Research map / Video / Video generation
Learning Zero-Shot Subject-Driven Video Generation Using 1% Compute
Daneul Kim, Jingxu Zhang, Wonjoon Jin, Sunghyun Cho, Qi Dai, Jaesik Park, Chong Luo
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 arbitrary-video motion task with stochastic switching to image-only training (ablation) | CogVideoX-5B · Ablation training strategy (120 videos) | Dynamic Degree: 0.84 -> 69.64 | Improved |
| add arbitrary-video motion task with stochastic switching to image-only training (ablation) | CogVideoX-5B · Ablation training strategy (120 videos) | Motion Smoothness: 99.60 -> 98.45 | Hurt |
| add arbitrary-video motion task with stochastic switching to image-only training (ablation) | CogVideoX-5B · Ablation training strategy (120 videos) | DINO-I: 43.19 -> 62.88 | Improved |
| use stochastic task switching instead of sequential two-stage fine-tuning (ablation) | CogVideoX-5B · Ablation training strategy (120 videos) | Motion Smoothness: 96.04 -> 98.45 | Improved |
| use stochastic task switching instead of sequential two-stage fine-tuning (ablation) | CogVideoX-5B · Ablation training strategy (120 videos) | CLIP-T: 28.96 -> 32.69 | Improved |
| use stochastic task switching instead of sequential two-stage fine-tuning (ablation) | CogVideoX-5B · Ablation training strategy (120 videos) | DINO-I: 76.13 -> 62.88 | Hurt |
| scale subject-image pairs from 2,000 to 200K (ablation) | CogVideoX-5B · Ablation pair scale (120 videos) | DINO-I: 57.03 -> 62.88 | Improved |
| scale subject-image pairs from 4,000 to 200K (ablation) | CogVideoX-5B · Ablation pair scale (120 videos) | Copy score (lower better): 41.09 -> 36.32 | Improved |
| scale subject-image pairs from 4,000 to 200K (ablation) | CogVideoX-5B · Ablation pair scale (120 videos) | Flow entropy: 58.91 -> 63.77 | Improved |
| scale subject-image pairs from 4,000 to 200K (ablation) | CogVideoX-5B · Ablation pair scale (120 videos) | Dynamic Degree: 78.33 -> 69.64 | Hurt |
| scale subject-image pairs from 2,000 to 4,000 (ablation) | CogVideoX-5B · Ablation pair scale (120 videos) | Dynamic Degree: 60.71 -> 78.33 | Improved |
| add Gradient Surgery with buffer 8 to stochastic task switching (ablation) | CogVideoX-5B · Gradient surgery comparison | Dynamic Degree: 69.64 -> 50.83 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VideoBooth — VideoBooth: Diffusion-based Video Generation with Image Prompts (CVPR 2024) · 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
- IM-Zero: Instance-level Motion Controllable Video Generation in a Zero-shot Manner (CVPR 2025 · similarity 0.87)
- TI2V-Zero: Zero-Shot Image Conditioning for Text-to-Video Diffusion Models (CVPR 2024 · similarity 0.87)
- SG-I2V: Self-Guided Trajectory Control in Image-to-Video Generation (ICLR 2025 · similarity 0.87)
- OpenS2V-Nexus: A Detailed Benchmark and Million-Scale Dataset for Subject-to-Video Generation (NeurIPS 2025 · similarity 0.86)
- CustomCrafter: Customized Video Generation with Preserving Motion and Concept Composition Abilities (AAAI 2025 · similarity 0.86)
- MotionCraft: Physics-Based Zero-Shot Video Generation (NeurIPS 2024 · similarity 0.86)
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.
