ACEMusic | Project | Hugging Face | ModelScope | Space Demo | Discord | Technical Report | Awesome ACE-Step

🎵 Want a faster & more stable experience? Try acemusic.ai — 100% free!

• [2026-04-02] 🎉 ACE-Step 1.5 XL (4B DiT) Released! — We introduce the XL series with a 4B-parameter DiT decoder for higher audio quality. Three variants available: xl-base, xl-sft, xl-turbo. Requires ≥12GB VRAM (with offload), ≥20GB recommended. All LM models fully compatible. See Model Zoo for details.

• 📰 News
• ✨ Features
• ⚡ Quick Start
• 🚀 Launch Scripts
• 📚 Documentation
• 📖 Tutorial
• 🏗️ Architecture
• 🦁 Model Zoo
• 🔬 Benchmark

🚀 We present ACE-Step v1.5, a highly efficient open-source music foundation model that brings commercial-grade generation to consumer hardware. On commonly used evaluation metrics, ACE-Step v1.5 achieves quality beyond most commercial music models while remaining extremely fast—under 2 seconds per full song on an A100 and under 10 seconds on an RTX 3090. The model runs locally with less than 4GB of VRAM, and supports lightweight personalization: users can train a LoRA from just a few songs to capture their own style.

🌉 At its core lies a novel hybrid architecture where the Language Model (LM) functions as an omni-capable planner: it transforms simple user queries into comprehensive song blueprints—scaling from short loops to 10-minute compositions—while synthesizing metadata, lyrics, and captions via Chain-of-Thought to guide the Diffusion Transformer (DiT). ⚡ Uniquely, this alignment is achieved through intrinsic reinforcement learning relying solely on the model's internal mechanisms, thereby eliminating the biases inherent in external reward models or human preferences. 🎚️

🔮 Beyond standard synthesis, ACE-Step v1.5 unifies precise stylistic control with versatile editing capabilities—such as cover generation, repainting, and vocal-to-BGM conversion—while maintaining strict adherence to prompts across 50+ languages. This paves the way for powerful tools that seamlessly integrate into the creative workflows of music artists, producers, and content creators. 🎸

• ✅ Ultra-Fast Generation — Under 2s per full song on A100, under 10s on RTX 3090 (0.5s to 10s on A100 depending on think mode & diffusion steps)
• ✅ Flexible Duration — Supports 10 seconds to 10 minutes (600s) audio generation
• ✅ Batch Generation — Generate up to 8 songs simultaneously

• ✅ Commercial-Grade Output — Quality beyond most commercial music models (between Suno v4.5 and Suno v5)
• ✅ Rich Style Support — 1000+ instruments and styles with fine-grained timbre description
• ✅ Multi-Language Lyrics — Supports 50+ languages with lyrics prompt for structure & style control

Star ACE-Step on GitHub and be instantly notified of new releases

🎵 Don't want to install locally? Try acemusic.ai — 100% free, no GPU required!

Requirements: Python 3.11-3.12, CUDA GPU recommended (also supports MPS / ROCm / Intel XPU / CPU)

Note: ROCm on Windows requires Python 3.12 (AMD officially provides Python 3.12 wheels only)

# 1. Install uv
curl -LsSf https://astral.sh/uv/install.sh | sh          # macOS / Linux
# powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"  # Windows

# 2. Clone & install
git clone https://github.com/ACE-Step/ACE-Step-1.5.git
cd ACE-Step-1.5
uv sync

# 3. Launch Gradio UI (models auto-download on first run)
uv run acestep

# Or launch REST API server
uv run acestep-api

Open http://localhost:7860 (Gradio) or http://localhost:8001 (API).

📦 Windows users: A portable package with pre-installed dependencies is available. See Installation Guide.

📦 MacOS users: A portable package with pre-installed dependencies is available. See Installation Guide.

📖 Full installation guide (AMD/ROCm, Intel GPU, CPU, environment variables, command-line options): English | 中文 | 日本語

XL (4B) models (acestep-v15-xl-*) offer higher audio quality with ~9GB VRAM for weights (vs ~4.7GB for 2B). They require ≥12GB VRAM (with offload + quantization) or ≥20GB (without offload). All LM models are fully compatible with XL.

The UI automatically selects the best configuration for your GPU. All settings (LM model, backend, offloading, quantization) are tier-aware and pre-configured.

📖 GPU compatibility details: English | 中文 | 日本語 | 한국어

Ready-to-use launch scripts for all platforms with auto environment detection, update checking, and dependency installation.

# Windows
start_gradio_ui.bat

# Linux
chmod +x start_gradio_ui.sh && ./start_gradio_ui.sh

# macOS (Apple Silicon)
chmod +x start_gradio_ui_macos.sh && ./start_gradio_ui_macos.sh

Recommended: Create a .env file to customize models, ports, and other settings. Your .env configuration will survive repository updates.

# Copy the example file
cp .env.example .env

# Edit with your preferred settings
# Examples in .env:
ACESTEP_CONFIG_PATH=acestep-v15-turbo
ACESTEP_LM_MODEL_PATH=acestep-5Hz-lm-1.7B
PORT=7860
LANGUAGE=en

📖 Script configuration & customization: English | 中文 | 日本語

🎯 Must Read: Comprehensive guide to ACE-Step 1.5's design philosophy and usage methods.

This tutorial covers: mental models and design philosophy, model architecture and selection, input control (text and audio), inference hyperparameters, random factors and optimization strategies.

📖 LoRA Training Tutorial — step-by-step guide covering data preparation, annotation, preprocessing, and training:

See also the LoRA Training tab in Gradio UI for one-click training, or Gradio Guide - LoRA Training for UI reference.

🔧 Advanced Training with Side-Step — CLI-based training toolkit with corrected timestep sampling, LoKR adapters, VRAM optimization, gradient sensitivity analysis, and more. See the Side-Step documentation.

XL models use a larger 4B-parameter DiT decoder (~9GB bf16) for higher audio quality. They require ≥12GB VRAM (with offload + quantization) or ≥20GB (without offload). All LM models are fully compatible.

ACE-Step 1.5 includes profile_inference.py, a profiling & benchmarking tool that measures LLM, DiT, and VAE timing across devices and configurations.

python profile_inference.py                        # Single-run profile
python profile_inference.py --mode benchmark       # Configuration matrix

📖 Full guide (all modes, CLI options, output interpretation): English | 中文

This project is licensed under MIT

ACE-Step enables original music generation across diverse genres, with applications in creative production, education, and entertainment. While designed to support positive and artistic use cases, we acknowledge potential risks such as unintentional copyright infringement due to stylistic similarity, inappropriate blending of cultural elements, and misuse for generating harmful content. To ensure responsible use, we encourage users to verify the originality of generated works, clearly disclose AI involvement, and obtain appropriate permissions when adapting protected styles or materials. By using ACE-Step, you agree to uphold these principles and respect artistic integrity, cultural diversity, and legal compliance. The authors are not responsible for any misuse of the model, including but not limited to copyright violations, cultural insensitivity, or the generation of harmful content.

🔔 Important Notice
The only official website for the ACE-Step project is our GitHub Pages site.
We do not operate any other websites.
🚫 Fake domains include but are not limited to: ac**p.com, a**p.org, a***c.org
⚠️ Please be cautious. Do not visit, trust, or make payments on any of those sites.

Check out Awesome ACE-Step — a curated list of community projects, alternative UIs, ComfyUI nodes, cloud deployments, training tools, and more built around ACE-Step.

This project is co-led by ACE Studio and StepFun.

If you find this project useful for your research, please consider citing:

@misc{gong2026acestep,
	title={ACE-Step 1.5: Pushing the Boundaries of Open-Source Music Generation},
	author={Junmin Gong, Yulin Song, Wenxiao Zhao, Sen Wang, Shengyuan Xu, Jing Guo}, 
	howpublished={\url{https://github.com/ace-step/ACE-Step-1.5}},
	year={2026},
	note={GitHub repository}
}