Research map / Video / Video generation
P-Flow: Prompting Visual Effects Generation
Rui Zhao, Mike Zheng Shou
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 noise prior enhancement to P-Flow (ablation) | P-Flow (Wan 2.1 14B) · Open-VFX overall | FVD: 1205.47 -> 1072.10 | Improved |
| add visual context of previous generated video (ablation) | P-Flow (Wan 2.1 14B) · Open-VFX overall | FVD: 1072.10 -> 953.10 | Improved |
| add logic context (full history of prompts and analyses) (ablation) | P-Flow (Wan 2.1 14B) · Open-VFX overall | FVD: 953.10 -> 882.63 | Improved |
| add logic context (ablation) | P-Flow (Wan 2.1 14B) · Open-VFX overall | Dynamic Degree: 0.81 -> 0.91 | Improved |
| add SVD projection of inverted noise (ablation) | P-Flow noise prior · Open-VFX image-to-video | FVD: 1052.80 -> 784.51 | Improved |
| blend inverted motion noise with random noise (alpha 0.001 vs random only) (ablation) | P-Flow noise prior · Open-VFX image-to-video | Dynamic Degree: 0.73 -> 0.94 | Improved |
| increase blending alpha from 0.001 to 0.01 (ablation) | P-Flow noise prior · Open-VFX image-to-video | FID-VID / Dynamic Degree: FID-VID 29.32 -> 29.21; Dyn. Degree 0.94 -> 0.88 | Mixed |
| prompt optimization alone without any module vs Wan 2.1 (ablation) | P-Flow · Open-VFX overall | Dynamic Degree: 0.31 -> 0.63 | Improved |
| replace one-shot human prompt revision with VLM test-time prompt optimization | Wan 2.1 (T2V) · Open-VFX text-to-video | FVD: 1632.10 -> 980.75 | Improved |
| human feedback prompt revision on baseline | Wan 2.1 · Open-VFX overall | FID-VID: 38.47 -> 75.33 | Hurt |
| P-Flow training-free vs training-based VFX Creator | Open-VFX image-to-video · Open-VFX image-to-video | Dynamic Degree: 0.63 -> 0.94 | Improved |
| P-Flow vs VFX Creator on FVD | Open-VFX image-to-video · Open-VFX image-to-video | FVD: 752.95 -> 784.51 | Hurt |
Closest papers by meaning
- Motion Prompting: Controlling Video Generation with Motion Trajectories (CVPR 2025 · similarity 0.88)
- Contextual Flow Matching: Adaptive Step Selection in Flow Models for Efficient Visual Generation (NeurIPS 2026 · similarity 0.87)
- Optical-Flow Guided Prompt Optimization for Coherent Video Generation (CVPR 2025 · similarity 0.87)
- Make Pixels Dance: High-Dynamic Video Generation (CVPR 2024 · similarity 0.86)
- EffectMaker: Unifying Reasoning and Generation for Customized Visual Effect Creation (CVPR 2026 · similarity 0.86)
- STOP: Integrated Spatial-Temporal Dynamic Prompting for Video Understanding (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.
