Research map / Video / Video generation
FlexComposer: Unified Video Compositing from Images to Dynamic Footage with Flexible Trajectory Control
Songchun Zhang, Sitong Guo, Xianghao Kong, Pengwei Liu, Yuwei Guo, lvmin zhang, Anyi Rao
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 canonical foreground representation (stabilization) (ablation) | FlexComposer · Ablation | EPE: 4.87 -> 2.39 | Improved |
| add canonical foreground representation (stabilization) (ablation) | FlexComposer · Ablation | FVD: 145.3 -> 79.8 | Improved |
| add static image expansion (ablation) | FlexComposer · Ablation | EPE: 2.94 -> 2.39 | Improved |
| add Gaussian noise injection in static expansion (ablation) | FlexComposer · Ablation | FVD: 118.2 -> 79.8 | Improved |
| add relighting augmentation (ablation) | FlexComposer · Ablation | LPIPS: 0.058 -> 0.036 | Improved |
| add relighting augmentation (ablation) | FlexComposer · Ablation | EPE: 2.61 -> 2.39 | No clear effect |
| add visibility gate (ablation) | FlexComposer · Ablation | FVD: 92.4 -> 79.8 | Improved |
| add parameter-free latent transport (vs learned adapter) (ablation) | FlexComposer · Ablation | EPE: 5.68 -> 2.39 | Improved |
| add parameter-free latent transport (vs learned adapter) (ablation) | FlexComposer · Ablation | FVD: 138.9 -> 79.8 | Improved |
| add real-data stage to synthetic-only training (ablation) | FlexComposer · Ablation | FVD: 154.2 -> 79.8 | Improved |
| add synthetic warm-up stage to real-only training (ablation) | FlexComposer · Ablation | EPE: 3.82 -> 2.39 | Improved |
| add generative data refinement stage (ablation) | FlexComposer · Ablation | FVD: 98.1 -> 79.8 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 5 of 5 on most shared results.
- ImageConductor — Image Conductor: Precision Control for Interactive Video Synthesis (AAAI 2025) · better on 4 of 4 shared results
- LeviTor — LeviTor: 3D Trajectory Oriented Image-to-Video Synthesis (CVPR 2025) · better on 4 of 4 shared results
- Tora — Tora: Trajectory-oriented Diffusion Transformer for Video Generation (CVPR 2025) · better on 4 of 4 shared results
- Wan-Move — Wan-Move: Motion-controllable Video Generation via Latent Trajectory Guidance (NeurIPS 2025) · better on 4 of 4 shared results
- GenCompositor — GenCompositor: Generative Video Compositing with Diffusion Transformer (ICLR 2026) · better on 3 of 3 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
- GenCompositor: Generative Video Compositing with Diffusion Transformer (ICLR 2026 · similarity 0.90)
- Wan-Move: Motion-controllable Video Generation via Latent Trajectory Guidance (NeurIPS 2025 · similarity 0.89)
- FlexAM: Flexible Appearance-Motion Decomposition for Versatile Video Generation Control (ECCV 2026 · similarity 0.89)
- FlexTraj: Image-to-Video Generation with Flexible Point Trajectory Control (CVPR 2026 · similarity 0.89)
- ONE-SHOT: Compositional Human-Environment Video Synthesis via Spatial-Decoupled Motion Injection and Hybrid Context Integration (NeurIPS 2026 · similarity 0.87)
- DynaTokens: Teaching Dynamics to Camera-Controlled Video Models at Test Time (NeurIPS 2026 · 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.
