---
license: mit
---

# Semantics Lead the Way: Harmonizing Semantic and Texture Modeling with Asynchronous Latent Diffusion

## 🚩 Overview
<p align="center">
  <img src="https://raw.githubusercontent.com/yuemingPAN/SFD/main/images/teaser_v5.png" width="95%">
</p>

<div align="center" style="max-width:900px; text-align:justify; font-size:14px; line-height:1.5;">
  <p>
    <strong>(a) Overview of Semantic-First Diffusion (SFD).</strong>
    Semantics (dashed curve) and textures (solid curve) follow asynchronous denoising trajectories.
    SFD operates in three phases:
    <span style="color:#d62728;">Stage I – Semantic initialization</span>, where semantic latents denoise first;
    <span style="color:#4472c4;">Stage II – Asynchronous generation</span>, where semantics and textures denoise jointly but asynchronously, with semantics ahead of textures;
    <span style="color:#2ca02c;">Stage III – Texture completion</span>, where only textures continue refining.
    After denoising, the generated semantic latent <b>s₁</b> is discarded, and the final image is decoded solely from the texture latent <b>z₁</b>.
    <strong>(b) Training convergence on ImageNet 256×256 without guidance.</strong>
    SFD achieves substantially faster convergence than DiT-XL/2 and LightningDiT-XL/1 by approximately <b>100×</b> and <b>33.3×</b>, respectively.
  </p>
</div>

---

## ✨ Highlights
- We propose **Semantic-First Diffusion (SFD)**, a novel latent diffusion paradigm that performs asynchronous denoising on semantic and texture latents, allowing semantics to denoise earlier and subsequently guide texture generation.  
- **SFD achieves state-of-the-art FID score of 1.04** on ImageNet 256×256 generation.  
- Exhibits **100×** and **33.3× faster** training convergence compared to **DiT** and **LightningDiT**, respectively.

---

## 🧪 Quantitative Results
Explicitly **leading semantics ahead of textures with a moderate offset (Δt = 0.3)** achieves an optimal balance between early semantic stabilization and texture collaboration, effectively harmonizing their joint modeling.
<p align="center">
  <img src="https://raw.githubusercontent.com/yuemingPAN/SFD/main/images/fid_vs_delta_t.png" width="50%">
</p>


### With AutoGuidance

| Model   | Epochs | #Params | FID (NPU) |
|:--------|:-------:|:--------:|:----------:|
| SFD-XL  | 80      | 675M     | 1.30 |
| SFD-XL  | 800     | 675M     | **1.06** |
| SFD-XXL | 80      | 1.0B     | 1.19 |
| SFD-XXL | 800     | 1.0B     | **1.04** |


---

## 🎨 Visual Results

<p align="center">
  <img src="https://raw.githubusercontent.com/yuemingPAN/SFD/main/images/demo_Sample.png" width="90%">
</p>

---

## 🔗 Links
- 🌐 **Project Page:** [https://yuemingpan.github.io/SFD.github.io/](https://yuemingpan.github.io/SFD.github.io/)
- 📄 **Paper (arXiv):** [https://arxiv.org/pdf/2512.04926](https://arxiv.org/pdf/2512.04926)
- 💾 **Code:** [https://github.com/yuemingPAN/SFD](https://github.com/yuemingPAN/SFD)
- 🧰 **License:** MIT

---

## 🧩 Citation
```bibtex
@article{Pan2025SFD,
  title={Semantics Lead the Way: Harmonizing Semantic and Texture Modeling with Asynchronous Latent Diffusion},
  author={Pan, Yueming and Feng, Ruoyu and Dai, Qi and Wang, Yuqi and Lin, Wenfeng and Guo, Mingyu and Luo, Chong and Zheng, Nanning},
  journal={arXiv preprint arXiv:2512.04926},
  year={2025}
}