AI Music
DiffRhythm
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 starsApache-2.0
View guide →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
ACE-Step is an open-source ai music project by ace-step (ace-step/ACE-Step on GitHub). It has 4,958 GitHub stars and 643 forks, is released under the Apache-2.0 license, and was last updated on GitHub on 15 February 2026.
Per the README installation section: clone the repo, create the environment it describes, and pip install -e . (Windows users also pip install triton-windows).
Start the app with the acestep command — the README-recommended consumer options are acestep --torch_compile true --cpu_offload true --overlapped_decode true — or use the Gradio UI.
Describe the style (tags, descriptive text or a use-case scenario), add lyrics if you want vocals, and generate — up to 4-minute tracks, then refine with repaint, variations or lyric editing.
Diffusion + DCAE + linear transformer design: ~4 minutes of music in ~20 seconds on an A100, without the structural drift of slower LLM-style generators.
Sung vocals with lyric alignment in 19 languages, plus dedicated lyric2vocal and singing2accompaniment tasks.
Voice cloning, lyric editing, remixing, repainting and variation generation are built into the model design, not bolted on.
Optimisations brought max VRAM to ~8GB with CPU offload; a ComfyUI node and Gradio UI cover different workflows.
The README installation route.
git clone https://github.com/ace-step/ACE-Step.git
cd ACE-Step
pip install -e .The README alternative (needs Git on your PATH; use a virtualenv).
pip install git+https://github.com/ace-step/ACE-Step.gitThe launch options the project recommends for memory-friendly speed.
acestep --torch_compile true --cpu_offload true --overlapped_decode trueThe ACE-Step project (ace-step on GitHub), with StepFun among the organisations behind it; the technical report is arXiv 2506.00045.
Yes — Apache-2.0 licensed. As with all music models, check the terms of any training data or voices you bring to it before commercial release.
This is the foundation model and engine itself. ace-step-ui and ACE-Step Studio are interfaces built on top of it — LibreAI listed those first; this entry is the missing engine underneath them.
Against DiffRhythm (also added this week), ACE-Step emphasises speed, lyric alignment and editing controls (repaint, lyric edit, cloning) as a foundation model; DiffRhythm emphasises dead-simple full-song latent diffusion. Against Suno, ACE-Step is the open, self-hosted, Apache-2.0 option you can fine-tune with LoRAs.
Clone, pip install -e ., then run acestep with the README consumer launch options and generate your first track in the UI.
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
Portable AI music generator — full songs with vocals, covers, music videos. One-click install, 100% offline, NVIDIA GPU.
🎵 The Ultimate Open Source Suno Alternative - Professional UI for ACE-Step 1.5 AI Music Generation. Free, local, unlimited. Stop paying for Suno!
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, …
Run the ACE-Step 1.5 music generator locally in C++.
Local AI music studio: ACE-Step plus YuE2, stems, MIDI, and a built-in DAW.
The most-starred other AI Music projects in the LibreAI catalogue — the realistic shortlist if ACE-Step is not the right fit.
YuE2: frontier music generation with symbolic planning, zero-shot covers, and agentic music editing.
🎵 The Ultimate Open Source Suno Alternative - Professional UI for ACE-Step 1.5 AI Music Generation. Free, local, unlimited. Stop paying for Suno!
Official DiffSinger code: diffusion-based singing voice synthesis and text-to-speech (AAAI 2022)
Use API to call the music generation AI of suno.ai, and easily integrate it into agents like GPTs.
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
Free and open — grab the source, star the repo, and start creating.