Skip to content

How-ToGPU

How to Generate Music Locally on Your PC With ACE-Step 1.5

ACE-Step 1.5 is a free, MIT-licensed music model that makes full songs with vocals from a description and lyrics on your own GPU. The developers recommend settings by VRAM, from a DiT-only mode for 6GB cards to XL models on 20GB-plus. Install the portable Windows package or the uv quick-start, or use ComfyUI's all-in-one checkpoint.

By , Editorial team

5 min read

How to Generate Music Locally on Your PC With ACE-Step 1.5

AI music generators such as Suno run in the cloud, but you can now make full songs, with vocals and lyrics, on your own PC. ACE-Step 1.5 is an open-source music model from ACE Studio and StepFun that turns a text description and optional lyrics into audio, and it's designed to run on consumer graphics cards. This guide covers the hardware you need and the two easiest ways to run it: the developers' own app and ComfyUI. It's based on the official ACE-Step documentation and ComfyUI's guide; Forge PC hasn't benchmarked ACE-Step itself, so the speed figures below are the developers' claims.

What ACE-Step 1.5 can do

According to the project's GitHub page, ACE-Step 1.5 can:

  • Generate songs from 10 seconds to 10 minutes long, with lyrics in more than 50 languages.

  • Take reference audio to guide the style, make covers, and "repaint" (regenerate) selected parts of a track.

  • Separate tracks into stems, add accompaniment to a vocal recording, and control BPM, key and time signature.

  • Train your own LoRA style add-ons; the developers say a LoRA from eight songs takes about an hour on an RTX 3090 with 12GB of VRAM in use.

The developers claim generation takes under 2 seconds per full song on an NVIDIA A100 data-centre GPU and under 10 seconds on a GeForce RTX 3090. Your results will depend on your GPU, the model you pick and the settings you use.

The model is released under the MIT licence, and its Hugging Face model card says the generated music can be used for commercial purposes. As always with AI-generated content, check the current licence and terms yourself before releasing anything commercially, and don't use it to imitate a specific artist.

What hardware do you need?

ACE-Step's own guidance on which model to use is based on your graphics card's VRAM. The app picks a configuration automatically, but this is the developers' table in simplified form:

Your GPU's VRAM

Recommended setup (per ACE-Step)

6GB or less

2B "turbo" music model only, with the language model planner off, INT8 quantisation and full CPU offload

6GB to 8GB

2B turbo model with the small 0.6B language model

8GB to 16GB

2B turbo or SFT model, with the 0.6B (8-12GB) or 1.7B (12-16GB) language model

16GB to 20GB

2B SFT or XL turbo with the 1.7B language model; XL needs CPU offload below 20GB

20GB or more

XL models fit without offload; 4B language model available at 24GB and up

The larger XL models, released in April 2026, use a 4-billion-parameter decoder for higher audio quality and need at least 12GB of VRAM with offloading, or 20GB without, according to the developers. ComfyUI's guide says the lightest configuration, which runs the music decoder alone, needs under 4GB of VRAM.

An NVIDIA graphics card is the most straightforward option (ACE-Step recommends CUDA), but the project also supports AMD cards through ROCm, Intel GPUs, Apple Silicon Macs and CPU-only mode. Our guides to running local AI on an AMD Radeon GPU and on Intel Arc cover the general setup for those cards, and how much VRAM you need explains memory in more detail.

Option 1: Install the ACE-Step app on Windows

The developers offer a portable Windows package with dependencies pre-installed, linked from the project's GitHub page. That's the simplest route if you don't want to use the command line.

If you prefer a standard install, the project's quick-start steps are:

  1. Install Python 3.11 or 3.12. The developers note that ROCm on Windows (for AMD cards) needs Python 3.12.

  2. Install the uv package manager. In PowerShell, the project gives this command: powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

  3. Install Git if you don't have it, then download the project and its dependencies:

  4. git clone https://github.com/ACE-Step/ACE-Step-1.5.git

  5. cd ACE-Step-1.5

  6. uv sync

  7. Start the app with uv run acestep. The models download automatically the first time, so expect a wait.

  8. Open http://localhost:7860 in your browser to use the web interface.

AMD and Intel users should follow the separate launch scripts and instructions in the project's installation guide.

Option 2: Use ACE-Step in ComfyUI

If you already use ComfyUI for image generation, it runs ACE-Step 1.5 natively. If you don't have it yet, see our guide to installing ComfyUI on Windows.

  1. Update ComfyUI. ComfyUI's guide says the templates won't appear if your copy is out of date.

  2. Download the all-in-one checkpoint, ace_step_1.5_turbo_aio.safetensors, which ComfyUI recommends for most users, and put it in ComfyUI/models/checkpoints.

  3. Open the Template Library and search for "ACE-Step 1.5 Music Generation AIO", or download the workflow JSON from ComfyUI's guide.

  4. Enter your style tags and lyrics, then run the workflow.

If you're short of VRAM, our ComfyUI low VRAM guide has general tips. One-click launchers can also install ComfyUI for you; see our Pinokio and Stability Matrix guides.

Writing your first prompt

ACE-Step works from two inputs: a description or tags for the style, and optional lyrics. Some tips based on the features the project documents:

  • Describe the genre, mood, instruments and vocal style in the caption or tags, for example "upbeat synth-pop, female vocals, punchy drums, 120 BPM".

  • Use the metadata controls to set the duration, BPM, key and time signature rather than relying on the text alone.

  • Use "Simple mode" if you only want to type a short description, and the language model will expand it into a fuller plan, according to the developers.

  • Generate a batch of versions at once (up to eight, according to the developers) and keep the best one.

Our verdict

ACE-Step 1.5 makes local AI music generation genuinely practical on a gaming PC. If you have an NVIDIA card with 8GB or more, start with the portable Windows package or the ComfyUI all-in-one checkpoint and the default turbo model. Owners of 6GB cards can still try it with the lightweight DiT-only setup, while 20GB-plus cards can run the higher-quality XL models without offloading. For other local AI audio tools, see our guide to transcribing audio locally with Whisper.

The bottom line

ACE-Step 1.5 makes local AI music generation genuinely practical on a gaming PC. If you have an NVIDIA card with 8GB or more, start with the portable Windows package or the ComfyUI all-in-one checkpoint and the default turbo model. Owners of 6GB cards can still try it with the lightweight DiT-only setup, while 20GB-plus cards can run the higher-quality XL models without offloading. For other local AI audio tools, see our guide to transcribing audio locally with Whisper.

Sources

  1. 01ACE-Step, ACE-Step 1.5 on GitHub
  2. 02ACE-Step, Ace-Step1.5 model card on Hugging Face
  3. 03ComfyUI Docs, ACE-Step 1.5 music generation
  4. 04ACE-Step, GPU compatibility guide

About the author

Forge PC Staff

Editorial team

The shared byline for articles researched, written and edited by the Forge PC editorial team. How we source, check and correct our work is set out in our editorial policy.

More from Forge PC Staff

Read next

Lemonade vs Ollama: Which Local AI Server Should You Use on an AMD or NVIDIA PC?

Versus

Lemonade vs Ollama: Which Local AI Server Should You Use on an AMD or NVIDIA PC?

Ollama is the safest default for local LLMs, with a big model library and support for NVIDIA, AMD and Apple hardware. Lemonade, an open-source server with AMD-backed optimisations, is the better choice on Ryzen AI and Radeon PCs if you want NPU support or speech and image models alongside LLMs. It also speaks the Ollama and OpenAI APIs.

Forge PC Staff5 min

Sleep vs Hibernate vs Shut Down in Windows 11: Which Should You Use on a Gaming PC?

Versus

Sleep vs Hibernate vs Shut Down in Windows 11: Which Should You Use on a Gaming PC?

Sleep keeps your session in RAM for the fastest resume but loses unsaved work in a power cut. Hibernate writes it to hiberfil.sys on your SSD and uses no power, but resumes more slowly. Shut down or restart for a clean start. On a gaming desktop, sleep for short breaks and hibernate for long ones; on a laptop, hibernate or shut down before travelling.

Forge PC Staff5 min

RTX 5090 vs RTX 4090 for Local AI: Is 32GB and 1.8TB/s Worth the Upgrade?

Versus

RTX 5090 vs RTX 4090 for Local AI: Is 32GB and 1.8TB/s Worth the Upgrade?

For local AI, the RTX 5090's 32GB of GDDR7 and 1,792GB/s of bandwidth beat the RTX 4090's 24GB and 1,008GB/s, so bigger models and longer contexts fit and token generation should be faster, with FP4 support too. It needs more power and NVIDIA recommends a 1,000W PSU. If your models fit in 24GB, a good-value RTX 4090 is still very capable.

Forge PC Staff5 min

How to Add Web Search to Your Local AI Chatbot in Open WebUI

How-To

How to Add Web Search to Your Local AI Chatbot in Open WebUI

Open WebUI can give a local model live web search. Start with the built-in DDGS engine, which needs no API key, move to a self-hosted SearXNG instance (with JSON output enabled) for more control, or use a hosted search API. Raise the model's context length to at least 16K tokens and use a model with good tool calling for useful answers.

Forge PC Staff5 min