Reproduce “MAVIN: Multi-Shot Audio-Visual Generation with Customized Narrative Control” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kaiqi Liu, Yunyao Mao, Ziqi Cai, Zheng Geng, Jing Wang, Qiulin Wang, Xintao Wang, Pengfei Wan and 3 more
MAVIN turns ideas into timed scripts for multi-shot audio-visual generation. Its boundary-aware routing limits which shot and role instructions guide video and dialogue at each moment, while optional image and voice references anchor identity. On a manually verified 1,000-sample benchmark, it scores 0.9897 on Shot Transition Accuracy, versus 0.4992 for OVI, without role references; this measures cut control, not film quality.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.