Research map / Video / Video generation
MVI2V: Human Centric Image to Video Generation with Multiview Consistent Appearance
Pengfei Liu, Mingyi Xu, Wentao Jiang, Tiezheng Ge, Ming Zeng
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 MVI2V architecture (dedicated reference stream) (ablation) | Wan2.1 · Multi-view benchmark | GPT garment: 0.409 -> 0.470 | Improved |
| add reference frame selection (ablation) | MVI2V-Wan2.1 · Multi-view benchmark | GPT garment: 0.470 -> 0.562 | Improved |
| add reference frame selection (ablation) | MVI2V-Wan2.1 · Multi-view benchmark | GPT person: 0.784 -> 0.853 | Improved |
| add inpainting subtask (ablation) | MVI2V-Wan2.1 · Multi-view benchmark | GPT garment: 0.562 -> 0.585 | Improved |
| add inpainting subtask (ablation) | MVI2V-Wan2.1 · Multi-view benchmark | DINO person: 0.743 -> 0.728 | Hurt |
| add inpainting subtask (ablation) | MVI2V-Wan2.1 · Multi-view benchmark | DINO garment: 0.638 -> 0.656 | Improved |
| use dedicated reference stream instead of Token-Concat | Wan2.1 · Multi-view benchmark | GPT garment: 0.512 -> 0.585 | Improved |
| use dedicated reference stream instead of Concat-ID | Wan2.1 · Multi-view benchmark | GPT person: 0.841 -> 0.862 | Improved |
| add Token-Concat multi-view references | Wan2.1 · Multi-view benchmark | Subject Consistency: 0.916 -> 0.902 | Hurt |
| add time projection layer for reference tokens (Concat-ID vs Token-Concat) | Token-Concat · Multi-view benchmark | Subject Consistency: 0.902 -> 0.918 | Improved |
| add MVI2V triple-stream extension | In-House · Multi-view benchmark | GPT person: 0.692 -> 0.868 | Improved |
| add MVI2V triple-stream extension | In-House · Multi-view benchmark | DINO garment: 0.601 -> 0.633 | Improved |
Closest papers by meaning
- FashionChameleon: Towards Real-Time and Interactive Human-Garment Video Customization (NeurIPS 2026 · similarity 0.87)
- MV-VTON: Multi-View Virtual Try-On with Diffusion Models (AAAI 2025 · similarity 0.86)
- MVHumanNet: A Large-scale Dataset of Multi-view Daily Dressing Human Captures (CVPR 2024 · similarity 0.86)
- Pursuing Temporal-Consistent Video Virtual Try-On via Dynamic Pose Interaction (CVPR 2025 · similarity 0.85)
- MEAT: Multiview Diffusion Model for Human Generation on Megapixels with Mesh Attention (CVPR 2025 · similarity 0.85)
- Robust-MVTON: Learning Cross-Pose Feature Alignment and Fusion for Robust Multi-View Virtual Try-On (CVPR 2025 · 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.
