Research map / Video / Video generation
VAFlow: Video-to-Audio Generation with Cross-Modality Flow Matching
Xihua Wang, Xin Cheng, Yuyue Wang, Ruihua Song, Yunfeng Wang
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 |
|---|---|---|---|
| use video latent prior from alignment VAE without joint tuning instead of Gaussian prior (ablation) | VAFlow-CLIP · VGGSound test (source distribution ablation) | FD (PaSST): 102.6 -> 111.8 | Hurt |
| add joint tuning of alignment VAE encoder with the estimator (ablation) | VAFlow-CLIP with video prior · VGGSound test (source distribution ablation) | FD (PaSST): 111.8 -> 87.7 | Improved |
| jointly tuned video prior instead of Gaussian prior (ablation) | VAFlow-CLIP · VGGSound test (source distribution ablation) | IS (mel): 67.6 -> 73.6 | Improved |
| add cross-attention layers with explicit video conditioning (ablation) | VAFlow-CLIP (no CFG) · VGGSound test (conditioning ablation) | sync accuracy: 31.7 -> 78.0 | Improved |
| add cross-attention layers with explicit video conditioning (ablation) | VAFlow-CLIP (no CFG) · VGGSound test (conditioning ablation) | IB-VA: 8.5 -> 21.0 | Improved |
| zero out video input to cross-attention at inference (ablation) | VAFlow-CLIP · VGGSound test (conditioning ablation) | sync accuracy: 78.0 -> 18.9 | Hurt |
| add classifier-free guidance (ablation) | VAFlow-CLIP with cross-attention · VGGSound test (conditioning ablation) | FD (PaSST): 133.8 -> 87.7 | Improved |
| add classifier-free guidance (ablation) | VAFlow-CLIP with cross-attention · VGGSound test (conditioning ablation) | sync accuracy: 78.0 -> 88.1 | Improved |
| use CAVP features instead of CLIP features (ablation) | VAFlow · VGGSound test | sync accuracy: 88.1 -> 96.3 | Improved |
| use CAVP features instead of CLIP features (ablation) | VAFlow · VGGSound test | FD (PaSST): 87.7 -> 91.8 | Hurt |
| use VidTok reconstruction features instead of CLIP (ablation) | VAFlow · VGGSound test | FD (PaSST): 87.7 -> 128.5 | Hurt |
| use VidTok reconstruction features instead of CLIP (ablation) | VAFlow · VGGSound test | IB-VA: 28.6 -> 12.9 | Hurt |
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.
- Frieren — Frieren: Efficient Video-to-Audio Generation Network with Rectified Flow Matching (NeurIPS 2024) · better on 2 of 3 shared results
- Diff-Foley — Diff-Foley: Synchronized Video-to-Audio Synthesis with Latent Diffusion Models (NeurIPS 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
- Frieren: Efficient Video-to-Audio Generation Network with Rectified Flow Matching (NeurIPS 2024 · similarity 0.89)
- Foley-Flow: Coordinated Video-to-Audio Generation with Masked Audio-Visual Alignment and Dynamic Conditional Flows (CVPR 2025 · similarity 0.89)
- Model-Guided Dual-Role Alignment for High-Fidelity Open-Domain Video-to-Audio Generation (NeurIPS 2025 · similarity 0.88)
- V2A-Mapper: A Lightweight Solution for Vision-to-Audio Generation by Connecting Foundation Models (AAAI 2024 · similarity 0.87)
- From Vision to Audio and Beyond: A Unified Model for Audio-Visual Representation and Generation (ICML 2024 · similarity 0.87)
- Seeing and Hearing: Open-domain Visual-Audio Generation with Diffusion Latent Aligners (CVPR 2024 · similarity 0.87)
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.
