![[2508.04228] LayerT2V: A Unified Multi-Layer Video Generation Framework](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2508.04228] LayerT2V: A Unified Multi-Layer Video Generation Framework
Summary
LayerT2V presents a novel framework for multi-layer video generation, enabling the creation of editable video layers that enhance professional workflows and improve visual fidelity.
Why It Matters
This research addresses the limitations of current text-to-video generation methods by introducing a unified framework that allows for multi-layer outputs. This innovation is significant for industries relying on video content creation, as it enhances flexibility and quality in video production.
Key Takeaways
- LayerT2V generates multiple semantically consistent video layers in one pass.
- The framework improves temporal coherence and cross-layer consistency.
- Introduces VidLayer, a large-scale dataset for multi-layer video generation.
- Utilizes a shared DiT backbone with enhancements for layer-aware processing.
- Demonstrates superior performance in visual fidelity compared to existing methods.
Computer Science > Computer Vision and Pattern Recognition arXiv:2508.04228 (cs) [Submitted on 6 Aug 2025 (v1), last revised 26 Feb 2026 (this version, v2)] Title:LayerT2V: A Unified Multi-Layer Video Generation Framework Authors:Guangzhao Li, Kangrui Cen, Baixuan Zhao, Yi Xin, Siqi Luo, Guangtao Zhai, Lei Zhang, Xiaohong Liu View a PDF of the paper titled LayerT2V: A Unified Multi-Layer Video Generation Framework, by Guangzhao Li and 7 other authors View PDF HTML (experimental) Abstract:Text-to-video generation has advanced rapidly, but existing methods typically output only the final composited video and lack editable layered representations, limiting their use in professional workflows. We propose \textbf{LayerT2V}, a unified multi-layer video generation framework that produces multiple semantically consistent outputs in a single inference pass: the full video, an independent background layer, and multiple foreground RGB layers with corresponding alpha mattes. Our key insight is that recent video generation backbones use high compression in both time and space, enabling us to serialize multiple layer representations along the temporal dimension and jointly model them on a shared generation trajectory. This turns cross-layer consistency into an intrinsic objective, improving semantic alignment and temporal coherence. To mitigate layer ambiguity and conditional leakage, we augment a shared DiT backbone with LayerAdaLN and layer-aware cross-attention modulation. LayerT2V i...