MOSS-SoundEffect v2.0

OpenMOSS's Apache 2.0 text-to-sound-effect model: ambience, creatures and actions up to 30 s at 48 kHz, from English or Chinese prompts.

In short. MOSS-SoundEffect v2.0 turns a sentence such as "a dog barking loudly in a park" into a sound effect of up to 30 seconds. It understands English and Chinese prompts, and it is Apache 2.0.

Summary

OpenMOSS released MOSS-SoundEffect v2.0 on 25 May 2026 as the text-to-sound member of its MOSS-TTS family. It replaces the first version's discrete-token autoregressive design with a Diffusion Transformer of 1.3 billion parameters trained with flow matching, paired with a DAC VAE and a Qwen3 text encoder. From a natural-language caption it generates environmental, urban, creature and human-action sounds, plus some short musical or percussive clips, at 48 kHz and up to 30 seconds per call. Prompts can be written in English or Chinese.

The card recommends 100 solver steps, a guidance scale of 4.0 and a default clip length of 10 seconds. It recommends a Python 3.12 environment with a CUDA build of PyTorch, wraps the model in torch.compile with Triton CUDA graphs, and warns that the first call can take a few minutes to compile. The repository is about 11 GB, and the card gives no VRAM figure and no benchmark scores, so test memory and quality on your own GPU. A Gradio demo and fine-tuning recipes are in the GitHub repository.

The model is Apache 2.0, and the card adds no extra terms. In the 30 days to 11 November 2026, Hugging Face counted about 6,000 downloads. For sound effects, Stable Audio 3 is the other open option on this site, with longer clips (up to 6 minutes 20 seconds for Medium) and eight sampling steps, but under a revenue-capped license.

Good for

  • Sound effects for games, video and apps, written as a text description
  • Short clips under a fully permissive Apache 2.0 license
  • English and Chinese prompts
  • Fine-tuning on your own sound library, with the recipes in the repository

Watch out for

  • Each call makes at most 30 seconds of audio
  • 100 sampling steps per clip make it slower than few-step models such as Stable Audio 3, which uses eight
  • The first run compiles for a few minutes, and the card recommends Python 3.12 with CUDA
  • No benchmark or VRAM figures are published, so test quality and memory yourself

Performance claims come from the developer and have not been independently verified.

How much memory does it need?

Memory figures for each variant of MOSS-SoundEffect v2.0. 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
V2.0 (1.3B DiT) FP32 1.3B 11 GB 13 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 license in plain words

One of the most widely used permissive licenses. You may use, change and sell products built on the model with only light conditions.

More about Apache 2.0

See also

Frequently asked questions

What is MOSS-SoundEffect v2.0?
MOSS-SoundEffect v2.0 turns a sentence such as "a dog barking loudly in a park" into a sound effect of up to 30 seconds. It understands English and Chinese prompts, and it is Apache 2.0.
Can I use MOSS-SoundEffect v2.0 commercially?
Yes. MOSS-SoundEffect v2.0 is released under Apache 2.0, a permissive license. You can use it commercially if you follow the standard conditions, such as keeping the license notice.
Is MOSS-SoundEffect v2.0 open source?
MOSS-SoundEffect v2.0 is released under Apache 2.0, 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 MOSS-SoundEffect v2.0 on my own computer?
Yes. The smallest variant, V2.0 (1.3B DiT), needs about 13 GB of memory at 4-bit precision, which fits a modern laptop or a mid-range graphics card.
When was MOSS-SoundEffect v2.0 released?
MOSS-SoundEffect v2.0 was released by OpenMOSS on May 25, 2026.

Official links

Sources

Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.

Release date and facts last verified November 11, 2026.

Related models

Model Released Parameters Context License Smallest variant needs
Stable Audio 3 Stability AI 460M – 1.4B – Stability Community (Conditional) 4.02 GB Laptop or phone
MOSS-TTS OpenMOSS 4.55B – 8.49B – Apache 2.0 (Permissive) 10 GB Laptop or phone
Magenta RealTime 2 Google DeepMind 230M – 2.4B – CC BY 4.0 (Permissive) 2.53 GB Laptop or phone
ACE-Step 1.5 ACE-Step 2B – 4B – MIT (Permissive) 4 GB Laptop or phone