Introducing ACE-Step 1.5, a music generation AI capable of generating high-quality vocal music at lightning speed. It can run locally on PCs with less than 4GB of VRAM and supports LoRA.

The music generation AI ' ACE-Step 1.5 ' was released as an open model on February 3, 2026. ACE-Step 1.5 can quickly generate high-quality music with vocals. It consumes less than 4GB of VRAM, so it can run on most PCs with graphics cards.

ACE-Step 1.5: Pushing the Boundaries of Open-Source Music Generation

We're releasing ACE-Step-v1.5(2B), a fast, high-quality open-source music model.

It runs locally on a consumer-grade GPU, generates a full song in under 2 seconds(on an A100), supports LoRA fine-tuning, and beats SUNO on common eval metrics.

GitHub: https://t.co/Y2CQETtltB

Key… pic.twitter.com/68OA1QQrrD

— ACE Music (@acemusicAI) February 3, 2026

ACE-Step 1.5 is a music generation AI with a hybrid architecture that combines a language model (LM) and a diffusion transformer (DiT). It can generate high-quality songs with vocals in response to prompts, and can also create songs with a similar atmosphere to existing songs using LoRA. It can generate songs from a few seconds of looped audio to 10 minutes of music, with quality exceeding that of commercial models. It can generate a song in less than 2 seconds on an NVIDIA A100, and less than 10 seconds on a GeForce RTX 3090.

There are several demo tracks available on the official website, so I tried playing them. It seems that it's possible to generate Japanese songs to a certain extent by entering the lyrics in romaji.

Demo sound sources of music generation AI 'ACE-Step 1.5' - YouTube


A demo app that allows you to run ACE-Step 1.5 on a browser has also been released, so I tried using it.

🎛️ ACE-Step V1.5 Playground💡
https://huggingface.co/spaces/ACE-Step/Ace-Step-v1.5

Enter a description of the entire song in the 'Prompt' field, enter lyrics in the 'Lyrics' field, and click 'Generate Music.'

The prompt I entered this time is below. I'll try to generate some electronic J-POP.

An energetic, J-pop vocal performance over a driving electronic beat. The track is built on a foundation of punchy synth bass, crisp drum machine rhythms, and layered synthesizers. A powerful female vocal delivers catchy melodies in Japanese.

The lyrics go like this

osusi daisuki. susi ga suki.
osoba daisuki. soba mo suki.
demo demo ichiban sukinanowa.
hamburger! hamburger! hamburger!
I like hamburger, my favorite food is hamburger!

The finished song is below. The Japanese is a bit iffy, but the English vocals are perfect.

I tried making Japanese music with music generation AI 'ACE-Step 1.5' - YouTube


The ACE-Step 1.5 model data is distributed free of charge at the following link under the MIT License.

ACE-Step/Ace-Step1.5 · Hugging Face
https://huggingface.co/ACE-Step/Ace-Step1.5

A Japanese tutorial is also available.

ACE-Step 1.5 Ultimate Guide (Must Read) | ACE-Step-1.5/docs/ja/Tutorial.md at main · ace-step/ACE-Step-1.5 · GitHub
https://github.com/ace-step/ACE-Step-1.5/blob/main/docs/ja/Tutorial.md

It's also already supported by the AI-powered app ComfyUI. According to the ComfyUI development team, a 4-minute song can be generated in under 10 seconds on a GeForce RTX 3090, and under 1 second on a GeForce RTX 5090.

ACE-Step 1.5 just dropped in ComfyUI. Full songs in under 10 seconds. Less than 4GB VRAM. Open-source music generation just got serious. pic.twitter.com/nLYsx52dkq

— ComfyUI (@ComfyUI) February 3, 2026