# COSY: Compositional 3DGS Synthesis for Disentangled Human Head Editing

arXiv: https://arxiv.org/abs/2605.24114 · HTML: https://mlsanity.com/paper/2605.24114/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 39 of its 47 references that are in the map.
- First posted on arXiv 22 May 2026.

## Builds on (most cited first, 30 of 39)

- [3D Gaussian Splatting for Real-Time Radiance Field Rendering](https://mlsanity.com/paper/2308.04079/) (2023)
- [Rethinking the Inception Architecture for Computer Vision](https://mlsanity.com/paper/1512.00567/) (2015)
- [Analyzing and Improving the Image Quality of StyleGAN](https://mlsanity.com/paper/1912.04958/) (2019)
- [Large Scale GAN Training for High Fidelity Natural Image Synthesis](https://mlsanity.com/paper/1809.11096/) (2018)
- [Conditional Generative Adversarial Nets](https://mlsanity.com/paper/1411.1784/) (2014)
- [Zero-1-to-3: Zero-shot One Image to 3D Object](https://mlsanity.com/paper/2303.11328/) (2023)
- [Point Transformer](https://mlsanity.com/paper/2011.00931/) (2020)
- [Efficient Geometry-aware 3D Generative Adversarial Networks](https://mlsanity.com/paper/2112.07945/) (2021)
- [FLUX.1 Kontext: Flow Matching for In-Context Image Generation and Editing in Latent Space](https://mlsanity.com/paper/2506.15742/) (2025)
- [PCT: Point cloud transformer](https://mlsanity.com/paper/2012.09688/) (2020)
- [pi-GAN: Periodic Implicit Generative Adversarial Networks for 3D-Aware Image Synthesis](https://mlsanity.com/paper/2012.00926/) (2020)
- [Point-E: A System for Generating 3D Point Clouds from Complex Prompts](https://mlsanity.com/paper/2212.08751/) (2022)
- [BiSeNet: Bilateral Segmentation Network for Real-time Semantic Segmentation](https://mlsanity.com/paper/1808.00897/) (2018)
- [SyncDreamer: Generating Multiview-consistent Images from a Single-view Image](https://mlsanity.com/paper/2309.03453/) (2023)
- [StyleGAN-XL: Scaling StyleGAN to Large Diverse Datasets](https://mlsanity.com/paper/2202.00273/) (2022)
- [InterFaceGAN: Interpreting the Disentangled Face Representation Learned by GANs](https://mlsanity.com/paper/2005.09635/) (2020)
- [StyleFlow: Attribute-conditioned Exploration of StyleGAN-Generated Images using Conditional Continuous Normalizing Flows](https://mlsanity.com/paper/2008.02401/) (2020)
- [SEAN: Image Synthesis with Semantic Region-Adaptive Normalization](https://mlsanity.com/paper/1911.12861/) (2019)
- [The Role of ImageNet Classes in Fr'echet Inception Distance](https://mlsanity.com/paper/2203.06026/) (2022)
- [FENeRF: Face Editing in Neural Radiance Fields](https://mlsanity.com/paper/2111.15490/) (2021)
- [PanoHead: Geometry-Aware 3D Full-Head Synthesis in 360circ](https://mlsanity.com/paper/2303.13071/) (2023)
- [GIRAFFE HD: A High-Resolution 3D-aware Generative Model](https://mlsanity.com/paper/2203.14954/) (2022)
- [GRAM-HD: 3D-Consistent Image Generation at High Resolution with Generative Radiance Manifolds](https://mlsanity.com/paper/2206.07255/) (2022)
- [DiffMorpher: Unleashing the Capability of Diffusion Models for Image Morphing](https://mlsanity.com/paper/2312.07409/) (2023)
- [SofGAN: A Portrait Image Generator with Dynamic Styling](https://mlsanity.com/paper/2007.03780/) (2020)
- [Seedream 2.0: A Native Chinese-English Bilingual Image Generation Foundation Model](https://mlsanity.com/paper/2503.07703/) (2025)
- [GGHead: Fast and Generalizable 3D Gaussian Heads](https://mlsanity.com/paper/2406.09377/) (2024)
- [StyleFusion: A Generative Model for Disentangling Spatial Segments](https://mlsanity.com/paper/2107.07437/) (2021)
- [Morphable Diffusion: 3D-Consistent Diffusion for Single-image Avatar Creation](https://mlsanity.com/paper/2401.04728/) (2024)
- [Gaussian3Diff: 3D Gaussian Diffusion for 3D Full Head Synthesis and Editing](https://mlsanity.com/paper/2312.03763/) (2023)

## Abstract (arXiv)

Recent 3D Gaussian Splatting (3DGS) GANs for human heads synthesize and render photorealistic 3D models in real-time and offer a vast variety in identity and appearance. However, controlling specific semantic attributes such as hair color or glasses remains challenging, as edits in the entangled latent space often induce unintended changes in identity or appearance. Although there are several methods that aim to disentangle the latent space post training by estimating directions that only modify certain features, these methods cannot guarantee complete disentanglement and often require pre-trained classifiers. In our approach, we propose a new generator architecture that synthesizes components, such as hair, skin, glasses, and torso, completely independently. This allows for changing the latent vector for one region while keeping the remaining parts fixed. Further, we achieve this separation using only sparse information such as the hair or skin color, eliminating the requirement of segmentation masks or geometric priors, often seen in prior work. To ensure matching shape and lighting conditions during editing, we allow minimal shared information via context tokens between the independent generators. These tokens even allow us to control the shape and light, without any prior annotation. Compared to existing works on GAN-based generation and editing, our method shows better disentanglement, more precise editing control, and competitive visual quality.