Meta research model that generates audio directly in raw waveform space, without a VAE, for video-to-audio synthesis.
Updated EU status checked Last shown
Sources: creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
WavFlow is a research audio generation model from Meta (FAIR authors) that synthesizes audio directly in raw waveform space rather than in a compressed latent space. It uses waveform patchifying and amplitude lifting to enable stable flow matching with direct x-prediction, removing the need for a VAE or other intermediate representation.
The project page reports results on VGGSound-Test for video-text-to-audio generation and on MovieGen-Audio-Bench. WavFlow variants (M-16kHz at 624M parameters, L-16kHz, L-44.1kHz) are compared against MMAudio, HunyuanVideo-Foley, Frieren, V2A-Mapper and MovieGen Audio. According to the page, it is competitive with MMAudio on several metrics, with training on about 11.1K hours of data versus about 8.2K for MMAudio.
The material provided shows a project page, demo samples and a paper citation (arXiv 2605.18749), but no code repository, model card or download link. A YouTube reviewer noted that the full production checkpoints were not released and that a piano example sounded poor. Access and licensing are therefore unconfirmed.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Video-to-audio model that generates sound directly in raw waveform space without a VAE or latent compression. Competitive with MMAudio, but the full production checkpoints were not released; piano example was poor.
Closest neighbours by meaning, not just by category.
Model · ByteDance
Research zero-shot TTS model that generates speech by modeling raw waveforms directly with flow-matching DiT.
Model · Meta
Meta and HKUST research method for fast artist-like 3D mesh generation using a MeshVAE and flow-based diffusion transformer.
Model · Baidu
Baidu ERNIE team's 6.3B joint audio-video generation model with native audio-visual alignment and timbre-controlled speech.
Model · Xiaomi
Xiaomi's LLM-driven model that generates mixed audio scenes (speech, music, SFX, ambience) from structured text captions.
Model
Text-to-human-motion flow-matching model that generates full-resolution motion directly, with training-free joint constraints.
Model · Meta
Meta speech-to-text model described as suited to real-time streaming transcription.
