Research map / Video / Video generation
FlowDirector: Training-Free Flow Steering for Precise Text-to-Video Editing
Guangzhao Li, Yanming Yang, Chenxi Song, Xiaohong Liu, Chi 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 |
|---|---|---|---|
| add direction-aware flow correction (DA-FC) (ablation) | FlowDirector · 150 video edits (41 frames) | Qedit: 25.42 -> 27.19 | Improved |
| add direction-aware flow correction (DA-FC) (ablation) | FlowDirector · 150 video edits (41 frames) | CLIP-T: 32.25 -> 34.64 | Improved |
| add motion-appearance decoupling (MAD-FC) (ablation) | FlowDirector · 150 video edits (41 frames) | WarpSSIM: 69.26 -> 78.49 | Improved |
| add motion-appearance decoupling (MAD-FC) (ablation) | FlowDirector · 150 video edits (41 frames) | Qedit: 24.04 -> 27.19 | Improved |
| add differential averaging guidance (DAG) (ablation) | FlowDirector · 150 video edits (41 frames) | Qedit: 27.11 -> 27.19 | Improved |
| add differential averaging guidance (DAG) (ablation) | FlowDirector · 150 video edits (41 frames) | WarpSSIM: 78.32 -> 78.49 | Mixed |
| remove parallel component of the editing flow (ablation) | DA-FC · 150 video edits (41 frames) | Qedit: 25.42 -> 26.04 | Improved |
| amplify opposing component with alpha 0.25 versus no amplification (ablation) | DA-FC · 150 video edits (41 frames) | Qedit: 26.04 -> 27.19 | Improved |
| increase alpha amplification (ablation) | DA-FC · 150 video edits (41 frames) | WarpSSIM: 78.85 -> 78.49 | Hurt |
| use direction-aware correction versus skipping early steps (ablation) | Director ODE (skip) · 150 video edits (41 frames) | CLIP-T: 32.23 -> 34.64 | Improved |
| skip early steps in direct ODE (ablation) | Director ODE · 150 video edits (41 frames) | WarpSSIM: 62.71 -> 78.90 | Improved |
| use DAG with 4 rounds instead of 20-round conventional averaging | conventional averaging · 41-frame editing time | editing time: 19 min 3s -> 2 min 54s | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- RAVE — RAVE: Randomized Noise Shuffling for Fast and Consistent Video Editing with Diffusion Models (CVPR 2024) · better on 4 of 4 shared results
- FLATTEN — FLATTEN: optical FLow-guided ATTENtion for consistent text-to-video editing (ICLR 2024) · better on 4 of 4 shared results
- VideoDirector — VideoDirector: Precise Video Editing via Text-to-Video Models (CVPR 2025) · 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
- VideoDirector: Precise Video Editing via Text-to-Video Models (CVPR 2025 · similarity 0.90)
- SplitFlow: Flow Decomposition for Inversion-Free Text-to-Image Editing (NeurIPS 2025 · similarity 0.88)
- Unveil Inversion and Invariance in Flow Transformer for Versatile Image Editing (CVPR 2025 · similarity 0.88)
- VINCIE-NExT: Unlocking Video Editing from Images via In-Context Modeling (NeurIPS 2026 · similarity 0.87)
- Dynamic Motion Blending for Versatile Motion Editing (CVPR 2025 · similarity 0.87)
- Taming Rectified Flow for Inversion and Editing (ICML 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.
