Research map / Vision-language and multimodal / Cross-modal learning
EchoMimicV3: 1.3B Parameters Are All You Need for Unified Multi-Modal and Multi-Task Human Animation
Rang Meng, Yan Wang, Weipeng Wu, Ruobing Zheng, Yuming Li, Chenguang Ma
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 hard-to-easy task schedule instead of easy-to-hard (ablation) | EchoMimicV3 · Ablation (Table 2) | Sync-C: 5.40 -> 5.49 | Improved |
| use hard-to-easy schedule instead of uniform random sampling (ablation) | EchoMimicV3 · Ablation (Table 2) | Sync-C: 4.98 -> 5.49 | Improved |
| add inter-task EMA (ablation) | EchoMimicV3 · Ablation (Table 2) | FVD: 508.82 -> 496.76 | Improved |
| allocate audio in early timestep phase (ablation) | EchoMimicV3 · Ablation (Table 2) | Sync-C: 4.76 -> 5.49 | Improved |
| include text in early/mid phases (ablation) | EchoMimicV3 · Ablation (Table 2) | HA: 0.81 -> 0.95 | Improved |
| include image in early phases (ablation) | EchoMimicV3 · Ablation (Table 2) | FVD: 540.80 -> 496.76 | Improved |
| use Negative DPO-SFT cycle instead of SFT only (ablation) | EchoMimicV3 · Ablation (Table 2) | FVD: 540.30 -> 496.76 | Improved |
| use Negative DPO instead of paired DPO (ablation) | EchoMimicV3 · Ablation (Table 2) | ID: 0.93 -> 1.00 | Improved |
| use paired DPO instead of Negative DPO (ablation) | EchoMimicV3 · Ablation (Table 2) | FVD: 480.98 vs 496.76 | Mixed |
| add Phase-aware Negative CFG (ablation) | EchoMimicV3 · Ablation (Table 2) | HA: 0.89 -> 0.95 | Improved |
| add Long Video CFG (ablation) | EchoMimicV3 · Ablation (Table 2) | FVD: 530.21 -> 496.76 | Improved |
| use 1.3B model with 5 steps instead of 14B models | EchoMimicV3 · Table 1 | inference speed: 18x speedup | Improved |
Closest papers by meaning
- EchoMimicV2: Towards Striking, Simplified, and Semi-Body Human Animation (CVPR 2025 · similarity 0.89)
- EchoMimic: Lifelike Audio-Driven Portrait Animations through Editable Landmark Conditions (AAAI 2025 · similarity 0.86)
- MegActor-Sigma: Unlocking Flexible Mixed-Modal Control in Portrait Animation with Diffusion Transformer (AAAI 2025 · similarity 0.86)
- HumanVid: Demystifying Training Data for Camera-controllable Human Image Animation (NeurIPS 2024 · similarity 0.85)
- Let Them Talk: Audio-Driven Multi-Person Conversational Video Generation (NeurIPS 2025 · similarity 0.85)
- DynaFLIP: Rethinking Robotics Perception via Tri-Modal-Dynamics Guided Representation (NeurIPS 2026 · similarity 0.85)
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.
