ACE-Step-1.5

<h1 align="center">ACE-Step 1.5</h1>
<h1 align="center">Pushing the Boundaries of Open-Source Music Generation</h1>
<p align="center">
    <a href="https://acemusic.ai">ACEMusic</a> |
    <a href="https://ace-step.github.io/ace-step-v1.5.github.io/">Project</a> |
    <a href="https://huggingface.co/ACE-Step/Ace-Step1.5">Hugging Face</a> |
    <a href="https://modelscope.cn/models/ACE-Step/Ace-Step1.5">ModelScope</a> |
    <a href="https://huggingface.co/spaces/ACE-Step/Ace-Step-v1.5">Space Demo</a> |
    <a href="https://discord.gg/PeWDxrkdj7">Discord</a> |
    <a href="https://arxiv.org/abs/2602.00744">Technical Report</a> |
    <a href="https://github.com/ace-step/awesome-ace-step">Awesome ACE-Step</a>
</p>

<p align="center">
    <img src="./assets/organization_logos.png" height="80" alt="StepFun Logo" style="vertical-align: middle;">
    &nbsp;&nbsp;
    <a href="https://acemusic.ai">
        <img src="./assets/acemusic-logo.svg" height="57" alt="ACEMusic - Try ACE-Step Online" style="vertical-align: middle; position: relative; top: 2px;">
    </a>
</p>

## 📰 News

> 🎵 **Want a faster & more stable experience? Try [acemusic.ai](https://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](https://huggingface.co/ACE-Step/acestep-v15-xl-base), [xl-sft](https://huggingface.co/ACE-Step/acestep-v15-xl-sft), [xl-turbo](https://huggingface.co/ACE-Step/acestep-v15-xl-turbo). Requires ≥12GB VRAM (with offload), ≥20GB recommended. All LM models fully compatible. See [Model Zoo](#-model-zoo) for details.

## Table of Contents

- [📰 News](#-news)
- [✨ Features](#-features)
- [⚡ Quick Start](#-quick-start)
- [🚀 Launch Scripts](#-launch-scripts)
- [📚 Documentation](#-documentation)
- [📖 Tutorial](#-tutorial)
- [🏗️ Architecture](#️-architecture)
- [🦁 Model Zoo](#-model-zoo)
- [🔬 Benchmark](#-benchmark)

## 📝 Abstract
🚀 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, there