Research map / Image generation / Autoregressive and latent generation
Semantics Lead the Way: Harmonizing Semantic and Texture Modeling with Asynchronous Latent Diffusion
Yueming Pan, Ruoyu Feng, Qi Dai, Yuqi Wang, Wenfeng Lin, Mingyu Guo, Chong Luo, Nanning Zheng
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add REPA (ablation) | LightningDiT-XL · ImageNet 256x256 (ablation, LightningDiT-XL 400K) | FID: 8.17 -> 7.08 | Improved |
| add SemVAE semantic latents (ablation) | LightningDiT-XL+REPA · ImageNet 256x256 (ablation, LightningDiT-XL 400K) | FID: 7.08 -> 5.24 | Improved |
| add semantic-first asynchronous denoising (ablation) | LightningDiT-XL+REPA+SemVAE · ImageNet 256x256 (ablation, LightningDiT-XL 400K) | FID: 5.24 -> 3.03 | Improved |
| use SemVAE instead of PCA compression (ablation) | SFD · ImageNet 256x256 (ablation, LightningDiT-XL 400K) | FID: 4.06 -> 3.03 | Improved |
| add semantic-first mechanism (ablation) | ReDi · ImageNet 256x256 (ablation, LightningDiT-XL 400K) | FID: 5.33 -> 4.41 | Improved |
| add SFD (ablation) | VA-VAE · ImageNet 256x256 (ablation, LightningDiT-XL 400K) | FID: 4.52 -> 4.14 | Improved |
| use SD-VAE texture latent instead of VA-VAE (ablation) | SFD · ImageNet 256x256 (ablation, LightningDiT-XL 400K) | FID: 4.14 -> 3.03 | Improved |
| temporal offset 0.3 instead of 0 (joint denoising) (ablation) | SFD · ImageNet 256x256 (400K iter) | FID: 3.03 | Improved |
| increase offset beyond 0.3 up to 1.0 (ablation) | SFD · ImageNet 256x256 (400K iter) | FID | Hurt |
| DINOv2-B target instead of MAE-B (ablation) | SemVAE · ImageNet 256x256 (ablation, LightningDiT-XL 400K) | FID: 6.29 -> 3.03 | Improved |
| scale target encoder from DINOv2-B to DINOv2-L (ablation) | SemVAE · ImageNet 256x256 (ablation, LightningDiT-XL 400K) | FID: 3.03 -> 2.97 | Improved |
| increase semantic channels from 2 to 16 (ablation) | SemVAE · ImageNet 256x256 (ablation, LightningDiT-XL 400K) | FID: 3.90 -> 3.03 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- REPA-E — REPA-E: Unlocking VAE for End-to-End Tuning of Latent Diffusion Transformers (ICCV 2025) · better on 1 of 1 shared result
- ReDi — ReDi: Efficient Learning-Free Diffusion Inference via Trajectory Retrieval (ICML 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Boosting Generative Image Modeling via Joint Image-Feature Synthesis (NeurIPS 2025 · similarity 0.90)
- Noise Diffusion for Enhancing Semantic Faithfulness in Text-to-Image Synthesis (CVPR 2025 · similarity 0.89)
- Nested Diffusion Models Using Hierarchical Latent Priors (CVPR 2025 · similarity 0.89)
- Single Mesh Diffusion Models with Field Latents for Texture Generation (CVPR 2024 · similarity 0.89)
- Würstchen: An Efficient Architecture for Large-Scale Text-to-Image Diffusion Models (ICLR 2024 · similarity 0.89)
- FAM Diffusion: Frequency and Attention Modulation for High-Resolution Image Generation with Stable Diffusion (CVPR 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
