Demucs or UVR: which tool splits vocals from music
Both do the same job: split a recording into vocals and everything else. Demucs is one clear model: MIT, tens of millions of parameters, splits a track into vocals, drums, bass and the rest, runs on a CPU, with version 4.1.0 released in 2026-07. UVR is not a model but a large community collection: MDX-Net, BS-RoFormer, Mel-RoFormer, SCNet, DTTNet, listed in the catalog as the quality leader for vocals among open solutions, with updates arriving constantly, the latest in 2026-08. The price is licensing: the code is MIT, but each set of weights carries its own terms and has to be checked individually. For a production pipeline Demucs is more predictable; for the best result on a specific recording it pays to try weights from UVR.
Comparison based on catalog data
| Parameter | Demucs | UVR / MDX-Net / RoFormer (разделение звука) |
|---|---|---|
| Category | Voice: speakers and sound, Music and sound | Voice: speakers and sound, Music and sound |
| Developer | Meta AI, then Alexandre Défossez, France | Community: Ultimate Vocal Remover (Anjok07), ZFTurbo, MVSep, International community |
| Releases | Dec 2022 – Jul 2026 | Dec 2022 – Aug 2026 |
| Sizes | tens of millions of parameters | from tens to hundreds of millions of parameters |
| Hardware | Laptop, 1 GPU | Laptop, 1 GPU |
| Commercial use | Commercial use allowed | Commercial use with conditions |
| License | MIT | Code MIT; individual weights have different terms, check each model |
| Russian | Not applicable | Not applicable |
| Ollama | No | No |
| Without GPU | Yes | Yes |
| Tasks |
|
|
Choose Demucs if
- You want one model under a clean MIT license, with no terms to untangle
- Separation is built into a service and must behave the same on every file
- You need all four stems: vocals, drums, bass and the rest
Choose UVR / MDX-Net / RoFormer (разделение звука) if
- You want the best vocal quality and will tune weights per recording
- You want a choice: MDX-Net, BS-RoFormer, Mel-RoFormer, SCNet, DTTNet
- Recency matters: the newest weights in the collection are dated 2026-08
Other comparisons
- Wan or LTX-Video: which to choose for business
- pyannote or WeSpeaker: which speaker tool to pick
- Qwen or GigaChat: which to choose for business
- Qwen or Llama: which to choose for business
- GigaChat or YandexGPT: which to choose for business
- DeepSeek or Qwen: which to choose for business
- Gemma or Llama: which to choose for business
- Mistral or Llama: which to choose for business
- Whisper or GigaAM: which to choose for business
- Whisper or Parakeet: which to choose for business
- Silero or Piper: which to choose for business
- FLUX or Stable Diffusion: which to choose for business
Need a model for your task?
An open model can run on your own server: data stays in-house, there is no per-request fee, and the model can be fine-tuned on your documents.
- SelectThe model and size for your task and hardware budget
- DeployOn your server or in a closed network, with an API
- Fine-tuneOn your data, or connect a knowledge base
- IntegrateInto your CRM, ERP, bot, website or team chat