Free audio stem separation for Windows — Demucs GUI, local AI
FrameShift splits any audio or video file into individual stems — vocals, drums, bass, instrumental — using HTDemucs running entirely on your machine. No upload, no subscription, no per-track charge. Right-click any file in Windows Explorer and the stems appear next to the source.
Stems you can extract
Vocals
Lead & backing vocals separated from the mix
Drums
Drum and percussion tracks isolated
Bass
Bass guitar and low-end separated
Other
Everything not in vocals, drums or bass
Instrumental
Full mix without vocals — karaoke-ready
Why separate audio locally?
No upload, complete privacy
Cloud services like LALAL.AI or Moises upload your audio to their servers. FrameShift processes everything on your machine — your music files never leave your disk.
Free, no limits
Cloud stem separators charge per minute of audio. FrameShift is completely free with no usage limits, no trial cap, no subscription.
Works offline
Once the HTDemucs model is downloaded, audio separation works entirely without internet. Ideal for studio or offline environments.
GPU accelerated
A DirectML GPU model is available for faster processing. A CPU model is included as fallback for machines without a compatible GPU.
Right-click from Explorer
No app to open, no import. Right-click any audio or video file and the stems appear next to the source file automatically.
Works on video files too
Right-click a video file and FrameShift extracts the audio track, separates it into stems, and saves all outputs next to the source.
How it works
Right-click an audio or video file in Windows Explorer and choose FrameShift → Audio Separation.
Choose your stems and engine — select which stems to extract and whether to use the CPU or GPU model.
Model preflight — FrameShift downloads the HTDemucs ONNX model on first use and verifies its integrity.
Separation runs locally — HTDemucs processes the audio on your machine. Time depends on track length and your hardware.
Stems appear next to the source — each stem is saved as a separate WAV file with a unique name alongside the original.
Frequently asked questions
Is there a free Demucs GUI for Windows?
Yes. FrameShift includes a free audio stem separation workflow powered by HTDemucs ONNX models running locally on Windows. No upload, no cloud, no subscription — right-click any audio or video file in Explorer and separate the stems.
What stems can FrameShift separate?
FrameShift can separate audio into vocals, drums, bass, other (everything else), and an instrumental track (full mix without vocals). The available stems depend on the selected model configuration.
What model does FrameShift use for audio separation?
FrameShift uses HTDemucs ONNX exports — the same model family behind the open-source Demucs project. Two variants are available: a CPU model for compatibility and a DirectML GPU model for faster processing.
Does audio separation require a GPU?
No. FrameShift includes both a CPU model and a DirectML GPU model. On machines without a compatible GPU, the CPU model runs automatically. Processing is slower on CPU but fully functional.
What audio formats does audio separation support?
FrameShift accepts MP3, WAV, FLAC, AAC, OGG and any format supported by FFmpeg — including video files, where the audio track is extracted automatically before separation.
How does FrameShift compare to LALAL.AI or Moises?
LALAL.AI and Moises are cloud services that charge per minute of audio and upload your files to their servers. FrameShift is completely free, processes audio locally and never uploads anything. For privacy, offline use or high-volume workflows, FrameShift is the better choice.
Separate audio stems locally — free, no upload
Download FrameShift and split vocals, drums and bass on Windows in one right-click.