Research map / Video / Video generation
Soul: Breathe Life into Digital Human for High-fidelity Long-term Multimodal Animation
Jiangning Zhang, Junwei Zhu, Zhenye Gan, Donghao Luo, Chuming Lin, FeiFan Xu, Xu Peng, Jianlong Hu, Yuansen Liu, Yijia Hong, Weijian Cao, Han Feng and 5 more
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 |
|---|---|---|---|
| train on Soul-1M and use Soul design instead of prior open-source avatar methods | Soul · Soul-Bench | Identity Consistence: 0.750 -> 0.763 | Improved |
| use Soul instead of Wan-S2V | Soul · Soul-Bench | LSE-D: 5.455 -> 0.130 | Improved |
| use Soul instead of InfiniteTalk | Soul · Soul-Bench | LSE-C: 8.48 -> 6.82 | Hurt |
| use Soul instead of StableAvatar | Soul · Soul-Bench | Video Quality: 71.40 -> 72.60 | Improved |
| add step and CFG distillation (ablation) | Soul · Soul-Bench | Full-model latency (s): 1019.2s -> 135.3s | Improved |
| add step and CFG distillation (ablation) | Soul · Soul-Bench | Identity Consistence: 0.763 -> 0.696 | Hurt |
| add step and CFG distillation (ablation) | Soul · Soul-Bench | Video Quality: 72.60 -> 71.90 | Hurt |
| add efficient eVAE decoder after distillation (ablation) | Soul · Soul-Bench | Full-model latency (s): 135.3s -> 89.4s | Improved |
| add efficient eVAE decoder after distillation (ablation) | Soul · Soul-Bench | Identity Consistence: 0.696 -> 0.702 | No clear effect |
| replace Wan2.2-5B VAE decoder with eVAE-35M (ablation) | Wan2.2-5B VAE · 720x1280 reconstruction | LPIPS: 0.0324 -> 0.052 | Hurt |
| replace Wan2.2-5B VAE decoder with eVAE-35M (ablation) | Wan2.2-5B VAE · 720x1280 reconstruction | Decoder speedup: 1.0x -> 8.1x | Improved |
| add threshold-aware codebook replacement (ablation) | Soul long-term generation · long video generation, qualitative | visual quality over time | 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.
- Sonic — Sonic: Shifting Focus to Global Audio Perception in Portrait Animation (CVPR 2025) · better on 3 of 4 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
- KeyFace: Expressive Audio-Driven Facial Animation for Long Sequences via KeyFrame Interpolation (CVPR 2025 · similarity 0.88)
- MVPortrait: Text-Guided Motion and Emotion Control for Multi-view Vivid Portrait Animation (CVPR 2025 · similarity 0.88)
- VLOGGER: Multimodal Diffusion for Embodied Avatar Synthesis (CVPR 2025 · similarity 0.87)
- Teller: Real-Time Streaming Audio-Driven Portrait Animation with Autoregressive Motion Generation (CVPR 2025 · similarity 0.87)
- OmniSync: Towards Universal Lip Synchronization via Diffusion Transformers (NeurIPS 2025 · similarity 0.87)
- Hallo2: Long-Duration and High-Resolution Audio-Driven Portrait Image Animation (ICLR 2025 · 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.
