Research map / Video / Video generation
M4V: Multimodal Mamba for Efficient Text-to-Video Generation
Jiancheng Huang, Gengwei Zhang, Zequn Jie, Siyu Jiao, Yinlong Qian, Ling Chen, Yunchao Wei, Lin 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 |
|---|---|---|---|
| add text token re-composition (ablation) | M4V MM-DiM block · VBench subset (fast eval) | Overall-Cons: 19.77 -> 21.23 | Mixed |
| add per-frame visual registers (ablation) | M4V MM-DiM block · VBench subset (fast eval) | average score: 55.70 -> 56.79 | Improved |
| add temporal branch to text+register model (ablation) | M4V MM-DiM block · VBench subset (fast eval) | average score: 57.10 -> 58.75 | Improved |
| replace attention with Mamba in all blocks (Full) instead of full attention (ablation) | PyramidFlow · VBench subset (fast eval, block design) | average score: 59.84 -> 57.10 (TFLOPs 55.44 -> 26.64) | Mixed |
| add parallel Mamba and attention in each block (ablation) | full attention model · VBench subset (fast eval, block design) | average score: 59.84 -> 59.97 | No clear effect |
| add reward learning post-training (ablation) | M4V (PyramidFlow) · VBench | Total Score: 81.55 -> 81.71 | Improved |
| add HunyuanVideo-generated training videos (ablation) | M4V (PyramidFlow) · VBench | Total Score: 81.55 -> 81.59 | No clear effect |
| combine reward learning with generated data (ablation) | M4V (PyramidFlow) · VBench | Semantic Score: 74.47 -> 76.10 | Improved |
| replace Wan2.1 self-attention with MM-DiM blocks | Wan2.1 · VBench | Total Score: 84.70 -> 86.14 | Improved |
| replace PyramidFlow attention with MM-DiM blocks | PyramidFlow · VBench | Total Score: 81.61 -> 81.55 | No clear effect |
Closest papers by meaning
- LinGen: Towards High-Resolution Minute-Length Text-to-Video Generation with Linear Computational Complexity (CVPR 2025 · similarity 0.88)
- OpenVid-1M: A Large-Scale High-Quality Dataset for Text-to-video Generation (ICLR 2025 · similarity 0.87)
- Vivid-ZOO: Multi-View Video Generation with Diffusion Model (NeurIPS 2024 · similarity 0.87)
- MicroCinema: A Divide-and-Conquer Approach for Text-to-Video Generation (CVPR 2024 · similarity 0.87)
- T2V-CompBench: A Comprehensive Benchmark for Compositional Text-to-video Generation (CVPR 2025 · similarity 0.87)
- LC-Mamba: Local and Continuous Mamba with Shifted Windows for Frame Interpolation (CVPR 2025 · similarity 0.86)
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.
