ACE-Step v1-3.5B

AI Model for Music Generation

An open-source 3.5B parameter text-to-music model developed by ACE Studio and StepFun that generates original music across diverse genres with impressive coherence and speed.

Listen to Demos Try Online Demo

Try ACE-Step Online

What is ACE-Step?

ACE-Step is a cutting-edge AI music generation model designed for creating original and high-quality music. Developed by ACE Studio and StepFun and released under the Apache 2.0 license, it delivers impressive musical coherence and generation speed, rivaling existing solutions.

  • High-fidelity music generation with rich harmony and melody
  • Controllable generation via text prompts for genre, mood, and instruments
  • Fast generation: 4-minute music in ~20s on A100
  • Open-source with Apache 2.0 license

Key Features

Superior Music Quality

ACE-Step produces incredibly rich and coherent music with complex melodies, harmonies, and rhythms. The advanced AI models create compositions that are both original and engaging.

Text-to-Music Generation

Easily generate music across various genres by providing text descriptions. Specify mood, instruments, tempo, and more to guide the creative process.

Fast and Efficient

Leveraging an optimized architecture, ACE-Step can generate several minutes of music in seconds on capable hardware, making it ideal for rapid prototyping and creative exploration.

Open Source

Released under the Apache 2.0 license, allowing free use for both personal and commercial purposes. Complete model weights and code are available on GitHub.

SONG SAMPLES

Dark Electro

Dark Electro - music generated by ACE-Step.

Disco

Disco - music generated by ACE-Step.

Electronic Rap Waves On

Electronic Rap Waves On - music generated by ACE-Step.

Hiphop Rap Shirt Song

Hiphop Rap Shirt Song - music generated by ACE-Step.

Opera Female

Opera Female - music generated by ACE-Step.

Dance Party

Dance Party - music generated by ACE-Step.

Minimal Techno

Minimal Techno - music generated by ACE-Step.

Psychedelic

Psychedelic - music generated by ACE-Step.

Saxphone Jazz

Saxphone Jazz - music generated by ACE-Step.

Orchestral Rock

Orchestral Rock - music generated by ACE-Step.

World Sad

World Sad - music generated by ACE-Step.

Acid House

Acid House - music generated by ACE-Step.

Acappella

Acappella - music generated by ACE-Step.

Bbox

Bbox - music generated by ACE-Step.

Drum Bass

Drum Bass - music generated by ACE-Step.

Female Nana

Female Nana - music generated by ACE-Step.

Sonata Piano Violin

Sonata Piano Violin - music generated by ACE-Step.

Tango Guitar

Tango Guitar - music generated by ACE-Step.

Trance

Trance - music generated by ACE-Step.

Violin Solo

Violin Solo - music generated by ACE-Step.

Dark Atmospheric

Dark Atmospheric - music generated by ACE-Step.

Electro House

Electro House - music generated by ACE-Step.

Folk Rnb Female

Folk Rnb Female - music generated by ACE-Step.

French Pop

French Pop - music generated by ACE-Step.

German Dance

German Dance - music generated by ACE-Step.

Hip-House

Hip-House - music generated by ACE-Step.

Italian Folk

Italian Folk - music generated by ACE-Step.

Jpop

Jpop - music generated by ACE-Step.

Kpop

Kpop - music generated by ACE-Step.

Mandopop

Mandopop - music generated by ACE-Step.

Portuguese Pop

Portuguese Pop - music generated by ACE-Step.

Russian Folk

Russian Folk - music generated by ACE-Step.

Spanish Song

Spanish Song - music generated by ACE-Step.

Original Sample

Original music sample by ACE-Step.

Retake Example: Variance1

Music retake variation: Variance1.

Retake Example: Variance2

Music retake variation: Variance2.

Retake Example: Variance3

Music retake variation: Variance3.

Retake Example: Variance4

Music retake variation: Variance4.

Retake Example: Variance5

Music retake variation: Variance5.

Retake Example: Variance6

Music retake variation: Variance6.

Repaint: A Original

Original audio for repaint A example.

Repaint: Female (0 30 Variance10)

Repainted segment (0 30 Variance10) changing to Female.

Repaint: Genre (0 30 Variance10)

Repainted segment (0 30 Variance10) changing to Genre.

Repaint: Lyrics (0 30 Variance10)

Repainted segment (0 30 Variance10) changing to Lyrics.

Repaint: Variance 7 (0-30s)

Repainted segment (0-30s) with variance 7.

Edit Example: A Orig

Audio edit demonstration: A Orig.

Edit Example: Kid

Audio edit demonstration: Kid.

Edit Example: Old

Audio edit demonstration: Old.

Edit Example: Spotify

Audio edit demonstration: Spotify.

Edit Example: Cry

Audio edit demonstration: Cry.

Edit Example: French

Audio edit demonstration: French.

Edit Example: German

Audio edit demonstration: German.

Edit Example: Ja

Audio edit demonstration: Ja.

Edit Example: Ko

Audio edit demonstration: Ko.

Edit Example: Zh

Audio edit demonstration: Zh.

Lyrics to Vocal: In You I See

Vocal generation from lyrics: In You I See.

Lyrics to Vocal: Lemonade

Vocal generation from lyrics: Lemonade.

Lyrics to Vocal: Turn Me On

Vocal generation from lyrics: Turn Me On.

Lyrics to Vocal: Whispering Shadows

Vocal generation from lyrics: Whispering Shadows.

Lyrics to Vocal: You Been Chosen Too

Vocal generation from lyrics: You Been Chosen Too.

Text to Samples: Acounstic Guitar

Generated sample: Acounstic Guitar.

Text to Samples: Bass

Generated sample: Bass.

Text to Samples: Drums

Generated sample: Drums.

Text to Samples: Drums Grooves

Generated sample: Drums Grooves.

Text to Samples: Edrum

Generated sample: Edrum.

Text to Samples: Electric Guitar

Generated sample: Electric Guitar.

Text to Samples: Erhu

Generated sample: Erhu.

Text to Samples: Flute

Generated sample: Flute.

Text to Samples: Guitar

Generated sample: Guitar.

Text to Samples: Hand Pan

Generated sample: Hand Pan.

Text to Samples: Koto

Generated sample: Koto.

Country Rock

Country Rock - music generated by ACE-Step.

Black Metal

Black Metal - music generated by ACE-Step.

Alternative Rock

Alternative Rock - music generated by ACE-Step.

Afro Cuban

Afro Cuban - music generated by ACE-Step.

Dead Rock

Dead Rock - music generated by ACE-Step.

Cyberpunk

Cyberpunk - music generated by ACE-Step.

Dubstep

Dubstep - music generated by ACE-Step.

Surf Music

Surf Music - music generated by ACE-Step.

Female Pop

Female Pop - music generated by ACE-Step.

Funk Pop Neon Lights

Funk Pop Neon Lights - music generated by ACE-Step.

ACE-Step Theme

Theme music for ACE-Step.

Text to Samples: Violin

Generated sample: Violin.

Text to Samples: Synth

Generated sample: Synth.

Text to Samples: Saxphone

Generated sample: Saxphone.

Text to Samples: Pad

Generated sample: Pad.

Text to Samples: Lead Guitar

Generated sample: Lead Guitar.

Note: For optimal results, experiment with detailed text prompts. The model is continually improving with community contributions.

How It Works: From Text to Expressive Music

ACE-Step Workflow Diagram

1. Craft Your Prompt

Write a detailed text description of the music you want to create. Specify genre, mood, instruments, tempo, and any other desired characteristics.

2. (Optional) Advanced Controls

For more fine-grained control, explore advanced parameters available in the generation interface or code for aspects like song structure or specific instrument nuances.

3. Generate the Music

Run the ACE-Step model (either locally via the app or using the online demo). The model processes your prompt and generates an original piece of music.

4. Listen and Iterate

Playback the generated audio. Refine your prompt and regenerate to explore different musical ideas. Download the audio file for your projects.

How to Install and Use ACE-Step: Quick Guide

1. Environment Setup

  • Prerequisites: Python 3.10, Conda (recommended) or venv, Git.
  • Create & Activate Environment: 
    # Conda
conda create -n ace_step python=3.10 -y
conda activate ace_step
    (See full guide for venv on different OS)

2. Installation

  • Clone Repository: git clone https://github.com/ace-step/ACE-Step.git && cd ACE-Step
  • Install PyTorch: Visit pytorch.org for system-specific command.
  • Install Requirements: pip install -r requirements.txt

3. Running ACE-Step

  • Launch: python app.py
  • Access UI: Open http://127.0.0.1:7865 in your browser.
  • (Advanced command-line args available for port, GPU, sharing, etc.)

4. Using the Interface (Key Features)

  • Text2Music: Generate audio from text prompts and lyrics.
  • Retake: Create variations of previous generations.
  • Repainting: Selectively regenerate parts of an audio track.
  • Edit: Modify tags or lyrics (option to preserve melody).
  • Extend: Lengthen existing audio tracks.

ACE-Step Technical Highlights

ACE-Step Application Map Diagram

Core Concept & Architecture

  • Open-Source Foundation Model: Aims to be the "Stable Diffusion for music."
  • Architecture: Integrates Diffusion-based generation, Deep Compression AutoEncoder (DCAE), and a lightweight Linear Transformer.
  • Semantic Alignment: Uses MERT & m-hubert (REPA) for enhanced training.

Key Performance

  • Speed: Generates ~4 minutes of music in 20 seconds (NVIDIA A100).
  • Quality: Superior musical coherence, lyric alignment, and fine-grained acoustic detail.

Training-Free Controllability

  • Variations Generation: Adjust noise mix for diverse outputs.
  • Repainting: Regenerate specific audio sections with mask constraints.
  • Lyric Editing: Modify lyrics preserving melody/vocals via flow-edit technology.

LoRa Fine-Tuned Applications

  • Lyric2Vocal: Generate vocal samples directly from lyrics.
  • Text2Samples: Create instrumental loops & SFX from text.

Noted Limitations

  • Output inconsistency (seed/duration sensitive).
  • Some style-specific weaknesses.
  • Vocal synthesis quality and control granularity are areas for ongoing improvement.
GitHub Repository Online Demo