Settings-Karaoke-Maker
Karaoke Maker Configuration

Karaoke Maker is an optional plugin: install it from the Plugin Browser (Media Library → Plugin Browser). Installing it also downloads the tools it needs (yt-dlp, deno, ffmpeg and the AI engine, about 2 GB, one time). Its settings are under Preferences → Plugins → Karaoke Maker (main menu → Preferences, or Ctrl+P).
The first time you make a track, QUP asks you to agree that you have the rights to use the song. Until you click I Agree, nothing is made.
Karaoke Maker is an AI-powered tool that creates karaoke tracks from regular songs. It downloads a song from YouTube (or uses a file you already have), separates the vocals from the instrumental using AI, and generates timed lyrics — producing a ready-to-sing karaoke track. Output settings like brightness, duet mode, and effects are chosen per song where you make the track: Media Library → Plugins → General → Karaoke Maker (see Karaoke Maker).
Automatic Setup
Repair Tools — Downloads again whatever tool has gone missing. You don't need it for a normal install, since the tools come with the plugin; use it if Check Tools reports something as NOT FOUND. When it finishes, the tool paths below are filled in for you.
Cancel — Cancels a repair in progress.
Hardware Acceleration
Shows whether your computer's hardware is being used, and why not:
- Vocal separation and lyrics — whether the AI engine runs on your graphics card or your processor. If you have an NVIDIA graphics card but the processor-only engine is installed, a Turn on graphics-card acceleration button appears. It downloads about 2.5 GB and makes vocal separation several times faster. On a Mac these steps always run on the processor.
- Video encoding — whether making the final video uses your hardware.
Hover over a line for the details. Re-check runs the check again, for example after updating your graphics driver.
Tool Paths
If you prefer to manage tools manually or they are installed elsewhere on your system, you can set the paths here.
yt-dlp — Path to the yt-dlp executable. Used to download audio from YouTube.
deno — Path to the Deno runtime. yt-dlp needs it for YouTube downloads.
demucs — Path to Demucs, the AI vocal separation tool.
whisper — Path to Whisper, the speech recognition used to generate timed lyrics.
ffmpeg — Path to the FFmpeg executable. Used for audio and video conversion.
ffprobe — Path to ffprobe. Used to inspect audio file metadata.
YouTube
Cookies from browser — Select which browser QUP should pull cookies from when downloading from YouTube (none, Chrome, Edge, Firefox, Brave, Opera or Vivaldi). This helps with age-restricted or region-locked content. Note: Chrome and Edge must be fully closed before QUP can read their cookies.
Cookies file — Alternatively, provide a direct path to a cookies.txt file exported via a browser extension. This overrides the browser option above.
Models
Demucs model — The AI model used for vocal separation: htdemucs (the default, fast), htdemucs_ft (best quality, slower) or mdx_extra.
Whisper model — The AI model used for lyrics transcription, from tiny to large-v3. "medium" (the default) is a good balance of speed and accuracy; "large-v3" provides the best accuracy but takes more processing time and memory.
Language (blank=auto) — The language of the song for lyrics transcription. Leave blank for automatic detection. Set a language code (e.g. "en", "es", "ja") if auto-detection struggles with a particular track.
Save Settings — Saves your Karaoke Maker configuration.
Check Tools — Checks that yt-dlp, demucs, whisper, ffmpeg and ffprobe can be found at their configured paths, and shows OK or NOT FOUND for each. Run this after changing any paths.

Updated on version 2.35.0, October 31, 2026