Research map / Architectures and efficiency / Pruning and efficient tuning
MagicPrompt: Ultra-Lightweight Prompt Tuning for Video Generation
Yinhan Zhang, DINGWEI TAN, Xianghao Kong, Yue Ma, Yeying Jin, Anyi Rao
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 soft prompts (ablation) | MagicPrompt · I2V, Wan2.1-1.3B | FID: 30.94 -> 29.23 | Improved |
| add soft prompts (ablation) | MagicPrompt · I2V, Wan2.1-1.3B | FVD: 284.16 -> 274.88 | Improved |
| add shift bias (ablation) | MagicPrompt · I2V, Wan2.1-1.3B | FID: 36.23 -> 29.23 | Improved |
| add shift bias (ablation) | MagicPrompt · I2V, Wan2.1-1.3B | FVD: 365.64 -> 274.88 | Improved |
| use 64 soft tokens instead of 4 (ablation) | MagicPrompt · I2V, Wan2.1-1.3B | FVD: 347.43 -> 274.88 | Improved |
| use 64 soft tokens instead of 8 (ablation) | MagicPrompt · I2V, Wan2.1-1.3B | FID: 34.73 -> 29.23 | Improved |
| use 64 soft tokens instead of 32 (ablation) | MagicPrompt · I2V, Wan2.1-1.3B | FVD: 353.72 -> 274.88 | Improved |
| add dual-space reward instead of no reward (ablation) | MagicPrompt · condition-guided I2V | FVD: 274.88 -> 141.44 | Improved |
| add pixel-space reward to latent-only reward (ablation) | MagicPrompt · condition-guided I2V | FID: 21.46 -> 17.92 | Improved |
| add latent-space reward to pixel-only reward (ablation) | MagicPrompt · condition-guided I2V | FVD: 142.06 -> 141.44 | No clear effect |
| replace LoRA with MagicPrompt on T2V | Wan2.1-1.3B · T2V | FVD: 853.56 -> 637.46 | Improved |
| replace LoRA with MagicPrompt on I2V | Wan2.1-1.3B · I2V | FVD: 269.02 -> 274.88 | Hurt |
Closest papers by meaning
- Diffusion-DRF: Free, Rich, and Differentiable Reward for Video Diffusion Fine-Tuning (NeurIPS 2026 · similarity 0.90)
- Optical-Flow Guided Prompt Optimization for Coherent Video Generation (CVPR 2025 · similarity 0.86)
- DiTCtrl: Exploring Attention Control in Multi-Modal Diffusion Transformer for Tuning-Free Multi-Prompt Longer Video Generation (CVPR 2025 · similarity 0.86)
- Improving Video Generation with Human Feedback (NeurIPS 2025 · similarity 0.86)
- Revisiting the Power of Prompt for Visual Tuning (ICML 2024 · similarity 0.86)
- Align-A-Video: Deterministic Reward Tuning of Image Diffusion Models for Consistent Video Editing (CVPR 2025 · 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.
