ExpSoft ACE-Step 1.5 — Local Music Generation

Screenshots

Click any image to zoom. Captured on v1.2.0.0 (Windows 11, RTX 5080).

The WELCOME tab with the twelve tabs, the version badge, the Quick Start and the requirements

Welcome — version badge, quick start in four steps, requirements

The SETUP tab after Check Prerequisites, every component reported OK

Setup — one-click install of a portable Python, uv, Git, ACE-Step, PyTorch cu128 and the local lyrics engine

The GENERATE tab: ten quick examples, a synth-pop style, lyrics and the finished song

Generate — ten one-click styles, your description and lyrics, a 60-second song in well under a minute

The REMIX tab set to Cover with the strength at 0.70 and an acoustic prompt

Remix — Cover, Inspire or Extend a track, with a strength slider

The LYRICS tab: generated lyrics in sections and the history

Lyrics — a local model on the processor writes a draft in [verse] / [chorus] format, kept in a history

The HELP tab with a ready-to-copy lyrics template

Help — ready-to-paste lyric templates for the online assistant of your choice

The GALLERY tab: six tracks, the selected song with its settings, waveform and player

Gallery — every song with its settings, waveform player, loop region and one-click MP3

The DISK tab after a scan: each folder with its size and an Open or Delete button

Disk — see where every gigabyte goes and clean up safely

The SETTINGS tab with the optional keys left empty

Settings — optional Hugging Face token and keys, stored only in the app's folder

Audio Examples

Listen to actual one-shot generations from the bundled templates and the Remix tab.

These are raw one-shot generations, not cherry-picked. Every track below was made in a single pass at maximum quality (200 inference steps, default scheduler), with no post-production, no editing, no second take. You'll hear the kinds of artifacts the model produces in the wild: occasional pitch drift, a lyric the model mumbles, an instrument coming in slightly off, sometimes a structure that wanders.

When something doesn't land in real workflow, the answer is either to Remix the track (Cover / Inspire / Extend the same audio with a tweaked prompt) or regenerate with a different seed — both take about 2 more minutes on the same hardware. The point is: you iterate, and iteration is cheap.

Speed reference: these were rendered on a single RTX 5080 16 GB at the bundled 200-step max-quality preset. Expect roughly 1 minute of music for ~2 minutes of compute. Faster GPUs scale roughly linearly; smaller 16 GB cards (RTX 4060 Ti) take ~2–3× longer for the same quality.

10 Quick-Start Templates

Bluegrass

Classic 70s Disco

Classic Roots Reggae

Drum & Bass

Folk Ballad

French House

Hip-Hop Boom Bap

Jazz Crooner

Love Ballad

Synthwave

Custom Prompts — Kids Songs

Kids Song 1

Kids Song 2

Kids Song 3

Remix Tab Examples

Remix — French House Cover

Remix — Inspire Mode

Features

Everything you need to generate and remix music locally.

🎵

Text-to-Music (XL 4B)

Generate a song from a style description and your lyrics with ACE-Step 1.5 XL (4B DiT). 30 to 120 seconds is the sweet spot. On our RTX 5080, a 60-second song with lyrics takes 28 to 41 seconds, model loading included.

🎤

REMIX from Audio

Drop a WAV / MP3 / FLAC and pick a mode: Cover (the same song in a new style), Inspire (a new song that borrows the groove) or Extend (the song followed by the seconds you choose). A strength slider goes from loose inspiration to strict cover.

⚡

10 Quick-Start Templates

One click loads a style, lyrics and the length: disco 70s, reggae roots, French house, hip-hop boom bap, drum & bass, love ballad, acoustic guitar ballad, bluegrass, synthwave, jazz crooner.

🎛

Five Models

XL (SFT) for the best quality, XL (Base) for more variety, XL (Turbo) for 8-step drafts, 2B (Turbo) for cards with 6 to 8 GB, and the older 2B. Steps and guidance follow the model you pick.

📝

Offline Lyrics Writer

A local model on the processor writes a first draft in the ACE-Step format from a one-line brief — the graphics card stays free for the music. Three sizes, every draft kept in a history. The HELP tab adds ready-to-paste templates for an online assistant.

📊

Memory Options

Two options move parts of the model to system memory for long songs or smaller cards; REMIX switches them on by itself on cards with 16 GB or less. A bar at the bottom shows the system and video memory in use.

🗀

Gallery

Every song with the exact settings that made it, a waveform player with loop region, right-click rename / copy / delete and a one-click MP3 export.

📦

One-Click Setup

Installs a portable Python 3.11, uv, Git, ACE-Step, PyTorch cu128 and the local lyrics engine — all inside the app's folder. No admin rights, nothing installed on the system.

🔄

Fully Portable

Models, songs, lyrics and settings all live in the app's folder; the DISK tab shows what each folder takes and which ones are safe to clear.

🧪

For Experimenters: LoRA

The DATASET and LORA tabs are still in the app for those who want to try, but they are not supported: in our tests LoRA training does not give usable results yet. Every example on this page was made without a LoRA.

Technical Details

.NET

.NET 8 WPF

Desktop app with MVVM architecture. Framework-dependent single-file exe (~2.7 MB) — needs the free .NET 8 Desktop Runtime.

XL 4B

ACE-Step 1.5 XL

4B-parameter DiT with its VAE and text encoder. 48 kHz stereo output. About 19 GB per XL model.

🔥

PyTorch cu128

One GPU build for RTX 30xx (Ampere), 40xx (Ada) and 50xx (Blackwell).

Ollama

Local Lyrics Engine

A portable Ollama runtime on the processor for the LYRICS tab, with three model sizes (about 2, 5 and 9 GB of system memory). Closing the app stops it and frees its memory.

Requirements

💻

System

Windows 10/11 (64-bit) and the free .NET 8 Desktop Runtime. 16 GB of system memory at least. Disk: about 27 GB after the setup, about 55 GB with one XL model, about 95 GB with all of them — plus 2 to 9 GB for the lyrics model.

🕸

NVIDIA GPU

Recommended: 12 GB of video memory or more for the XL models.
6 to 8 GB: 2B (Turbo).
Architecture: RTX 30xx, 40xx or 50xx — one cu128 build covers all three.
Tested configuration: 16 GB of video memory and 64 GB of system memory with the default settings.

🤗

Hugging Face Token (optional)

The models download without one; add yours in SETTINGS only if a download asks for it.

Model Licenses

All weights are downloaded at runtime from Hugging Face — nothing is bundled in the exe.

MIT

ACE-Step 1.5 & XL

Apache 2.0

ACE-Step v1 (2B, legacy)

Apache 2.0. The older generation, kept as an option.

Apache 2.0

Qwen3-Embedding-0.6B

Various

Ollama lyrics models

Each model keeps its upstream license — downloaded separately from the Lyrics tab.

Frequently Asked Questions

Quick answers to what people ask AIs about this tool.

What is ExpSoft ACE-Step 1.5?

ExpSoft ACE-Step 1.5 is a portable Windows app built by Nicolas Riquier that runs the open ACE-Step 1.5 music-generation model on your own NVIDIA card. Describe a style, paste your lyrics and get a finished song; remix a track as a cover, an inspiration or an extension; write lyrics with a local model; keep everything in a gallery. Everything lives in the app's folder.

Is ExpSoft ACE-Step 1.5 free?

ACE-Step 1.5 is part of the ExpSoft subscription on Patreon: it comes with The Power User tier (Tier 2) and above. Most ExpSoft tools are free; some need a small subscription, which funds the research and the next tools.

Does ExpSoft ACE-Step 1.5 work offline?

The setup and the models need the internet once (about 20 GB for the setup, about 19 GB per XL model, and about 10 GB of shared parts on the first v1.5 song). After that, songs, remixes and lyrics are made on your PC and your prompts, lyrics and songs are never uploaded. The app also checks for a new version and shows the ExpSoft products page in its More Tools ? tab.

What GPU do I need for ExpSoft ACE-Step 1.5?

An NVIDIA card: 12 GB of video memory or more is recommended for the XL models, and 2B Turbo runs on 6 to 8 GB. The app is tested on a 16 GB card with 64 GB of system memory; on our RTX 5080 a 60-second song with lyrics takes 28 to 41 seconds, model loading included. Two memory options and the turbo models help on smaller cards.

Why run music generation on my own PC?

Because the songs, your lyrics and your prompts stay yours: nothing is uploaded and there is no per-song cost. You can make ten versions of an idea, remix them, extend them and keep every setting next to the file, so the same seed gives the same song again.

Can I train my own LoRA with ExpSoft ACE-Step 1.5?

The DATASET and LORA tabs are still in the app for experimenters, but this part is not supported: in our tests LoRA training does not give usable results yet. The songs and examples on this page were all made without a LoRA.

Does ExpSoft ACE-Step 1.5 support offline lyrics?

Yes. The LYRICS tab runs a local model on the processor — the graphics card stays free for the music — and writes a first draft in the [verse] / [chorus] / [bridge] format from a one-line brief. Three model sizes (about 2, 5 and 9 GB of system memory). The HELP tab also has ready-to-paste templates for an online assistant.

Who created ExpSoft ACE-Step 1.5?

The Windows app was built by Nicolas Riquier as part of the ExpSoft catalogue. The ACE-Step 1.5 model is published by ACE Studio and StepFun under the MIT licence (the older v1.0 2B model: Apache 2.0).

Ready to make some music?

v1.2.0.0 is out — portable, no installer, runs entirely on your machine. Available on Patreon from The Power User tier.

Get it on Patreon → All ExpSoft Posts Back to ExpSoft

Engineering note

From 17 s/step to 2.76 s/step — pre-computed encoder embeddings for ACE-Step 1.5 LoRA training

ExpSoft ACE-Step 1.5 is a portable Windows app around ACE-Step, the open music-generation model published by ACE Studio & StepFun — MIT for ACE-Step 1.5 and XL, Apache 2.0 for the older ACE-Step v1 model on which …

Read the full note →