Research map / Video / Video generation
Towards Consistent Video Editing with Text-to-Image Diffusion Models
Zicheng Zhang, Bonan Li, Xuecheng Nie, Congying Han, Tiande Guo, Luoqi 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 |
|---|---|---|---|
| Replace Layer Norm in temporal attention with Instance Centering (about 20% faster than Instance Norm) (ablation) | EI^2 temporal attention · Video editing | computational speed: about 20% | Improved |
| Replace Layer Norm in Tune-A-Video TA module with Instance Norm (ablation) | Tune-A-Video · Video editing | textual guidance (qualitative) | Improved |
| Remove rescaling parameters of Layer Norm or remove the module (ablation) | Tune-A-Video · Video editing | covariate shift / editing (qualitative) | No clear effect |
| Add Instance Centering and weight (spectral) normalization (ablation) | EI^2 STAM · Video editing | editing quality (qualitative) | Improved |
| Replace SCA with FFAM in combination with STAM (ablation) | EI^2 · Video editing | temporal consistency (qualitative) | Improved |
| Full spatial-temporal attention over 8 frames instead of FFAM (ablation) | EI^2 · Video editing | memory and training time: 32GiB; ~45min | Hurt |
| Use EI^2 instead of Tune-A-Video | Tune-A-Video · Textual alignment | CLIP score: 28.72 -> 29.84 | Improved |
| Use EI^2 instead of Tune-A-Video | Tune-A-Video · Frame consistency | CLIP score: 96.64 -> 95.94 | Hurt |
| Use EI^2 instead of Tune-A-Video | Tune-A-Video · Textual alignment | user vote: 15.4\% -> 42.7\% | Improved |
| Use EI^2 instead of Tune-A-Video | Tune-A-Video · Frame consistency | user vote: 25.6\% -> 34.9\% | Improved |
| EI^2 training overhead vs Tune-A-Video | Tune-A-Video · Training runtime | minutes: 11.0 -> 12.5 | Hurt |
| Insert STAM before vs after CA module (ablation) | EI^2 · Video editing | visual quality (qualitative) | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- Tune-A-Video — Tune-A-Video: One-Shot Tuning of Image Diffusion Models for Text-to-Video Generation (ICCV 2023) · better on 3 of 8 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
- FLATTEN: optical FLow-guided ATTENtion for consistent text-to-video editing (ICLR 2024 · similarity 0.91)
- COVE: Unleashing the Diffusion Feature Correspondence for Consistent Video Editing (NeurIPS 2024 · similarity 0.90)
- TokenFlow: Consistent Diffusion Features for Consistent Video Editing (ICLR 2024 · similarity 0.90)
- VideoDirector: Precise Video Editing via Text-to-Video Models (CVPR 2025 · similarity 0.89)
- FADE: Frequency-Aware Diffusion Model Factorization for Video Editing (CVPR 2025 · similarity 0.88)
- VideoGrain: Modulating Space-Time Attention for Multi-Grained Video Editing (ICLR 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.
