HunyuanVideo-Foley
TencentHunyuanVideo-Foley is Tencent's model for generating sound effects and ambient audio that match a video, guided by an optional text prompt. It was trained on a curated 100,000-hour text, video and audio collection and produces 48 kHz audio synchronized to what happens on screen. Tencent published it in August 2025, adding a smaller XL model that runs in 8 GB of video memory in September.
Released under the Tencent Hunyuan Community License, which does not apply in the European Union, the United Kingdom or South Korea and needs a separate license above 100 million monthly active users. It loads Google's SigLIP 2 and LAION's CLAP encoders, both Apache 2.0. HunyuanVideo, Tencent's video generation line, is a separate entry.
Openness
3 high confidence- weights
- open (the XXL and XL checkpoints, with the audio VAE and Synchformer weights, download from the Hub without a gate)
- data
- described (the paper and README describe the pipeline that curated a 100,000-hour text-video-audio set; the data is not released)
- code
- partial (inference code and a Gradio app; no training code)
- license
- Tencent-Hunyuan-Community-License (the weights and code; excludes the European Union, the United Kingdom and South Korea, and needs a separate license above 100 million monthly active users)
HunyuanVideo-Foley's weights and inference code are published under Tencent's community license, which allows commercial use but does not apply in the European Union, the United Kingdom or South Korea. No training code is published, and the training data is described but not released.
- https://arxiv.org/abs/2508.16930 recorded 2026-11-12
Paper abstract: "a scalable data pipeline curating 100k-hour multimodal datasets through automated annotation".
- https://huggingface.co/api/models/google/siglip2-base-patch16-512?expand[]=cardData&expand[]=gated recorded 2026-11-12
Hugging Face API record for google/siglip2-base-patch16-512: license apache-2.0, gated false.
- https://huggingface.co/api/models/laion/larger_clap_general?expand[]=cardData&expand[]=gated recorded 2026-11-12
Hugging Face API record for laion/larger_clap_general: license apache-2.0, gated false.
- https://huggingface.co/api/models/tencent/HunyuanVideo-Foley?expand[]=downloads&expand[]=cardData&expand[]=gated&expand[]=siblings&expand[]=createdAt recorded 2026-11-12
Hugging Face API record for tencent/HunyuanVideo-Foley: license_name tencent-hunyuan-community, gated false; hunyuanvideo_foley.pth, hunyuanvideo_foley_xl.pth, vae_128d_48k.pth and synchformer_state_dict.pth.
- https://raw.githubusercontent.com/Tencent-Hunyuan/HunyuanVideo-Foley/main/hunyuanvideo_foley/utils/model_utils.py recorded 2026-11-12
Model loader: weights from the repository plus "google/siglip2-base-patch16-512" and "laion/larger_clap_general" from the Hub.
- https://raw.githubusercontent.com/Tencent-Hunyuan/HunyuanVideo-Foley/main/LICENSE recorded 2026-11-12
LICENSE body: "TENCENT HUNYUAN COMMUNITY LICENSE AGREEMENT / Tencent HunyuanVideo-Foley Release Date: August 28, 2025"; "DOES NOT APPLY IN THE EUROPEAN UNION, UNITED KINGDOM AND SOUTH KOREA"; a separate license above "100 million monthly active users".
- https://raw.githubusercontent.com/Tencent-Hunyuan/HunyuanVideo-Foley/main/README.md recorded 2026-11-12
README news: "[2025.9.29] HunyuanVideo-Foley-XL Model Release"; "[2025.8.28] HunyuanVideo-Foley Open Source Release - Inference code and model weights publicly available"; a "data processing pipeline for high-quality text-video-audio datasets".
- https://ungh.cc/repos/Tencent-Hunyuan/HunyuanVideo-Foley/files/main recorded 2026-11-12
Repository file list: infer.py, gradio_app.py, the hunyuanvideo_foley/ package and tests; no training script.
Adoption
1 high confidenceAdoption is measured as Hugging Face downloads of the single HunyuanVideo-Foley repository, which holds .pth checkpoints the Hub's counter may undercount.
- https://huggingface.co/api/models?author=tencent&search=Foley&sort=downloads&direction=-1&limit=100 recorded 2026-11-12
Listing: tencent/HunyuanVideo-Foley, 1121 downloads, the only Foley repository on the account.
Capability
1 medium confidenceHunyuanVideo-Foley does the same job as MMAudio, sound for silent video, and like it has no public arena to rank on.
- https://artificialanalysis.ai/music/leaderboard/instrumental recorded 2026-11-12
The "AA-Music-Instrumental v1.1" board lists 19 music models and no HunyuanVideo-Foley entry.
- https://arxiv.org/abs/2508.16930 recorded 2026-11-12
Paper abstract: "an end-to-end text-video-to-audio framework that synthesizes high-fidelity audio precisely aligned with visual dynamics and semantic context".
Verified 2026-11-12