Docker container for running ACE-Step 1.5 Gradio UI - A highly efficient open-source music foundation model that brings commercial-grade music generation to consumer hardware.
ACE-Step v1.5 is a cutting-edge music foundation model that achieves commercial-grade quality while remaining extremely fast and accessible:
- Ultra-Fast Generation: Under 2 seconds per full song on A100, under 10 seconds on RTX 3090
- Commercial-Grade Quality: Performance beyond most commercial music models (between Suno v4.5 and Suno v5)
- Lightweight: Runs locally with less than 4GB of VRAM
- Flexible Duration: Generate music from 10 seconds to 10 minutes (600s)
- Multi-Language Support: Lyrics support for 50+ languages
- Rich Style Support: 1000+ instruments and styles with fine-grained timbre description
- Advanced Capabilities: Cover generation, audio editing, track separation, vocal-to-BGM conversion, LoRA training, and more
At its core lies a novel hybrid architecture where the Language Model (LM) functions as an omni-capable planner that transforms simple user queries into comprehensive song blueprints. The model synthesizes metadata, lyrics, and captions via Chain-of-Thought to guide the Diffusion Transformer (DiT), achieving alignment through intrinsic reinforcement learning.
This section provides step-by-step examples to help you get started with ACE-Step 1.5, whether you have GPU support or not.
- Docker installed on your system (Get Docker)
- At least 4GB of available RAM
- For GPU support: NVIDIA GPU with CUDA support and NVIDIA Container Toolkit installed
If you don't have a GPU or want to run on CPU only, follow these steps:
docker pull ghcr.io/dotnetautor/ace-step-1.5-docker:latestdocker run -p 7860:7860 ghcr.io/dotnetautor/ace-step-1.5-docker:latestNote: CPU-only mode will be slower than GPU-accelerated mode. Music generation may take several minutes per song.
Open your web browser and navigate to:
http://localhost:7860
You should see the Gradio web interface where you can start generating music!
If you have an NVIDIA GPU, you can significantly accelerate music generation. The configuration depends on your GPU's VRAM capacity.
First, ensure your system has GPU support:
docker run --rm --gpus all nvidia/cuda:11.8.0-base-ubuntu22.04 nvidia-smiIf this command shows your GPU information, you're ready to proceed!
docker pull ghcr.io/dotnetautor/ace-step-1.5-docker:latestChoose the appropriate configuration based on your GPU's VRAM:
Use DiT-only mode for lower memory usage:
docker run --gpus all -p 7860:7860 ghcr.io/dotnetautor/ace-step-1.5-docker:latest \
uv run acestep \
--server-name 0.0.0.0 \
--port 7860 \
--language en \
--download-source auto \
--init_service true \
--config_path acestep-v15-turbo \
--init_llm falseFeatures:
- Uses
--init_llm falseto disable the language model (DiT-only mode) - Faster generation with lower memory footprint
- Typical generation time: ~10-20 seconds (times vary by song length and complexity)
Use the default turbo configuration with the compact language model:
docker run --gpus all -p 7860:7860 ghcr.io/dotnetautor/ace-step-1.5-docker:latestFeatures:
- Uses the default 1.7B language model (
acestep-5Hz-lm-1.7B) - Balanced performance and quality
- Typical generation time: ~5-15 seconds (times vary by song length and complexity)
Use the larger 4B language model for maximum quality:
docker run --gpus all -p 7860:7860 ghcr.io/dotnetautor/ace-step-1.5-docker:latest \
uv run acestep \
--server-name 0.0.0.0 \
--port 7860 \
--language en \
--download-source auto \
--init_service true \
--config_path acestep-v15-turbo \
--lm_model_path acestep-5Hz-lm-4BFeatures:
- Uses the larger
acestep-5Hz-lm-4Bmodel for best quality - Superior audio comprehension and music planning
- Typical generation time: <2 seconds on A100, ~5-10 seconds on RTX 3090/4090 (times vary by song length and complexity)
Open your web browser and navigate to:
http://localhost:7860
The interface will be the same as the CPU version, but music generation will be much faster!
| GPU VRAM | Configuration | Language Model | Typical Generation Speed* | Best For |
|---|---|---|---|---|
| ≤6GB | --init_llm false |
Disabled (DiT-only) | ~10-20s | Low-end GPUs, minimal VRAM |
| 8-16GB | Default | 1.7B (compact) | ~5-15s | Most users, balanced performance |
| ≥16GB | --lm_model_path acestep-5Hz-lm-4B |
4B (large) | ~2-10s | High-end GPUs, maximum quality |
*Generation times are approximate and vary based on song length, complexity, and prompt requirements.
For more advanced configuration options and detailed documentation, see the ACE-Step 1.5 README.
docker build -t ace-step-1.5-docker .
docker run -p 7860:7860 ace-step-1.5-dockerThis Docker image provides:
- ACE-Step 1.5 music generation system
- Gradio web interface for easy interaction
- All necessary dependencies pre-installed via uv package manager
- Ready-to-use environment with Python 3.11
- Docker installed on your system
- At least 4GB of available RAM (more recommended for better performance)
- GPU support optional but recommended for faster generation
MIT License - see LICENSE file for details
This Docker image packages the ACE-Step 1.5 music generation system created by ACE Studio and StepFun.
- GitHub Repository: https://github.com/ACE-Step/ACE-Step-1.5
- Project Website: https://ace-step.github.io/ace-step-v1.5.github.io/
- Hugging Face: https://huggingface.co/ACE-Step/Ace-Step1.5
- ModelScope: https://modelscope.cn/models/ACE-Step/Ace-Step1.5
- Demo Space: https://huggingface.co/spaces/ACE-Step/Ace-Step-v1.5
- Technical Report: https://arxiv.org/abs/2602.00744
- Discord Community: https://discord.gg/PeWDxrkdj7
If you use ACE-Step 1.5 in your research or project, please cite:
@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}
}ACE-Step 1.5 is co-led by ACE Studio and StepFun. This Docker containerization is an independent packaging effort to make the system more accessible for deployment.