Research map / Video / Video generation
Syncphony: Synchronized Audio-to-Video Generation with Diffusion Transformers
Jibin Song, Mingi Kwon, Jaeseok Jeong, Youngjung Uh
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 Motion-aware Loss (ablation) | Syncphony · AVSync15 | CycleSync: 15.18 -> 16.48 | Improved |
| add Motion-aware Loss (ablation) | Syncphony · AVSync15 | FVD: 305.9 -> 293.1 | Improved |
| add Audio Sync Guidance with w=2 (ablation) | Syncphony · AVSync15 | CycleSync: 15.31 -> 16.48 | Improved |
| add Audio Sync Guidance with w=2 (ablation) | Syncphony · AVSync15 | FID: 8.5 -> 8.5 | No clear effect |
| increase ASG scale from 2 to 4 (ablation) | Syncphony · AVSync15 | FVD: 293.1 -> 298.3 | Hurt |
| increase ASG scale from 2 to 4 (ablation) | Syncphony · AVSync15 | CycleSync: 16.48 -> 16.26 | No clear effect |
| add Audio RoPE to audio cross-attention (ablation) | Syncphony · AVSync15 | CycleSync: 14.41 -> 15.31 | Improved |
| add audio cross-attention layers to the off-sync backbone (ablation) | Syncphony · AVSync15 | CycleSync: 12.34 -> 15.31 | Improved |
| use pretrained video backbone with audio cross-attention instead of AVSyncD's image backbone | Syncphony · TheGreatestHits | CycleSync: 9.89 -> 16.18 | Improved |
| increase ASG scale from 0 to 4 (ablation) | Syncphony · TheGreatestHits | CycleSync: 11.70 -> 17.71 | Improved |
| evaluate with AV-Align instead of CycleSync | synchronization metrics · AVSync15 | metric ranking of fine-tuned audio-free Pyramid Flow: AV-Align 20.69 (highest) vs CycleSync 12.34 (lowest) | Hurt |
Closest papers by meaning
- Audio-Sync Video Generation with Multi-Stream Temporal Control (NeurIPS 2025 · similarity 0.90)
- MMAudio: Taming Multimodal Joint Training for High-Quality Video-to-Audio Synthesis (CVPR 2025 · similarity 0.89)
- DiVAS: Video and Audio Synchronization with Dynamic Frame Rates (CVPR 2024 · similarity 0.88)
- Native Audio-Visual Alignment for Generation (NeurIPS 2026 · similarity 0.88)
- Seeing and Hearing: Open-domain Visual-Audio Generation with Diffusion Latent Aligners (CVPR 2024 · similarity 0.88)
- Animate and Sound an Image (CVPR 2025 · 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.
