Research map / Video / Video generation
JavisDiT++: Unified Modeling and Optimization for Joint Audio-Video Generation
Kai Liu, Yanhao Zheng, Kai Wang, Shengqiong Wu, Rongjunchen Zhang, Jiebo Luo, Dimitrios Hatzinakos, Ziwei Liu, Hao Fei, Tat-Seng Chua
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 modality-specific MoE FFNs instead of shared DiT with LoRA (ablation) | JavisDiT++ · Architecture ablation | JavisScore: 0.098 -> 0.153 | Improved |
| use modality-specific MoE FFNs instead of shared DiT with full fine-tuning (ablation) | JavisDiT++ · Architecture ablation | FVD: 269.3 -> 221.3 | Improved |
| use modality-specific MoE FFNs instead of shared DiT with full fine-tuning (ablation) | JavisDiT++ · Architecture ablation | FAD: 5.66 -> 5.51 | Improved |
| add AV-DPO with modality-aware micro reward (ablation) | JavisDiT++ · AV-DPO reward design | FVD: 221.3 -> 198.5 | Improved |
| add AV-DPO with modality-aware micro reward (ablation) | JavisDiT++ · AV-DPO reward design | DeSync: 0.807 -> 0.776 | Improved |
| use average-micro (modality-agnostic) reward in DPO instead of none (ablation) | JavisDiT++ · AV-DPO reward design | DeSync: 0.807 -> 0.810 | No clear effect |
| add reward normalization in AV-DPO pair ranking (ablation) | JavisDiT++ · AV-DPO reward design | FVD: 210.0 -> 198.5 | Improved |
| include ground-truth samples in AV-DPO pairs (ablation) | JavisDiT++ · AV-DPO reward design | FVD: 234.7 -> 198.5 | Improved |
| use interleave+offset TA-RoPE instead of vanilla position IDs (ablation) | JavisDiT++ · Positional encoding ablation | DeSync: 0.918 -> 0.807 | Improved |
| use interleave+offset TA-RoPE instead of vanilla position IDs (ablation) | JavisDiT++ · Positional encoding ablation | FAD (AudioCaps): 6.03 -> 4.65 | Improved |
| use interpolation instead of vanilla positional encoding (ablation) | JavisDiT++ · Positional encoding ablation | FAD (AudioCaps): 6.03 -> 12.87 | Hurt |
| train on 330K medium-quality data instead of 120K high-quality data (ablation) | JavisDiT++ · AV-SFT data composition | FVD: 239.9 -> 221.3 | Improved |
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.
- JavisDiT — JavisDiT: Joint Audio-Video Diffusion Transformer with Hierarchical Spatio-Temporal Prior Synchronization (ICLR 2026) · 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
- VinTAGe: Joint Video and Text Conditioning for Holistic Audio Generation (CVPR 2025 · similarity 0.90)
- Diverse and Aligned Audio-to-Video Generation via Text-to-Video Model Adaptation (AAAI 2024 · similarity 0.90)
- MMAudio: Taming Multimodal Joint Training for High-Quality Video-to-Audio Synthesis (CVPR 2025 · similarity 0.89)
- JavisGPT: A Unified Multi-modal LLM for Sounding-Video Comprehension and Generation (NeurIPS 2025 · similarity 0.89)
- Native Audio-Visual Alignment for Generation (NeurIPS 2026 · similarity 0.89)
- Read, Watch and Scream! Sound Generation from Text and Video (AAAI 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.
