Directory
Published directory entries matching your search.
MMAudio is a CVPR 2025 project for video-to-audio synthesis that generates synchronized audio from video and/or text inputs. It uses multimodal joint training and a synchronization module to align generated audio with video frames. A pretrained large_44k_v2 model is used by default, with inference around 6GB of GPU memory in 16-bit mode.
MMAudio is a research project on multimodal joint training for video-to-audio synthesis. It generates synchronized audio given video and/or text inputs. The work is presented at CVPR 2025 by authors from the University of Illinois Urbana-Champaign and Sony.
MMAudio is a Hugging Face Space that generates synchronized audio from video or text. Users can upload a short video and write a prompt, including a negative prompt, to add realistic sound, or create a standalone audio clip from a text description.
