ACE-Step 1.5
An MIT-licensed music foundation model: full songs in seconds, from about 4 GB of VRAM, with covers, repainting and 50+ languages.
In short. ACE-Step 1.5 makes complete songs with vocals from a text description and lyrics in a few seconds, and it runs on an ordinary gaming GPU. Both the code and the weights are MIT-licensed, so commercial use is straightforward.
Summary
ACE Studio and StepFun released ACE-Step 1.5 on 23 January 2026, and a larger 4B XL version followed on 2 April. A language model (0.6B, 1.7B or 4B, built on Qwen3) acts as a planner: it turns a short request into a detailed song blueprint with metadata, lyrics and a caption, and a diffusion transformer then renders the audio. The developers say the training data is licensed music, royalty-free and public-domain recordings and synthetic audio made from MIDI. Songs run from 10 seconds to 10 minutes, lyrics work in more than 50 languages, and the model also handles covers, repainting part of a track, stem separation, vocal-to-accompaniment and LoRA training on a few songs.
Speed and size are the selling points. The developers say a full song takes under 2 seconds on an A100 and under 10 seconds on an RTX 3090, and that the model runs in under 4 GB of VRAM. Their guide pairs the 2B turbo DiT without the language model with GPUs up to 6 GB and recommends the XL models at 20 GB or more without offloading, and the XL DiT needs at least 12 GB with INT8 quantization and CPU offload. The project rates its own quality as between Suno v4.5 and Suno v5. In the YuE2 team's WildSongBench, ACE-Step 1.5 scores 6.01 on the SongBench average, behind YuE2 (6.73) and MiniMax Music 3 (6.28), though its 7.46% phoneme error rate is among the lowest of the open models there.
The code and weights are MIT-licensed, and the developers say the music you generate can be used commercially. The claim that the training data was licensed, royalty-free and synthetic is theirs, and we have not verified it. The technical report is on arXiv (2602.00744). In the 30 days to 11 November 2026, Hugging Face counted about 97,000 downloads of the main repository, 9,100 of the XL turbo model and 816,000 of the ComfyUI team's repackaged copy. Download counts may include automated pulls.
Good for
- Quick song drafts on a gaming GPU or a laptop with 4 to 8 GB of VRAM
- Covers, repainting parts of a track, stem separation and vocal-to-accompaniment
- Personalizing the style with a LoRA trained on a few songs
- Commercial projects that want MIT-licensed code and weights
Watch out for
- The Suno comparison is the developers' own, and the YuE2 team's benchmark puts ACE-Step 1.5 below YuE2 and MiniMax Music 3 on overall quality
- The 4 GB figure is for the 2B DiT without the language model, while the XL models need 12 GB with offloading and about 20 GB for comfortable use
- Stem extraction and similar tasks need the base model, because the faster turbo and SFT versions do not support them
- The "licensed and royalty-free data" claim comes from the developers, so check copyright for your own use
Performance claims come from the developer and have not been independently verified.
How much memory does it need?
Memory figures for each variant of ACE-Step 1.5. Image, video, speech and music pipelines also load a text encoder and a decoder, so the memory needed can be larger than the main checkpoint.
| Variants | Parameters | Download size | Memory needed | Fits on |
|---|---|---|---|---|
| 1.5 Turbo (2B DiT, with 1.7B LM) BF16 | 2B | 10 GB | 4 GB | Laptop or phone |
| 1.5 XL Turbo (4B DiT) FP32 | 4B | 20 GB | 12 GB | Laptop or phone |
Download size covers every component needed to run the model, such as the generator, text encoder and decoder. Memory needed is a stated or published measured minimum where one exists; otherwise it is the download size plus about 15%. Larger images, longer clips and higher resolutions need more. The developer states a minimum of 4 GB of GPU memory.
The license in plain words
The shortest and simplest permissive license. You can do almost anything with the model as long as you keep the copyright and license notice.
See also
- ACE-Step 1.5 vs MiniMax Music 3: which open song model?
- Best open-source music and sound-generation models in 2026
- Open music models grew up in 2026: seven models for songs, instruments and sound effects
Frequently asked questions
- What is ACE-Step 1.5?
- ACE-Step 1.5 makes complete songs with vocals from a text description and lyrics in a few seconds, and it runs on an ordinary gaming GPU. Both the code and the weights are MIT-licensed, so commercial use is straightforward.
- Can I use ACE-Step 1.5 commercially?
- Yes. ACE-Step 1.5 is released under MIT, a permissive license. You can use it commercially if you follow the standard conditions, such as keeping the license notice.
- Is ACE-Step 1.5 open source?
- ACE-Step 1.5 is released under MIT, a license approved by the Open Source Initiative. Training data and training code may not be public, so it is best described as open weights under an open-source license.
- Can I run ACE-Step 1.5 on my own computer?
- Yes. The smallest variant, 1.5 Turbo (2B DiT, with 1.7B LM), needs about 4 GB of memory at 4-bit precision, which fits a modern laptop or a mid-range graphics card.
- When was ACE-Step 1.5 released?
- ACE-Step 1.5 was released by ACE-Step on January 23, 2026.
Official links
Sources
Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.
- ACE-Step 1.5 model card (Hugging Face) huggingface.co
- ACE-Step 1.5 XL Turbo model card, with GPU requirements (Hugging Face) huggingface.co
- ACE-Step 1.5 repository: features, GPU guide and release news (GitHub) github.com
- ACE-Step 1.5: Pushing the Boundaries of Open-Source Music Generation (arXiv) arxiv.org
- ACE-Step 1.5 license, MIT (GitHub) github.com
- YuE2-3B model card, WildSongBench comparison table (Hugging Face) huggingface.co
Release date and facts last verified November 11, 2026.
Related models
| Model | Released | Parameters | Context | License | Smallest variant needs |
|---|---|---|---|---|---|
| YuE2 M-A-P | 3.63B | – | CC BY-NC (Non-commercial) | 24 GB Gaming GPU | |
| MiniMax Music 3 MiniMax | 11.1B | – | MiniMax Music3 Community (Conditional) | 24 GB Gaming GPU | |
| HeartMuLa HeartMuLa | 3.94B | – | Apache 2.0 (Permissive) | 26 GB Gaming GPU | |
| Stable Audio 3 Stability AI | 460M – 1.4B | – | Stability Community (Conditional) | 4.02 GB Laptop or phone |