Garment Particles: A 2D–3D Symmetric Garment Representation for Generation and Editing

Kiyohiro Nakayama1       I-Chao Shen2       Ruofan Liu3       Yiming Wang4       Gordon Wetzstein1       Takeo Igarashi2

1Stanford University 2The University of Tokyo 3Institute of Science Tokyo 4ETH Zurich

Paper arXiv Code Dataset (HF) Video

SIGGRAPH Conference Papers 2026 · Los Angeles, CA

Supplementary Video

Garment Particles is a garment representation that models both the sewing pattern and its draped garment geometry in a symmetric, 2D–3D point cloud. (a) shows the garment particles representation. The color on the 3D garment (left) and the 2D sewing pattern (right) indicate the same points. Garment Particles Flow (GPF), a generative framework, generates garment particles from multimodal inputs. More importantly, the prior space of GPF enables versatile editing in both 3D garment geometry and 2D sewing pattern domains. Finally, Particles-to-Pattern Flow (PPF) converts the generated particles to simulation-ready sewing patterns. (b) shows the various editing applications enabled by GPF.

Abstract

Practical garment design spans two modes: intuitive creation from high-level intent, such as a reference image or text description, and complex low-level editing across 2D sewing patterns and 3D draped geometry, which requires professional training to navigate their complex interdependencies. Yet existing frameworks address only part of this challenge, offering either garment generation from casual inputs or direct editing on sewing patterns. To support both ends of the spectrum, we propose Garment Particles, a 5D point-cloud representation that jointly encodes 2D sewing patterns and 3D geometry. This representation enables Garment Particles Flow (GPF), a rectified flow framework that supports intuitive generation from high-level inputs (text, images, sketches) and various editing operations on 2D sewing patterns and 3D geometries via diffusion posterior sampling. Finally, we introduce Particles-to-Pattern Flow that converts generated garment particles into curved-based patterns for simulation. We validate our model's generation ability on multiple datasets, achieving state-of-the-art garment generation results against competitive baselines. Our model also enables many garment editing scenarios, including garment interpolation, sewing pattern editing, point-cloud- and silhouette-conditioned garment generation.


Method

Our framework has three parts: a unified Garment Particles representation that jointly samples 2D sewing patterns and 3D draped geometry as a 5D point cloud; Garment Particles Flow (GPF), a rectified flow generative model over this representation that supports text-, image-, and sketch-conditioned generation as well as diffusion posterior sampling (DPS) for editing; and Particles-to-Pattern Flow (PPF), which recovers simulation-ready curved sewing patterns from the generated particles.

Pipeline Overview
End-to-end pipeline. Multimodal inputs (text, images, sketches, or partial 3D/2D edits) condition GPF, which generates a set of garment particles jointly encoding the sewing pattern and the draped 3D geometry. PPF then converts the generated particles into curve-based sewing patterns ready for physical simulation and fabrication.
Garment Particles Representation
(Left) We model garments as the graph Γ of the parametric function mapping a sewing pattern U in ℝ2 to its draped geometry r(U) in ℝ3. (Right) We discretize Γ with point samples XΓ. Points with the same color in 2D and 3D represent corresponding points; black points mark the boundary of U.
Diffusion Trajectory
We visualize the joint generation process of Garment Particles Flow for two samples in parallel. Each column shows the diffusion of the 2D sewing pattern (top) and 3D draped geometry (bottom) at 3× speed. After the trajectory finishes, the 3D view transitions to a 360° render of the diffused point cloud and then of the reconstructed garment, while the 2D sewing pattern remains on its final state.

Garment Particles Flow & DPS-based Editing
(Left) Using a trained GPF model, we optimize the posterior mean at each sampling step against an observation, guiding generation toward a sample that minimizes a user-specified objective. (Right) Adjusting the hyperparameter stop_t produces more faithful (higher stop_t) or more diverse (lower stop_t) results from different random noise inputs.

Results

We evaluate Garment Particles on garment generation (text-, image-, and sketch-conditioned) and on a wide range of editing operations enabled by DPS, including point-cloud-conditioned generation, sewing-pattern editing, silhouette-conditioned generation, and interpolation. We benchmark against competitive baselines on GarmentCodeDatav2, 4DDress, and in-the-wild images.

Text-conditioned Garment Generation
Qualitative comparison on text-conditioned garment generation against baselines. Our method produces sewing patterns whose draped geometry faithfully matches the input description.
Image-conditioned Garment Generation
Compared to baselines, which exhibit incorrect pattern style and stitching, our method correctly generates a sewing pattern that yields a draped garment matching the input image, for both sketch and GCDv2 image inputs.
In-the-wild Image-conditioned Generation
Our method produces more plausible sewing patterns than ChatGarment on in-the-wild images, capturing the displayed garment style.
Point-Cloud-Conditioned Garment Editing
Point-cloud-based garment editing applications enabled by GPF. (a) Users directly edit an existing 3D garment to guide its generation; addition and deletion of points are performed via our 3D interface. (b) Garment mixing: components of two existing 3D garments are combined to generate a new garment. (c) Text-conditioned generation from an incomplete 3D garment. Numbers indicate the number of garment particles used.
Sewing Pattern Editing
Given a generated garment shown in grey, we edit the sewing pattern (red parts illustrate the user's additions through our 2D UI) and use the edited pattern to guide the garment generation process.
Silhouette-Conditioned Garment Generation
The user paints a 2D projection to guide garment generation through our 2D interface. The user can control the complexity of the result by changing the number of particles. Numbers indicate the number of garment particles used.
Multi-step Garment Editing
A garment editing sequence combining several of the editing methods enabled by GPF.
Fabrication
Generated garments can be fabricated from the sewing patterns recovered by Particles-to-Pattern Flow.

Citation
@inproceedings{nakayama2026garment,
  title     = {Garment Particles: A 2D--3D Symmetric Garment Representation for Generation and Editing},
  author    = {Nakayama, Kiyohiro and Shen, I-Chao and Liu, Ruofan and Wang, Yiming and Wetzstein, Gordon and Igarashi, Takeo},
  booktitle = {SIGGRAPH Conference Papers '26},
  year      = {2026},
  doi       = {10.1145/3799902.3811102}
}


Acknowledgements

This research is supported by JST ASPIRE (JPMJAP2401), the Initiative on Recommendation Program for Young Researchers and Woman Researchers, the Information Technology Center of The University of Tokyo, LVMH, Google, and the National Science Foundation Graduate Research Fellowship Program.