AI Music
ACE-Step
ACE-Step — the open foundation model for music generation: full songs with vocals and lyrics in 19 languages, up to 4 minutes in about 20 seconds, Apache-2.0
★ 5.0k starsApache-2.0
View guide →// ai-music · ASLP-lab/DiffRhythm
DiffRhythm
DiffRhythm is an open-source ai music project by ASLP-lab (ASLP-lab/DiffRhythm on GitHub). It has 2,350 GitHub stars and 279 forks, is released under the Apache-2.0 license, and was last updated on GitHub on 27 November 2025.
DiffRhythm — blazingly fast, embarrassingly simple full-length song generation with latent diffusion: complete songs with vocals from lyrics and a style prompt, Apache-2.0
★ 2.4k stars⑂ 279 forksApache-2.0updated 2025-11
▦ visual — this project's README ships no embeddable screenshot,
so here's a placeholder. The guide below is complete regardless.
so here's a placeholder. The guide below is complete regardless.
// project facts
Project facts
- Repository
- ASLP-lab/DiffRhythm
- GitHub stars
- 2,350
- Forks
- 279
- License
- Apache-2.0
- Last updated on GitHub
- 27 November 2025
- Category
- AI Music
- Listing
- Full install guide
// how it works
How it works
- Clone and install system dependency
Clone the repo and install espeak-ng (apt/brew, or the .msi on Windows) — inference depends on it, per the README.
- Create the environment
conda create -n diffrhythm python=3.10 (or a plain venv), activate it, then pip install -r requirements.txt. On Windows, set the eSpeak NG path in your environment variables first, per the README.
- Generate a song
Run the inference script with your lyrics and either a reference WAV file or a text style prompt; or use the Docker compose route in the docker folder.
// features
Why people love it
Full-length songs, end to end
Generates complete songs with vocals in one latent-diffusion model instead of stitched sections.
Lyrics + style in
Condition on lyrics and a style described in text, or steer the sound with a reference audio file.
Simple by design
One repo, one requirements file, one inference script — the project pitch is that song generation does not need a pipeline empire.
A growing family
DiffRhythm+ and DiffRhythm 2 extend the line, with papers, demos and a Hugging Face Space to try in the browser.
Advertisement
// run it
Run it yourself
- Clone
Get the code first.
git clone https://github.com/ASLP-lab/DiffRhythm.git cd DiffRhythm - Environment
The README inference setup: espeak-ng, then conda + requirements.
sudo apt-get install espeak-ng conda create -n diffrhythm python=3.10 conda activate diffrhythm pip install -r requirements.txt - Docker alternative
The README docker route: edit the compose bindings in the docker folder, compose up, then exec into the container.
docker compose up -d docker exec -it DiffRhythm bash
// use cases
What you can build
- Generating complete demo songs from finished lyrics
- Style-exploration: the same lyrics rendered in several prompted styles
- Reference-guided generation that borrows the feel of a reference track
- Research into latent diffusion for long-form music
Advertisement
// faq
Questions, answered
Who makes DiffRhythm?
ASLP-lab (the Audio, Speech and Language Processing lab team behind the repo) with named authors on the papers — ASLP-lab/DiffRhythm on GitHub.
Is DiffRhythm free?
Yes — Apache-2.0 licensed. Check rights for any reference audio or lyrics you feed it before releasing output commercially.
What does it need to run?
A Python 3.10 environment, the requirements file, espeak-ng installed, and a GPU for practical generation speeds — or use the Hugging Face Space demo to try it with none of that.
How does it compare to a well-known alternative?
Against ACE-Step (also added this week), DiffRhythm is the simpler, single-purpose song generator — lyrics and style in, full song out — while ACE-Step is the broader foundation model with editing, cloning and repaint controls. Against Suno, DiffRhythm is open, self-hosted and Apache-2.0.
How do I get started?
Clone, install espeak-ng, conda create -n diffrhythm python=3.10, pip install -r requirements.txt, then run inference with your lyrics and a style prompt.
// keep exploring
Similar tools
AI Music
YuE2-Studio
Local AI song generator with an editable score — YuE2 on your GPU: full songs with vocals, sheet music, covers, exact replay. Native Windows app, no Python, …
★ 424 starsMIT
View guide →AI Music
ACE-Step-Studio
Portable AI music generator — full songs with vocals, covers, music videos. One-click install, 100% offline, NVIDIA GPU.
★ 374 starsMIT
View guide →AI Music
suno-cli
Terminal client for Suno v5.5 — tags, vocal control, and full generation flags.
★ 56 starsMIT
View guide →AI Music
ComfyUI_FL-HeartMuLa
ComfyUI nodes for multilingual AI songs with lyrics via HeartMuLa.
★ 133 starsCustom
View guide →AI Music
melodyc
Open-source AI music SaaS: full tracks with cover art from a text prompt.
★ 49 starsMIT
View guide →// alternatives
Alternatives in AI Music
The most-starred other AI Music projects in the LibreAI catalogue — the realistic shortlist if DiffRhythm is not the right fit.
- YuE★ 10.8k
YuE2: frontier music generation with symbolic planning, zero-shot covers, and agentic music editing.
- ace-step-ui★ 5.0k
🎵 The Ultimate Open Source Suno Alternative - Professional UI for ACE-Step 1.5 AI Music Generation. Free, local, unlimited. Stop paying for Suno!
- ACE-Step★ 5.0k
ACE-Step — the open foundation model for music generation: full songs with vocals and lyrics in 19 languages, up to 4 minutes in about 20 seconds, Apache-2.0
- DiffSinger★ 4.9k
Official DiffSinger code: diffusion-based singing voice synthesis and text-to-speech (AAAI 2022)
- Stable Audio Tools★ 3.9k
Stable Audio Tools — training and inference code for Stability's audio generation models, including Stable Audio Open, MIT
// more in ai-music
More in AI Music
Nearby AI Music projects in the catalogue, closest in stars to DiffRhythm — keep exploring the shelf either side of it.
- suno-api★ 3.2k
Use API to call the music generation AI of suno.ai, and easily integrate it into agents like GPTs.
- MSST-WebUI★ 1.3k
A WebUI app for Music-Source-Separation-Training and we bundled UVR together!
- Stable Audio Tools★ 3.9k
Stable Audio Tools — training and inference code for Stability's audio generation models, including Stable Audio Open, MIT
- BS-RoFormer★ 965
A working implementation of Band Split Roformer
- DiffSinger★ 4.9k
Official DiffSinger code: diffusion-based singing voice synthesis and text-to-speech (AAAI 2022)
- audio-ai-hub★ 960
The hub for audio AI research: papers, open models, benchmarks & datasets across audio LLMs, speech recognition, TTS, music & audio generation.
Get DiffRhythm
Free and open — grab the source, star the repo, and start creating.