ACE-Step 1.5
App : A product you use directly
Full music generation (vocals, structure, style) running locally, open source.
The best local music generator in 2026: a full song in a few seconds on a consumer GPU.
Security reflects where data goes, requested permissions and incidents on record at the verification date. It is not a guarantee of zero risk: check the deployment mode and the connections you enable.
What it does, concretely
Creating complete music with vocals, instruments and coherent structure usually requires expensive professional software or online services that retain your data. ACE-Step 1.5 addresses this by offering an open-source music generator that runs entirely on your own machine.
The model combines a diffusion transformer with a language model that acts as a planner: it turns a simple text description into a complete song with lyrics, metadata and musical structure. Generation takes under 10 seconds on an RTX 3090 GPU and under 2 seconds on a professional A100. The model runs with as little as 4 GB of VRAM.
ACE-Step 1.5 is aimed at musicians, producers and content creators who want to generate original tracks, create accompaniments for an existing vocal recording, edit a specific section of a song, or train a personalised style from their own songs using LoRA.
However, the tool is not suited to users without a dedicated graphics card (a CPU mode exists but is significantly slower), nor to those looking for a ready-to-use interface without any installation. Quality can vary by genre, with some styles such as Chinese rap being less well handled, and results remain sensitive to the choice of random seed.
What it offers
- ·Full song generation from a text description, with lyrics, structure and style, in under 10 seconds on a consumer GPU.
- ·Support for over 50 languages for lyrics, with fine-grained control over BPM, key, time signature and duration (from 10 seconds to 10 minutes).
- ·Vocal-to-accompaniment conversion (Vocal2BGM): the model automatically generates an instrumental track suited to an existing vocal recording.
- ·Local repaint editing: only a specific section of a track can be regenerated without altering the rest.
- ·One-click personalised LoRA training from around 8 songs, taking about one hour on an RTX 3090 with 12 GB of VRAM, to capture a particular musical style.
- ·Audio stem separation and multi-track generation to add instrumental layers.
- ·XL variant (4 billion parameters) available for higher audio quality, requiring at least 12 GB of VRAM.
In practice
- Price
- Free
- Commercial use
- Commercial use: unclear
- Available on
- windows, linux, macos
- Licence
- voir dépôt
- Runs locally
- Yes, 4 GB VRAM recommended
- FR interface
- No
Security and data
- Rating
- Security: nothing to flag
- Your data
- Data stays local
- Known incidents
- Licence des poids à vérifier pour un usage commercial.
- Maintenance
- Active
- Last release
- 18 May 2026
Links and repository checked on 25 October 2026 · added on 1 May 2025
How to install ACE-Step 1.5
git clone https://github.com/ACE-Step/ACE-Step-1.5.git && cd ACE-Step-1.5 && uv sync && uv run acestep