TANGLED: Generating 3D Hair Strands from Images with Arbitrary Styles and Viewpoints

TL;DR

TANGLED tackles the challenge of generating realistic 3D hair strands from images across arbitrary styles and viewpoints by introducing a MultiHair dataset, a lineart-conditioned latent diffusion model, and a parametric braid inpainting module. The approach encodes hair strands as polylines mapped to scalp UV space and denoises a 2D latent representation conditioned on multi-view lineart features via cross-attention, enabling robust generation from sparse or diverse inputs. A dedicated braid inpainting stage imposes parametric braid constraints to preserve coherence in complex braided styles. Experiments show improved geometric and semantic fidelity over baselines, with strong user preference for the generated hairstyles, highlighting potential for culturally inclusive avatars and sketch-based editing in animation and AR contexts.

Abstract

Hairstyles are intricate and culturally significant with various geometries, textures, and structures. Existing text or image-guided generation methods fail to handle the richness and complexity of diverse styles. We present TANGLED, a novel approach for 3D hair strand generation that accommodates diverse image inputs across styles, viewpoints, and quantities of input views. TANGLED employs a three-step pipeline. First, our MultiHair Dataset provides 457 diverse hairstyles annotated with 74 attributes, emphasizing complex and culturally significant styles to improve model generalization. Second, we propose a diffusion framework conditioned on multi-view linearts that can capture topological cues (e.g., strand density and parting lines) while filtering out noise. By leveraging a latent diffusion model with cross-attention on lineart features, our method achieves flexible and robust 3D hair generation across diverse input conditions. Third, a parametric post-processing module enforces braid-specific constraints to maintain coherence in complex structures. This framework not only advances hairstyle realism and diversity but also enables culturally inclusive digital avatars and novel applications like sketch-based 3D strand editing for animation and augmented reality.

Figures (10)

• Figure 1: TANGLED brings creativity to life by generating high-quality 3D hairstyles from images of any style or viewpoint, seamlessly integrating into existing CG pipelines and delivering breathtakingly detailed hair assets.
• Figure 2: Dataset Annotation Process. Our annotation pipeline begins by processing rendered 3D hair strands with a line-art detector, and line-art sketches are combined with OpenPose cao2017realtime skeletal data for conditioning ControlNet. To enrich dataset diversity, we further synthesize multi-view images, to cover variations in lighting, texture, and perspective. Finally, GPT-4 chatgpt2025 generates detailed textual annotations for each hairstyle, including attributes such as length, curliness, density, and cultural style.
• Figure 3: Architecture of our TANGLED . Our model takes hair images with arbitrary styles and viewpoints as conditions, and generate the 3D hair latent through the diffusion process. The conditions are randomly masked and cross-attention with the latent. At inference, we sample hair latent maps and feed the upsampled hair latent map to the strand decoder to extract the 3D hair strands.
• Figure 4: Lineart extracted for various images. For the same hairstyle under different image domains (realistic, anime and oil painting), the extracted lineart effectively captures consistent hair structure and features.
• Figure 5: Application showcase.Row 1 show the generated hairstyles from hand-drawn sketches. Row 2 illustrate hairstyle modifications(adding pigtails) by altering specific parts in the sketches from Row 1. Row 3-4 depict the process of generating outputs with braid using guidelines (highlighted in red).