DiffRhythm ASLP-lab/DiffRhythm

Diffusion model that generates full-length songs (up to 4m45s) from timed lyrics and a text or audio style reference.

Creator kit + Compare GitHub ↗ Install guide (official) ↗

DiffRhythm at a glance

Checked 15 Oct 2026

DiffRhythm from ASLP-lab generates songs with vocals from LRC-timed lyrics plus a text style prompt or a reference clip, with an instrumental mode; v1.2 (May 2025) adds song editing and continuation. Base models make 1m35s, full models 4m45s. It needs about 8 GB of VRAM with chunked decoding and espeak-ng installed. Code and DiT weights are Apache-2.0, but the VAE uses the Stability AI Community licence. There is no Gradio UI; you run shell scripts. Last code push was November 2025, and DiffRhythm 2 is announced elsewhere.

Cite this pageRepoQuarry. “DiffRhythm.” Checked 15 Oct 2026. https://repoquarry.dev/projects/diffrhythm

How much of Suno does it replace?

3 of 5 jobs. Songs, instrumentals and continuation from scripts; lyrics must be timed LRC; no covers or stems.

Full songs with vocals from a prompt or lyrics Partial replacement

Instrumental tracks (Instrumental mode) Partial replacement

Covers and restyles of an uploaded song (Cover) Not covered

Extending or regenerating part of a song (Extend Partial replacement

Splitting a song into vocal and instrument stems (Get Stems) Not covered

For covers and restyles of an uploaded song (Cover), splitting a song into vocal and instrument stems (Get Stems), see other Suno alternatives. You may also need to build an interface, an API server and hosting yourself.

Facts, with sources

Every answer links to where we found it. We read the documentation; we did not install it.

Report a correction

QuestionAnswerSourceChecked

Runs on my machine? Yes, inference scripts or Docker on your own machine Following the steps below to clone the repository and install the environment. github.com · DiffRhythm#inference ↗ 1d ago

Code licence Apache-2.0 DiffRhythm (code and DiT weights) is released under the Apache License 2.0. github.com · DiffRhythm#license--disclaimer ↗ 1d ago

Model weights licence DiT weights Apache-2.0; the DiffRhythm VAE is under the Stability AI Community licence huggingface.co · DiffRhythm-vae ↗ 1d ago

Commercial use Free for commercial use under USD 1M annual revenue because of the VAE's Stability AI Community licence; larger firms need a Stability licence For researchers, developers, small businesses, and creators with less than $1M in annual revenue. stability.ai · license ↗ 1d ago

Paid services needed? None required github.com · DiffRhythm#inference ↗ 1d ago

Hardware At least 8 GB VRAM with the --chunked option; more without it Note that DiffRhythm-base requires a minimum of 8G of VRAM. github.com · DiffRhythm#inference ↗ 1d ago

Operating systems Linux, Windows (batch scripts, espeak-ng setup) and macOS DiffRhythm can now run on MacOS! github.com · DiffRhythm#inference ↗ 1d ago

Setup effort Medium; conda or venv plus espeak-ng, or Docker compose You just need the 3 files inside the folder docker. github.com · DiffRhythm#docker-installation ↗ 1d ago