Local and private music generation
Run the model on compatible local hardware when projects or source recordings should not be uploaded to a closed consumer service.
Independent tool overview
Ace-Step-1.5 at a glance
ACE-Step 1.5 is an open-source music foundation model that can generate complete stereo songs from prompts and lyrics, edit existing audio, separate tracks, analyze musical attributes, and train small LoRA adaptations on local hardware.
Visit the official Ace-Step-1.5 site ↗
- Product type
- Open-source AI music generation and editing model
- License
- MIT
- Audio output
- 48 kHz stereo
- Song duration
- About 10 seconds to 10 minutes
- Languages
- More than 50
- Local memory
- Under 4 GB VRAM with offloading for standard models
- Model options
- Base, SFT, Turbo, LM, LoRA, and newer XL variants
- Best for
- Technical musicians, developers, researchers, and custom workflows
- Last reviewed
- August 30, 2026
Overview
What Ace-Step-1.5 is
ACE-Step 1.5 is built for people who want more control over AI music generation than a closed, browser-only service usually provides. It can create 48 kHz stereo songs from text and optional lyrics, supports more than 50 languages, and handles durations from roughly 10 seconds to 10 minutes. The broader toolkit includes reference-audio conditioning, covers, repainting, track separation, vocal-to-accompaniment generation, multi-track work, audio analysis, and local LoRA training.
The tradeoff is that this is a model and developer project, not a polished all-in-one music business. Setup, checkpoint selection, hardware configuration, generation parameters, editing, mixing, and rights review remain the user's responsibility. The standard release can run with less than 4 GB of VRAM when offloading is enabled, while larger XL checkpoints require substantially more memory. It is a strong choice for technical musicians and researchers, but a hosted tool such as Suno or Udio will be easier for someone who only wants to type a prompt and receive a finished song.
Use cases
Who Ace-Step-1.5 is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Technical music creators
Control lyrics, duration, style, musical metadata, checkpoints, seeds, inference settings, and downstream editing in a reproducible workflow.
Long-form song experiments
Generate complete pieces ranging from short clips to tracks of about 10 minutes rather than limiting every output to a brief loop.
Audio editing and transformation
Use repainting, cover generation, reference audio, track separation, vocal-to-background-music, and multi-track tools alongside generation.
Custom artist or genre research
Train a LoRA from a small, properly licensed dataset to study a consistent sound without retraining the full foundation model.
Developer integrations
Build custom interfaces, batch pipelines, research tools, or creative applications around open weights and documented inference code.
Capabilities
Core Ace-Step-1.5 features
Text-to-song generation
Create a complete piece from a style prompt, optional lyrics, duration, and musical parameters.
48 kHz stereo output
The official pipeline produces stereo audio at a 48 kHz sample rate for music-focused workflows.
Long duration range
Generate material from roughly 10 seconds to 10 minutes, covering samples, cues, and full song structures.
Multilingual lyrics
The project reports support for lyrics in more than 50 languages.
Broad style coverage
Official documentation describes more than 1,000 instruments and styles that can be expressed through prompting.
Reference-audio conditioning
Use an authorized audio reference to guide the generated result while retaining control over the prompt and other inputs.
Cover generation
Transform an input performance or composition into a new rendition, subject to the rights attached to the source material.
Repainting
Regenerate a selected portion of an existing track instead of rebuilding the entire song for a local correction.
Track and stem tools
Separate material and work with vocal, accompaniment, or multi-track components in more involved production workflows.
Audio understanding
Analyze attributes such as BPM, key or scale, time signature, captions, and lyric timestamps.
Local LoRA training
Adapt the model from a small collection of authorized songs; the project documents an example using eight songs in about an hour on an RTX 3090.
Multiple checkpoints
Choose among Base, SFT, Turbo, language-model, LoRA, and newer 4B XL variants according to quality, speed, memory, and control needs.
Cross-hardware support
The project documents options for NVIDIA CUDA, Apple silicon, AMD, and Intel environments, although performance and setup vary.
Batch generation
Generate multiple candidates in one run to compare interpretations and reduce the cost of serial experimentation.
Process
How the Ace-Step-1.5 workflow works
Step 1
Choose the deployment path
Decide whether to use a hosted interface or install the open-source project locally for privacy, customization, and deeper control.
Step 2
Match a checkpoint to the hardware
Select a standard or XL model based on available VRAM, system memory, desired speed, and whether CPU offloading is acceptable.
Step 3
Install and verify the environment
Follow the official setup for the operating system and accelerator, download the required weights, and test a small known-good generation.
Step 4
Describe the music
Enter the genre, instrumentation, mood, arrangement, vocal direction, language, lyrics, duration, and optional musical metadata.
Step 5
Generate several candidates
Compare seeds, prompts, checkpoints, and inference settings rather than treating the first render as a final master.
Step 6
Edit and arrange
Repaint weak sections, separate or replace tracks, refine lyrics and timing, then assemble the preferred structure in an audio workstation.
Step 7
Finish and clear the release
Mix and master the audio, document the model and source inputs, and confirm that lyrics, references, samples, voices, and training data are authorized for the intended use.
Cost
Ace-Step-1.5 pricing and free plan
ACE-Step 1.5's code and model are available under the MIT license, so there is no software subscription for local use. Real cost depends on the hardware, electricity, storage, setup time, and any hosted compute or third-party interface selected.
Local open source
Free software
Download and run ACE-Step 1.5 on compatible hardware under the MIT license.
- Hardware and electricity are not included
- Standard models can use under 4 GB VRAM with offloading
- Installation, updates, and troubleshooting are self-managed
Larger XL models
Free software; higher compute cost
Run the newer 4B XL checkpoints for additional model capacity.
- Official guidance calls for at least 12 GB VRAM with offloading
- 20 GB or more VRAM is recommended
- Storage and generation time vary by checkpoint and hardware
Hosted or cloud use
Provider-specific
Use a hosted ACE-Step interface or rent GPU compute instead of maintaining a local environment.
- ACEMusic is linked by the project as a hosted option
- Availability, limits, privacy, and commercial terms can change
- Check the provider's current terms before uploading private or licensed audio
Pricing checked . Check current pricing at the source ↗
Assessment
Ace-Step-1.5 strengths and limitations
Where it stands out
- Open-source code and weights under a permissive MIT license
- Can run locally for privacy and workflow control
- Generates complete 48 kHz stereo songs rather than only short loops
- Supports tracks up to about 10 minutes and lyrics in more than 50 languages
- Combines generation with repainting, reference audio, covers, separation, and multi-track workflows
- Standard configurations can operate with less than 4 GB VRAM through offloading
- Multiple checkpoints let users trade speed, quality, memory, and editability
- Local LoRA training enables controlled adaptation from a small authorized dataset
- Official documentation covers musicians as well as developers
- Batch generation supports efficient comparison of multiple candidates
What to consider
- Local setup is substantially more technical than using a consumer web music generator
- Performance, installation steps, and stability vary across NVIDIA, Apple, AMD, Intel, and operating-system combinations
- CPU offloading reduces VRAM needs but can slow generation and increase system-memory pressure
- The 4B XL family needs considerably more memory than the standard configurations
- Model downloads, checkpoints, outputs, caches, and training data can consume significant storage
- Generated vocals may contain pronunciation, intelligibility, timing, or consistency problems
- Long songs can still drift in structure, instrumentation, dynamics, or lyrical coherence
- Repainting, separation, and cover outputs usually require manual listening and production cleanup
- Reference audio, cover generation, lyrics, voices, samples, and LoRA datasets must be used only with appropriate rights and permissions
- An open-source license for the model does not automatically clear copyright, publicity, trademark, contractual, or dataset issues in a particular output
- Commercial users should assess provenance, applicable law, platform rules, and client requirements rather than assuming every generated track is exclusive or copyrightable
- The repository evolves quickly, so checkpoint compatibility and documented requirements can change
- There is no bundled distribution, royalty administration, mastering, collaboration, or music-release service
- Hosted third-party implementations may have different privacy, retention, usage, and pricing terms from the open-source project
Compare
Ace-Step-1.5 alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Content Creator
Suno AI
Suno provides a simpler hosted workflow for generating complete songs without installing or maintaining a local model.
Explore Suno AI →Content Creator
Udio
Udio is another consumer-friendly hosted option for prompt-based song creation and iterative extension.
Explore Udio →Content Creator
Stable Audio 3.0
Stable Audio 3.0 is an alternative music and sound-generation model ecosystem for creators comparing model access and production workflows.
Explore Stable Audio 3.0 →Questions
Ace-Step-1.5 FAQs
What is ACE-Step 1.5?
ACE-Step 1.5 is an open-source AI music foundation model for generating complete songs and working with existing audio through tools such as reference conditioning, covers, repainting, separation, analysis, and local adaptation.
Is ACE-Step 1.5 free?
The official code and model are available under the MIT license without a software subscription. You still pay for local hardware and electricity or any cloud GPU and third-party service you choose.
Can ACE-Step 1.5 run locally?
Yes. Local operation is one of its main advantages. The standard project can run with less than 4 GB of VRAM when offloading is enabled, although speed and system-memory use depend on the computer and checkpoint.
How long can its songs be?
Official documentation describes a generation range of approximately 10 seconds to 10 minutes.
Does it support lyrics?
Yes. Lyrics are optional, and the project reports support for more than 50 languages. Generated vocals and timing still need human review.
What is the difference between Base, SFT, Turbo, and XL?
They are different checkpoints and model families. Base emphasizes flexible generation and editing, SFT is instruction-tuned, Turbo favors speed, and the newer 4B XL variants offer greater model capacity while requiring more memory. Use the current official model guide to match a checkpoint to the task and hardware.
Can I train ACE-Step 1.5 on my own music?
The project supports local LoRA training from a small music collection. Only use recordings, compositions, performances, voices, and metadata you own or have permission to use, and test the result carefully before release.
Can I use ACE-Step music commercially?
The MIT license is permissive for the software and model, but it does not guarantee that every input or output is commercially cleared. Commercial use requires a separate review of source rights, references, lyrics, voices, samples, contracts, platform terms, and applicable law.
Is ACE-Step 1.5 better than Suno or Udio?
It is better suited to local control, open customization, checkpoint selection, and technical production workflows. Suno and Udio are generally easier for people who prefer a polished hosted interface and do not want to manage models or hardware.
Bottom line
Our Ace-Step-1.5 verdict
ACE-Step 1.5 is one of the more capable open options for technically comfortable music creators: it offers long-form stereo generation, local privacy, editing tools, multiple checkpoints, and small-dataset adaptation without a subscription. Its value comes with real operational responsibility. Expect setup and production work, choose hardware and checkpoints carefully, and treat rights clearance as part of the workflow—not something the model license solves for you.
Visit Ace-Step-1.5 website ↗Similar Content Creator Tools
View all Content Creator AI tools →
Eleven v3
Eleven v3 - ElevenLabs’ most expressive AI voice model, now commercially available
Kling 3.0
Kling 3.0 - Kling's new AI video model with upgraded consistency, realism 15 second clips, native audio, and more
Wonda
Wonda - Wondercraft's AI agent for video editing and creative direction
Krea Realtime
Krea Realtime - Krea's real-time, long-form AI video generation model
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.