Research map / Video / Video generation
LayerT2V: A Unified Multi-Layer Video Generation Framework
Guangzhao Li, Kangrui Cen, Baixuan Zhao, Yi Xin, Siqi Luo, Guangtao Zhai, Lei Zhang, Xiaohong Liu
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 LayerAdaLN to vanilla temporal concatenation (ablation) | LayerT2V · VidLayer held-out, FG layer | subject consistency: 0.931 -> 0.973 | Improved |
| add LayeredCrossAttention on top of LayerAdaLN (ablation) | LayerT2V · VidLayer held-out, full-video layer | text alignment: 0.195 -> 0.214 | Improved |
| add LayeredCrossAttention on top of LayerAdaLN (ablation) | LayerT2V · VidLayer held-out, FG layer | subject consistency: 0.973 -> 0.983 | Improved |
| fine-tune Wan VAE decoder with LoRA instead of training a native Mask VAE from scratch (ablation) | LayerT2V · VidLayer held-out, FG layer | aesthetic quality: 0.4432 -> 0.4971 | Improved |
| use VAE LoRA instead of native Mask VAE (ablation) | LayerT2V · VidLayer held-out, BG layer | text alignment: 0.2102 -> 0.2307 | Improved |
| use VAE LoRA instead of native Mask VAE (ablation) | LayerT2V · VidLayer held-out, BG layer | temporal flickering: 0.9837 vs 0.9830 | No clear effect |
| extend 3D RoPE to 4D with a layer axis (ablation) | LayerT2V · qualitative | visual quality | Hurt |
| replace LayerFlow with LayerT2V | LayerFlow · VidLayer held-out, FG layer | aesthetic quality: 0.4591 -> 0.4971 | Improved |
| replace LayerFlow with LayerT2V | LayerFlow · VidLayer held-out, BG layer | subject consistency: 0.9682 -> 0.9838 | Improved |
| replace LayerFlow with LayerT2V | LayerFlow · User study | preference for aesthetic quality: 0.120 -> 0.724 | Improved |
| use native Mask VAE instead of LayerFlow | LayerT2V (Native Mask VAE) · VidLayer held-out, full-video layer | motion smoothness: 0.9788 -> 0.9776 | Hurt |
Closest papers by meaning
- VideoTetris: Towards Compositional Text-to-Video Generation (NeurIPS 2024 · similarity 0.89)
- StreamingT2V: Consistent, Dynamic, and Extendable Long Video Generation from Text (CVPR 2025 · similarity 0.89)
- Mask^2DiT: Dual Mask-based Diffusion Transformer for Multi-Scene Long Video Generation (CVPR 2025 · similarity 0.89)
- DiTCtrl: Exploring Attention Control in Multi-Modal Diffusion Transformer for Tuning-Free Multi-Prompt Longer Video Generation (CVPR 2025 · similarity 0.87)
- Snap Video: Scaled Spatiotemporal Transformers for Text-to-Video Synthesis (CVPR 2024 · similarity 0.87)
- Encapsulated Composition of Text-to-Image and Text-to-Video Models for High-Quality Video Synthesis (CVPR 2025 · similarity 0.87)
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.
