Research map / Video / Video generation
GenCompositor: Generative Video Compositing with Diffusion Transformer
Shuzhou Yang, Xiaoyu Li, Xiaodong Cun, Guangzhi Wang, Lingen Li, Ying Shan, Jian Zhang
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 |
|---|---|---|---|
| replace cross-attention injection with DiT fusion block (token concatenation + full self-attention) (ablation) | GenCompositor · HYouTube | PSNR: 19.8940 -> 42.0010 | Improved |
| add background preservation branch (BPBranch) (ablation) | GenCompositor · Trajectory-controlled generation | Background consistency: 89.62% -> 93.43% | Improved |
| add BPBranch (ablation) | GenCompositor · HYouTube | PSNR: 40.0099 -> 42.0010 | Improved |
| add luminance (gamma) augmentation of foreground during training (ablation) | GenCompositor · HYouTube | LPIPS: 0.0520 -> 0.0385 | Improved |
| add mask inflation with Gaussian filter (ablation) | GenCompositor · HYouTube | PSNR: 41.8553 -> 42.0010 | Improved |
| add mask inflation (ablation) | GenCompositor · Trajectory-controlled generation | Background consistency: 91.62% -> 93.43% | Improved |
| use ERoPE stream-specific position shift instead of shared RoPE for foreground tokens (ablation) | GenCompositor · training loss | training loss | Improved |
| shift ERoPE along height vs width vs time (ablation) | ERoPE · training loss | training loss | No clear effect |
| encode foreground with CLIP instead of VAE under cross-attention (ablation) | w/o fusion block variant · qualitative | foreground injection | Hurt |
| GenCompositor vs video harmonization methods Harmonizer and VTT | video harmonization · HYouTube | PSNR: 40.0251 -> 42.0010 | Improved |
| GenCompositor vs VACE for trajectory-controlled compositing | trajectory-controlled generation · 40 video sets | FVD: 942.52 -> 535.71 | Improved |
Later papers that compared with this method
1 later accepted paper reports GenCompositor as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- FlexComposer: Unified Video Compositing from Images to Dynamic Footage with Flexible Trajectory Control (ECCV 2026 · better on 3 of 3 shared results)
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.
- Tora — Tora: Trajectory-oriented Diffusion Transformer for Video Generation (CVPR 2025) · better on 2 of 2 shared results
- Revideo — ReVideo: Remake a Video with Motion and Content Control (NeurIPS 2024) · better on 2 of 2 shared results
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
- DiTCtrl: Exploring Attention Control in Multi-Modal Diffusion Transformer for Tuning-Free Multi-Prompt Longer Video Generation (CVPR 2025 · similarity 0.89)
- Generative Video Propagation (CVPR 2025 · similarity 0.89)
- VideoPoet: A Large Language Model for Zero-Shot Video Generation (ICML 2024 · similarity 0.88)
- VD3D: Taming Large Video Diffusion Transformers for 3D Camera Control (ICLR 2025 · similarity 0.88)
- Generative Rendering: Controllable 4D-Guided Video Generation with 2D Diffusion Models (CVPR 2024 · similarity 0.88)
- Video Motion Transfer with Diffusion Transformers (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.
