ACM SIGGRAPH 2026

HairPort

In-context 3D-aware Hair Import and Transfer for Images

HairPort transfers a reference hairstyle onto a source face while explicitly handling large pose and scale differences through 3D-aware alignment before image synthesis.

Code Dataset Weights BibTeX arXiv Paper DOI Video coming soon
HairPort hairstyle transfer examples across different identities, poses, scales, and visual styles.
HairPort transfers hairstyles across challenging differences in identity, pose, scale, and style.

Abstract

3D-aware hairstyle transfer from a single reference image.

Transferring hairstyles between images is an important but challenging task in computer graphics, computer vision, and visual effects. It enables users to explore new looks without physically altering their hair, with applications in virtual try-on systems, augmented reality, and entertainment.

Most prior works operate best under small pose gaps and fall short under large viewpoint and scale differences, where missing hair content must be synthesized rather than transferred. HairPort addresses these issues by explicitly separating hair removal from transfer and enforcing geometric consistency before synthesis.

HairPort introduces a Bald Converter that produces realistic bald versions of faces through LoRA-based in-context adaptation of FLUX, trained with Baldy, a dataset of 6,400 paired bald and original images. A 3D-aware transfer pipeline reconstructs and re-renders the reference hairstyle from the target viewpoint before compositing it onto the source image. A conditional flow-matching generator then synthesizes the final result conditioned on the bald source, pose-aligned hair rendering, original reference image, and text prompt.

Method

A staged pipeline for pose-consistent hair import.

HairPort first creates a clean bald source canvas, aligns the reference hairstyle in 3D, then synthesizes a final identity-preserving transfer.

HairPort method diagram showing inputs, bald conversion, 3D-aware hair transfer, and hair synthesis.
Method overview: source bald conversion, 3D-aware reference-hair alignment, and final flow-matching synthesis.
  1. 01

    Bald Converter

    Removes source hair while preserving facial identity, expression, lighting, and background context.

  2. 02

    Baldy Dataset

    Provides 6,400 paired bald and original portraits for supervised in-context bald generation.

  3. 03

    3D-aware Transfer

    Reconstructs and re-renders reference hair from the source viewpoint to handle pose and scale gaps.

  4. 04

    Flow-Matching Synthesis

    Fuses the bald source, aligned hair rendering, reference image, and prompt into a coherent edit.

Transfer Results

Reference hair in, source identity out.

Each carousel panel shows a face image, a reference hairstyle, and the final HairPort result. The examples stress changes in color, style, viewpoint, scale, and visual domain.

Qualitative examples

The gallery includes the curated transfer examples from the project assets.

1 / 21
HairPort transfer result showing source, reference hairstyle, and final synthesis.
FaceReference HairResult
HairPort transfer result showing source, reference hairstyle, and final synthesis.
FaceReference HairResult
HairPort transfer result showing source, reference hairstyle, and final synthesis.
FaceReference HairResult
HairPort transfer result showing source, reference hairstyle, and final synthesis.
FaceReference HairResult
HairPort transfer result showing source, reference hairstyle, and final synthesis.
FaceReference HairResult
HairPort transfer result showing source, reference hairstyle, and final synthesis.
FaceReference HairResult
HairPort transfer result showing source, reference hairstyle, and final synthesis.
FaceReference HairResult
HairPort transfer result showing source, reference hairstyle, and final synthesis.
FaceReference HairResult
HairPort transfer result showing source, reference hairstyle, and final synthesis.
FaceReference HairResult
HairPort transfer result showing source, reference hairstyle, and final synthesis.
FaceReference HairResult
HairPort transfer result showing source, reference hairstyle, and final synthesis.
FaceReference HairResult
HairPort transfer result showing source, reference hairstyle, and final synthesis.
FaceReference HairResult
HairPort transfer result showing source, reference hairstyle, and final synthesis.
FaceReference HairResult
HairPort transfer result showing source, reference hairstyle, and final synthesis.
FaceReference HairResult
HairPort transfer result showing source, reference hairstyle, and final synthesis.
FaceReference HairResult
HairPort transfer result showing source, reference hairstyle, and final synthesis.
FaceReference HairResult
HairPort transfer result showing source, reference hairstyle, and final synthesis.
FaceReference HairResult
HairPort transfer result showing source, reference hairstyle, and final synthesis.
FaceReference HairResult
HairPort transfer result showing source, reference hairstyle, and final synthesis.
FaceReference HairResult
HairPort transfer result showing source, reference hairstyle, and final synthesis.
FaceReference HairResult
HairPort transfer result showing source, reference hairstyle, and final synthesis.
FaceReference HairResult

Bald Converter Results

A clean source canvas before transfer.

The Bald Converter removes source hair while preserving face structure, expression, clothing, lighting, and background context. The paired images are displayed as interactive before/after comparisons.

Original portrait before bald conversion.
Generated bald portrait after Bald Converter.
Original Bald
Stylized lighting and strong color contrast remain coherent after conversion.
Original portrait before bald conversion.
Generated bald portrait after Bald Converter.
Original Bald
Facial structure, expression, and background context stay stable.
Original portrait before bald conversion.
Generated bald portrait after Bald Converter.
Original Bald
Close crops and high-detail face regions remain usable for transfer.
Original portrait before bald conversion.
Generated bald portrait after Bald Converter.
Original Bald
The conversion remains robust to masks, saturated color, and graphic portrait styles.
Original portrait before bald conversion.
Generated bald portrait after Bald Converter.
Original Bald
Long hair is removed while preserving identity cues and lighting.
Original portrait before bald conversion.
Generated bald portrait after Bald Converter.
Original Bald
Illustrative inputs can still become clean bald-source canvases.
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald
Original and bald portrait pair from the Bald Converter gallery.
OriginalBald

Bald Converter and Baldy

A dedicated bald-generation stage for in-context transfer.

The Bald Converter produces realistic bald versions of source portraits using LoRA-based in-context adaptation of FLUX. Baldy provides paired bald/original supervision across diverse identities and imaging conditions.

Baldy dataset generation and Bald Converter finetuning diagram.
Baldy dataset generation and LoRA-based Bald Converter finetuning.
  • 6,400paired images
  • FLUXbase model
  • LoRAin-context adaptation
Baldy dataset LoRA weights

Code and Assets

Source preview, dataset, and weights are available.

The official SIGGRAPH 2026 implementation snapshot is online. Packaging, dependency manifests, and console-script entry points are still being finalized.

Paper arXiv preprint arXiv:2606.12562 Read on arXiv Publication ACM DOI 10.1145/3799902.3811046 Open DOI GitHub HairPort source code Official SIGGRAPH 2026 implementation snapshot. Open repository README Install and run Setup, source-preview caveats, and usage examples. Read setup Dataset Baldy 6,400 paired bald and original portraits. Download dataset Weights Bald Converter LoRAs Adapters for in-context bald generation. Download weights

News

Latest updates.

  1. HairPort preprint is now available on arXiv:2606.12562.
  2. ACM assigned the HairPort DOI: 10.1145/3799902.3811046.
  3. HairPort will appear at ACM SIGGRAPH 2026.
  4. Project page launched at deepmancer.github.io/HairPort.
  5. Initial source preview released; packaging and dependency manifests will be finalized soon.
  6. Baldy dataset and Bald Converter LoRA weights released on Hugging Face.

Acknowledgements

With thanks.

We thank the anonymous reviewers for their insightful comments and constructive feedback, and Xuebin Qin for valuable early discussions related to this work. This work was supported in part by NSERC.

Citation

Use HairPort in your research?

If our work helped, please cite the ACM SIGGRAPH 2026 paper.

DOI: 10.1145/3799902.3811046. ACM ISBN: 979-8-4007-2554-8/2026/07.

Preprint: arXiv:2606.12562.

BibTeX 
@inproceedings{heidari2026hairport,
  title     = {HairPort: In-context 3D-aware Hair Import and Transfer for Images},
  author    = {A. Heidari and A. Alimohammadi and W. Michel Pinto Lira and A. Bar-Lev and A. Mahdavi-Amiri},
  booktitle = {Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers (SIGGRAPH Conference Papers '26)},
  year      = {2026},
  isbn      = {979-8-4007-2554-8/2026/07},
  doi       = {10.1145/3799902.3811046},
  url       = {https://doi.org/10.1145/3799902.3811046},
  location  = {Los Angeles, CA, USA}
}