Research map / Video / Video generation
MEVG : Multi-event Video Generation with Text-to-Video Models
Gyeongrok Oh, Jaehwan Jeong, Sieun Kim, Wonmin Byeon, Jinkyu Kim, Sungwoong Kim, Sangpil Kim
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 structure-guided sampling on top of LFAI and dynamic noise (ablation) | MEVG · Ablation user study | Preference (1-5): 2.78 -> 3.27 | Improved |
| add structure-guided sampling on top of LFAI and dynamic noise (ablation) | MEVG · Ablation user study | Semantics (1-5): 3.58 -> 3.47 | Hurt |
| add dynamic noise alone to base inversion (ablation) | MEVG · Ablation user study | Preference (1-5): 2.89 -> 2.51 | Hurt |
| add last frame-aware inversion to dynamic noise (ablation) | MEVG · Ablation user study | Preference (1-5): 2.51 -> 2.78 | Improved |
| add all three modules (LFAI, DN, SGS) to base DDIM inversion (ablation) | MEVG · Ablation user study | Temporal consistency (1-5): 3.42 -> 3.70 | Improved |
| use a constant kappa instead of the decreasing noise schedule (ablation) | Dynamic noise · Qualitative ablation | visual quality | Hurt |
| apply last-frame guidance to more than the first frame (ablation) | Last frame-aware inversion · Qualitative ablation | motion diversity | Hurt |
| raise SGS guidance scale from 7 to 50 (ablation) | Structure-guided sampling · Hyper-parameter analysis | CLIP-Text / CLIP-Image: CLIP-Text 30.674 -> 30.102; CLIP-Image 0.945 -> 0.954 | Mixed |
| MEVG vs Text2Video-Zero on CLIP-Text | MEVG · Multi-event video generation | CLIP-Text: 0.322 -> 0.309 | Hurt |
| MEVG vs Gen-L-Video and VidRD on human evaluation | MEVG · Multi-event video generation, human evaluation | Temporal consistency (1-5): 3.40 -> 3.82 | Improved |
Closest papers by meaning
- EIDT-V: Exploiting Intersections in Diffusion Trajectories for Model-Agnostic, Zero-Shot, Training-Free Text-to-Video Generation (CVPR 2025 · similarity 0.91)
- Grid Diffusion Models for Text-to-Video Generation (CVPR 2024 · similarity 0.91)
- FreeNoise: Tuning-Free Longer Video Diffusion via Noise Rescheduling (ICLR 2024 · similarity 0.90)
- Vivid-ZOO: Multi-View Video Generation with Diffusion Model (NeurIPS 2024 · similarity 0.90)
- VideoTetris: Towards Compositional Text-to-Video Generation (NeurIPS 2024 · similarity 0.90)
- A Recipe for Scaling up Text-to-Video Generation with Text-free Videos (CVPR 2024 · similarity 0.89)
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.
