All guides

How do I use AI stems to prepare vocals and instruments?

AI stems are estimated audio parts separated from a mixed recording, such as vocals, drums or bass. Use an AI stem separator when you need to isolate an existing performance. AI Music.Video's Add Instrumental and Add Vocals tools generate new parts; they do not split a finished recording into its original instruments.

By AI Music.Video · Updated:

Before you start

Start with the best source recording you can use and keep an untouched copy. Decide whether you need vocal stems, a karaoke backing track or a specific instrument. Check which outputs the external separator actually supports before uploading or installing anything.

Steps

  1. Choose a separator with the required outputs. Audacity's OpenVINO overview describes vocal/instrumental and four-part separation options. Check current platform and version compatibility before installation. A combined other-instruments output does not promise an isolated piano.
  2. Separate a short, representative section and listen to every output. Compare quiet passages, loud choruses and sustained notes. Keep the original timing and start position when exporting stems so they can be aligned again in an audio editor.
  3. For karaoke, follow how to get the instrumental of a song. For removing the sung words, use how to delete lyrics from a song. If the result contains strong vocal remnants, try a cleaner source or an authorized instrumental version.
  4. Use the prepared part in the appropriate tool. Open Add Instrumental for a vocal recording that needs accompaniment, or Add Vocals for an instrumental that needs singing. Supply the required title, style and exclusions, review the displayed charge, then listen to the generated result.

Worked example

Suppose you have a vocal recording and want a new acoustic arrangement. You can send that vocal source to Add Instrumental without removing instruments that are not present. If you start with a full band recording, separate and inspect the vocal first. These inputs ask the model to do different jobs.

Check the result

You should have clearly named parts and a mix that passes a listening check. For a longer backing track, use the song extension workflow. For a vocal ensemble direction, read the female trio guide, which explains the limits of a voice preference.

If something goes wrong

To extract piano from a song, the separator needs an explicit piano output and the result still needs review. A piano remover must isolate that part before you can mute it. Learning how to separate instruments in a song starts with checking supported outputs, not assuming every instrument can be recovered cleanly.

More questions

How do I isolate instruments from a song?

Choose a model that names the instrument you need, export its estimated stem and listen for other sounds leaking into it. Bass and drums outputs do not imply piano or guitar outputs. Keep the original mix so you can judge whether the isolated part remains usable.

How do I get rid of instruments in a song?

Separate the supported parts, mute the unwanted stem in an audio editor and audition the remaining mix. Instruments that share one output cannot be muted independently without another separation method. Strong processing can damage sounds you wanted to keep.

Can a music maker game supply a backing track?

A loop game can help you explore an arrangement, but export and reuse options depend on the game. Read music maker games and production tools before treating a playful sketch as a finished recording.