Research map / Video / Video generation
Talking Head Generation with Probabilistic Audio-to-Visual Diffusion Priors
Zhentao Yu, Zixin Yin, Deyu Zhou, Duomin Wang, Finn Wong, Baoyuan 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 diffusion prior instead of auto-regressive prior for non-lip motion (ablation) | audio-driven talking head · VoxCeleb2 | SND: 7.65 -> 4.94 | Improved |
| use diffusion prior instead of auto-regressive prior (ablation) | audio-driven talking head · VoxCeleb2 | FID: 24.31 -> 14.58 | Improved |
| use diffusion prior instead of auto-regressive prior (ablation) | audio-driven talking head · VoxCeleb2 user study | Naturalness MOS: 2.53 -> 3.82 | Improved |
| add velocity loss to diffusion prior (ablation) | diffusion prior · VoxCeleb2 | SND: 5.29 -> 4.94 | Improved |
| add mask editing during training (ablation) | diffusion prior · VoxCeleb2 | SND: 8.76 -> 4.94 | Improved |
| add orthogonal loss between non-lip and audio-lip features (ablation) | lip/non-lip disentanglement · VoxCeleb2 | SyncNet confidence Sc: 2.84 -> 7.48 | Improved |
| add memory banks for orthogonal loss (ablation) | lip/non-lip disentanglement · VoxCeleb2 | pose distance Pd: 0.0293 -> 0.0019 | Improved |
| diffusion prior versus PC-AVS in image quality (FID) | audio-driven talking head · VoxCeleb2 | FID: 14.36 (PC-AVS) vs 14.58 | No clear effect |
| diffusion prior versus Wav2Lip on SyncNet confidence | audio-driven talking head · VoxCeleb2 | SyncNet confidence Sc: 9.23 (Wav2Lip) vs 7.57 | Hurt |
Closest papers by meaning
- FaceTalk: Audio-Driven Motion Diffusion for Neural Parametric Head Models (CVPR 2024 · similarity 0.91)
- Long-Term TalkingFace Generation via Motion-Prior Conditional Diffusion Model (ICML 2025 · similarity 0.91)
- Diffusion-based Realistic Listening Head Generation via Hybrid Motion Modeling (CVPR 2025 · similarity 0.91)
- MoDiTalker: Motion-Disentangled Diffusion Model for High-Fidelity Talking Head Generation (AAAI 2025 · similarity 0.91)
- DAWN: Dynamic Frame Avatar with Non-autoregressive Diffusion Framework for Talking head Video Generation (ICLR 2025 · similarity 0.91)
- EmotiveTalk: Expressive Talking Head Generation through Audio Information Decoupling and Emotional Video Diffusion (CVPR 2025 · similarity 0.90)
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.
