Reproduce “Wan-Alpha++: Animating a Single RGBA Asset into Controllable Transparent Video Effects” 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
Haotian Dong, Xin Wang, Wenjing Wang, Chen Li, Jing Lyu, Yan Tina Luximon, Ping Li, Wanchao Su and 2 more
Wan-Alpha++ generates asset-conditioned RGBA videos that emerge from and return to transparency. It cyclically shifts transparent-visible-transparent training clips so visible content reaches both conditioned boundaries, guides a selected interval toward transparency during sampling, then reverses the shift. On the tested Wan2.1-FLF2V-plus-MatAnyone baseline, masked DINO-I rose from 0.6679 to 0.9227; guidance adds inference computation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.