusion VAE, hair strands) and attach 3D Gaussians to each layer, optimizing them with SDS from a diffusion model and driving them via physics/neural simulation, with a hair opacity regularizer.
Key results
- User study with 540 votes from 18 users: SimAvatar preferred over TADA 89.55% (appearance) and 91.87% (motion).
- Preferred over GAvatar 87.03% (appearance) and 94.47% (motion) in the user study.
- Preferred over Fantasia3D 100% for both appearance and motion, though Fantasia3D cannot be animated.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use SimAvatar instead of TADA | text-driven 3D avatar generation · 22 prompts, appearance preference | preference: 89.55 | Helped |
| use SimAvatar instead of Fantasia3D | text-driven 3D avatar generation · 22 prompts, appearance preference | preference: 100 | Helped |
| use SimAvatar instead of GAvatar | text-driven 3D avatar generation · 22 prompts, appearance preference | preference: 87.03 | Helped |
| use SimAvatar instead of TADA | text-driven 3D avatar generation · 22 prompts, motion preference | preference: 91.87 | Helped |
| use SimAvatar instead of Fantasia3D | text-driven 3D avatar generation · 22 prompts, motion preference | preference: 100 | Helped |
| use SimAvatar instead of GAvatar | text-driven 3D avatar generation · 22 prompts, motion preference | preference: 94.47 | Helped |
Limitations
- Hair and garment generators trained on specific datasets, limiting diversity.
- Hair and garments are simulated sequentially, which fails for cases like hoods.
- Accessories and footwear remain entangled with body or garment layers.
Experimental setup
Datasets: Garment Pattern Generator (GPG) CLOTH3D
Benchmarks: 22 text prompts user study
Metrics: user preference percentage
Baselines: Fantasia3D TADA GAvatar
Builds on: Score Distillation Sampling HOOD HAAR GAvatar EDM
Closest papers by meaning
- MPMAvatar: Learning 3D Gaussian Avatars with Accurate and Robust Physics-Based Dynamics (NeurIPS 2025 · similarity 0.91)
- DiffAvatar: Simulation-Ready Garment Optimization with Differentiable Simulation (CVPR 2024 · similarity 0.88)
- GAvatar: Animatable 3D Gaussian Avatars with Implicit Mesh Learning (CVPR 2024 · similarity 0.88)
- Text-Conditioned Generative Model of 3D Strand-based Human Hairstyles (CVPR 2024 · similarity 0.88)
- LUCAS: Layered Universal Codec Avatars (CVPR 2025 · similarity 0.88)
- DGH: Dynamic Gaussian Hair (NeurIPS 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.