Research map / Video / Video generation
Customize-A-Video: One-Shot Motion Customization of Text-to-Video Diffusion Models
Yixuan Ren, Yang Zhou, Jimei Yang, Jing Shi, Difan Liu, Feng Liu, Mingi Kwon, Abhinav Shrivastava
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 TextInv appearance absorber before training Temporal LoRA (ablation) | Customize-A-Video (T-LoRA) · LOVEU-TGVE subset | text alignment (CLIPScore): 31.687 -> 32.632 | Improved |
| add S-LoRA appearance absorber before training Temporal LoRA (ablation) | Customize-A-Video (T-LoRA) · LOVEU-TGVE subset | text alignment (CLIPScore): 31.687 -> 31.913 | Improved |
| use dual appearance absorbers (S-LoRA + TextInv) instead of single (ablation) | Customize-A-Video · LOVEU-TGVE subset | diversity (LPIPS): 0.621 -> 0.631 | Improved |
| use dual appearance absorbers instead of TextInv only (ablation) | Customize-A-Video · LOVEU-TGVE subset | text alignment (CLIPScore): 32.632 -> 32.193 | Hurt |
| apply LoRA on spatial attention (S-LoRA or ST-LoRA) instead of temporal attention only (ablation) | motion LoRA · qualitative ablation | motion modeling | Hurt |
| use T-LoRA with appearance absorber instead of ModelScope text prompt only | ModelScope T2V · LOVEU-TGVE subset | diversity (LPIPS): 0.636 -> 0.621 | Hurt |
| use Customize-A-Video instead of Tune-A-Video/Video-P2P without DDIM inversion | one-shot motion customization · LOVEU-TGVE subset | diversity (LPIPS): 0.348 -> 0.621 (Tune-A-Video w/o DDIM) | Improved |
| use Customize-A-Video instead of MotionDirector | one-shot motion customization · user study | motion fidelity: 3.33 -> 3.72 | Improved |
| use Customize-A-Video instead of MotionDirector | one-shot motion customization · LOVEU-TGVE subset | diversity (LPIPS): 0.606 -> 0.621 | Improved |
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.
- Video-P2P — Video-P2P: Video Editing with Cross-attention Control (CVPR 2024) · better on 2 of 2 shared results
- Tune-A-Video — Tune-A-Video: One-Shot Tuning of Image Diffusion Models for Text-to-Video Generation (ICCV 2023) · better on 1 of 1 shared result
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
- VMC: Video Motion Customization using Temporal Attention Adaption for Text-to-Video Diffusion Models (CVPR 2024 · similarity 0.93)
- CustomTTT: Motion and Appearance Customized Video Generation via Test-Time Training (AAAI 2025 · similarity 0.92)
- VideoMage: Multi-Subject and Motion Customization of Text-to-Video Diffusion Models (CVPR 2025 · similarity 0.91)
- AnimateDiff: Animate Your Personalized Text-to-Image Diffusion Models without Specific Tuning (ICLR 2024 · similarity 0.89)
- CustomCrafter: Customized Video Generation with Preserving Motion and Concept Composition Abilities (AAAI 2025 · similarity 0.88)
- Space-Time Diffusion Features for Zero-Shot Text-Driven Motion Transfer (CVPR 2024 · similarity 0.88)
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.
