ACE-Step 1.5 Released: Open Source AI Music Generator Running on 4GB VRAM, A Strong Rival to Suno?

This is news that will make music creators and AI enthusiasts smile.

To be honest, over the past year or two, we’ve watched commercial giants like Suno and Udio conquer the market. Although the quality of the music they generate is amazing, the feeling of “look but don’t touch” is always a bit itching. After all, these models are locked behind paywalls, and we can’t run them on our own computers, let alone fine-tune them for our own styles.

But now, the rules have changed.

ACE-Step 1.5, jointly launched by ACE Studio and StepFun, is officially open source. This is not just a new model; it is a declaration to break commercial monopolies. Imagine that you don’t need to rent expensive cloud servers, or even top-tier graphics cards. You only need an ordinary gaming graphics card to train your own AI music producer at home.

Does this sound a bit too good to be true? Let’s see what it’s really capable of.

Speed and Threshold: Fast Enough to Make You Doubt Reality

First, let’s talk about its speed, which is really crazy.

In the past, high-quality AI music generation often meant long waits or required expensive computing support. But ACE-Step 1.5 kicks this threshold to the floor. According to official data, if you have an NVIDIA RTX 3090 at hand, generating a complete song takes less than 10 seconds.

If you have a monster card like the A100? That’s even more exaggerated, finishing a song in less than 2 seconds. What does this mean? It means your inspiration hasn’t cooled down yet, and the music is already done.

Even better, its hardware requirements are incredibly friendly. You don’t need a company-level workstation. As long as your graphics card has 4GB VRAM, this model can run on your local machine. This is absolutely a huge boon for independent developers or students with limited budgets. It turns AI music generation from an “aristocratic sport” into a “national sport”.

Quality Showdown: Can Open Source Really Beat Commercial Models?

Usually, when we hear the words “open source,” we might expect the quality to be slightly discounted. But ACE-Step 1.5 doesn’t seem to intend to go down that path.

Judging from the evaluation data released on Hugging Face, this model has shown amazing strength in multiple indicators. Especially in SongEval, an indicator for evaluating overall music quality, ACE-Step 1.5’s score even surpassed Suno v5.

Of course, data is cold, and listening is believing. The music generated by this model is very close to, and even surpasses in some styles, the current commercial dominators in terms of structural coherence and sound quality clarity. It is no longer that kind of experimental product full of background noise and chaotic structure, but a creation tool that can truly be used.

LoRA Fine-tuning: Create Your Exclusive Music Soul

This is probably the most exciting feature of ACE-Step 1.5 for creators.

Although current commercial models are powerful, they are usually “black boxes.” You can only “draw cards” through text prompts. If you are lucky, you get what you like; if you are unlucky, you have to keep trying. You cannot make Suno truly learn “your” style.

ACE-Step 1.5 supports LoRA (Low-Rank Adaptation) fine-tuning. What does this mean? You can feed it a few songs of a specific style you like, or your own original works. With just a small amount of data, it can learn specific instrument timbres, arrangement habits, and even the singer’s singing style.

This is true “customization.” You can train a model specifically for writing Lo-Fi Hip Hop, or an assistant specializing in 80s Japanese City Pop. This return of control is the core value of the open source community.

Technology Decoded: Perfect Cooperation Between Planner and Executor

Why can it run so fast and the quality is still so good? This is due to its unique “hybrid architecture.”

ACE-Step 1.5 does not work blindly like traditional models. It uses a clever method of division of labor:

1. Language Model (LM) is the “Brain”: It acts as an omnipotent planner. After you enter the prompt, it doesn’t rush to make a sound, but first uses Chain-of-Thought technology to plan the blueprint of the whole song. This includes the structure of the lyrics, the arrangement of paragraphs, the direction of the style, and so on. It’s like a senior music producer writing the score before entering the recording studio.
2. Diffusion Transformer (DiT) is the “Hand”: Once the blueprint is established, this part is responsible for execution, converting the plan into high-quality audio.

This mode of thinking first and then executing, coupled with a unique internal reinforcement learning mechanism (not relying on external reward models), allows it to accurately restore the user’s intention while maintaining extremely high efficiency.

Not Just Generation: Powerful Post-Editing Capabilities

For professional musicians, simple “text-to-music” is actually not enough. We often need to modify and fine-tune. ACE-Step 1.5 obviously took this into account, providing a complete set of productivity tools:

• Cover Generation: You can throw a song in and let it re-interpret it in a completely different style.
• Repainting: Feel that a certain bar of the guitar solo is not good enough? You can modify only that segment without redoing the whole song.
• Vocal-to-BGM: This feature is very practical. It can automatically convert the vocal track into background accompaniment, which is very convenient for making Karaoke or mixing.

Moreover, it supports over 50 languages. Whether you want to make Chinese pop songs, Japanese rock, or French Chanson, it can handle it with ease. You can go to their GitHub page or Demo website to listen to the actual effects. Those Heavy Metal or Cantopop examples will definitely impress you.

Commercial Application and Copyright: Guarantee for Peace of Mind in Creation

In this moment when AI copyright disputes are constant, ACE-Step 1.5 gives a reassuring answer.

It uses the MIT License, which is one of the most permissive open source licenses. This means you can completely use the generated music for commercial purposes without worrying about receiving a lawyer’s letter one day.

The official emphasizes that the training data comes from legally licensed music tracks, royalty-free music, and high-quality synthetic data. For creators who want to use AI music in games, videos, or advertisements, this solves the biggest worry.

Frequently Asked Questions (FAQ)

To help everyone get started faster, I have compiled some common questions about ACE-Step 1.5:

Q1: Is the hardware requirement for ACE-Step 1.5 really that low? Yes. According to tests, as long as your graphics card has more than 4GB of VRAM, you can run the model locally. Of course, if you want to pursue the ultimate generation speed (such as generating a full song in 2 seconds), using a higher-end graphics card (such as RTX 3090 or A100) will make a significant difference, but the entry threshold is indeed very low.

Q2: Can I make money with the music generated by this model? Absolutely. ACE-Step 1.5 adopts the MIT license, and the official explicitly states that the model was designed for creators. You can use the generated music for commercial projects, and the training data source is compliant, significantly reducing copyright risks.

Q3: What is its biggest advantage compared to Suno or Udio? Besides being “free” and “running locally,” the biggest advantage lies in controllability. Through the LoRA fine-tuning function, you can let the model learn specific styles, which is impossible for current closed commercial models. In addition, its editing functions (such as Repainting and Cover) also provide more detailed creative control.

Q4: Where can I download and try it? You can directly visit the official GitHub repository to get the code, or download the model weights on Hugging Face. For users who are not familiar with code, the official also provides relevant guidelines, and there is even a Windows portable package available.

The emergence of ACE-Step 1.5 may mark a new stage in AI music generation. It is no longer the patent of tech giants, but has returned to the hands of every creator. Whether you want to make a catchy TikTok soundtrack or seriously want to produce a concept album, now, the tools are ready, the rest is up to your imagination.