#mgaudio
MGAudio, a flow-based video-to-audio framework, uses model-guided dual-role alignment; outperforms classifier-free guidance baselines on VGGSound; generalizes to UnAV-100.
Model-Guided Dual-Role Alignment for High-Fidelity Open-Domain Video-to-Audio Generation
Kang Zhang, Trung X. Pham, Suyeon Lee, Axi Niu, Arda Senocak, Joon Son Chung
arxiv.org
October 29, 2025 at 8:35 AM