ACE-Step-1.5
lastmod 2026-02-19created 2026-02-193 edits
The most powerful local music generation model that outperforms most commercial alternatives, supporting Mac, AMD, Intel, and CUDA devices.
A local music generation model that outperforms paid services
About
- Open-source music generation model that achieves commercial-grade music generation quality like Suno even on ordinary consumer hardware
- Fast generation of a full song in under 10 seconds (RTX 3090)
- Can run locally even with 4GB VRAM or less, and supports LoRA-based personalized training to reflect the user's musical style
- Supports cover generation, repainting (partial regeneration), vocal→BGM conversion, track separation, and multitrack synthesis
- Structure and style can be controlled with lyric prompts in 50+ languages
- Performance and quality characteristics
- Delivers audio quality and style diversity at or above commercial models (Suno v4.5–v5 level)
- Supports 1000+ instruments and genres, with fine-grained timbre control
- Supports generating audio up to 10 minutes (600 seconds) long and batch generation of 8 songs at once
- One-click annotation and training in the Gradio UI
- Training on 8 songs completes within 1 hour on an RTX 3090 (12GB)
- Supports advanced LoRA/LoKR fine-tuning and VRAM optimization through the Side-Step module
- Gradio Web UI: intuitive web-based music generation interface
- Studio UI: DAW-like advanced editing environment
- Python API / REST API / CLI: supports service integration and automation
Documentations
- https://arxiv.org/abs/2602.00744
- [2602.00744] ACE-Step 1.5 - Pushing the Boundaries of Open-Source Music Generation