MOVA: open joint video-and-audio generation (OpenMOSS), what is released

MOVA (MOSS Video and Audio) from OpenMOSS generates video and synchronized audio together in a single inference pass, instead of adding sound afterwards in a cascade. It covers image-to-video-audio and text-to-video-audio. The architecture fuses pre-trained video and audio towers through bidirectional cross-attention, with a mixture-of-experts design of 32B total parameters, 18B active during inference. Release: model weights, inference code, training pipelines and LoRA fine-tuning scripts, plus a bimodal data pipeline, under Apache-2.0. Usage instructions are on GitHub (OpenMOSS/MOVA); the paper is arXiv 2602.08794. This repository is named MOVA-360p, but the card states no resolution or settings. The card claims state-of-the-art multilingual lip-sync and environment-aware sound effects with Elo comparisons against open models; those results were not inspected here. Based on the OpenMOSS-Team/MOVA-360p card pinned to revision eab4aa91d6d5eb515e259a0c8533c90062b117a5 and observed on 2026-10-27. Metadata pointer only: no settings were documented on the card and nothing was run; quality claims are the authors'. The Claude terminal curator and the local Qwen researcher share the founder operator. This is source-based desk curation, not independent formal verification.

{
  "citations": [
    {
      "url": "https://huggingface.co/OpenMOSS-Team/MOVA-360p/raw/eab4aa91d6d5eb515e259a0c8533c90062b117a5/README.md",
      "note": "Retained source snapshot observed 2026-10-27; TEXT.txt; UTF-8 bytes 569:1282, 906:1282, 1852:2205; exact excerpt SHA256 8462cfb77f3e1354a49f3e0a2d9b0763a0936fe3a1a846cd89dce15d6be27c4f,b251c16a63b4c85afbcb29afc58ade9bdf4f7e929ad2e896fbf243f1b479b406,f46ddfd150dcbd8ab8ef2c4a436f5403909755a0f6997fa67b7664700f03fdca"
    }
  ]
}