DiffRhythm
ASLP LabDiffRhythm is a song generation model line from ASLP Lab at Northwestern Polytechnical University. It writes a complete song, vocals and accompaniment together, from lyrics and a style prompt or reference clip. DiffRhythm 2, released with Xiaomi in October 2025, generates semi-autoregressively in blocks to keep the singing aligned with the lyrics.
The first DiffRhythm (March 2025) and its 1.2 update remain on the Hub with training code. DiffRhythm 2 was developed with Xiaomi and ships inference code only; it loads the separately published MuQ-MuLan style encoder.
Openness
2 medium confidence- weights
- open (the DiffRhythm 2 checkpoint and decoder download from the Hub without a gate)
- data
- closed (no training songs are released)
- code
- partial (DiffRhythm 2 publishes inference code only; training code was published for the first DiffRhythm)
- license
- Apache-2.0 (the DiffRhythm 2 code and weights)+CC-BY-NC-4.0 (the MuQ-MuLan-large style encoder the inference script downloads and loads on every run)
DiffRhythm 2's own code and weights are Apache 2.0, but generating a song requires the MuQ-MuLan style encoder, which its authors license for non-commercial use only, so the model as run cannot be used commercially. Only inference code is published for this release, and the training data is not released.
- https://arxiv.org/abs/2510.22950 recorded 2026-11-12
DiffRhythm 2 paper abstract: a semi-autoregressive architecture based on block flow matching, a 5 Hz music VAE and cross-pair preference optimization; no training dataset is released.
- https://huggingface.co/api/models/ASLP-lab/DiffRhythm2?expand[]=downloads&expand[]=cardData&expand[]=gated&expand[]=siblings&expand[]=createdAt recorded 2026-11-12
Hugging Face API record for ASLP-lab/DiffRhythm2: license apache-2.0, gated false; model.safetensors and decoder.bin.
- https://huggingface.co/OpenMuQ/MuQ-MuLan-large/raw/main/README.md recorded 2026-11-12
MuQ-MuLan card: "license: cc-by-nc-4.0"; "The model weights (MuQ-large-msd-iter, MuQ-MuLan-large) are released under the CC-BY-NC 4.0 license."
- https://raw.githubusercontent.com/ASLP-lab/DiffRhythm2/main/inference.py recorded 2026-11-12
inference.py loads the style encoder on every run: "mulan = MuQMuLan.from_pretrained("OpenMuQ/MuQ-MuLan-large", ...)", then downloads decoder.bin from the DiffRhythm 2 repository.
- https://raw.githubusercontent.com/ASLP-lab/DiffRhythm2/main/LICENSE recorded 2026-11-12
LICENSE body: "Copyright 2025 ASLP Lab and Xiaomi Inc."; Apache License, Version 2.0.
- https://raw.githubusercontent.com/ASLP-lab/DiffRhythm2/main/README.md recorded 2026-11-12
README: "2025.10.30 We released the DiffRhythm2 paper, demo code, and model weights"; "DiffRhythm 2 (code and weights) is released under the Apache License 2.0"; inference instructions only.
- https://ungh.cc/repos/ASLP-lab/DiffRhythm2/files/main recorded 2026-11-12
DiffRhythm2 file list: inference.py, inference.sh, the diffrhythm2/ model code, a bigvgan/ decoder and g2p/; no training script.
Adoption
1 high confidenceAdoption is measured as Hugging Face downloads summed across ASLP Lab's DiffRhythm checkpoints, most of them DiffRhythm 2.
- https://huggingface.co/api/models?author=ASLP-lab&search=DiffRhythm&sort=downloads&direction=-1&limit=100 recorded 2026-11-12
Listing sorted by downloads: DiffRhythm2 1294, DiffRhythm-1_2 318, DiffRhythm-base 156, DiffRhythm-1_2-full 131, DiffRhythm-full 129, DiffRhythm-vae 0; 2,028 in total.
Capability
1 medium confidenceDiffRhythm 2 generates full songs with vocals but has no placement on the public music arenas, where MusicGen at least appears.
- https://artificialanalysis.ai/music/leaderboard/instrumental recorded 2026-11-12
The "AA-Music-Instrumental v1.1" board lists 19 models, from "Suno v6" first to "MusicGen" 19th, and no DiffRhythm entry.
- https://artificialanalysis.ai/music/leaderboard/vocals recorded 2026-11-12
The "AA-Music-Vocal v1.1" board lists 14 models, from "ACE Music v1" first to "Melodia v3" 14th, and no DiffRhythm entry.
- https://raw.githubusercontent.com/ASLP-lab/DiffRhythm2/main/README.md recorded 2026-11-12
README: DiffRhythm 2 "is capable of generating full-length songs with precise lyric alignment and coherent musical structures".
Verified 2026-11-12