10 tips for cleaner stem separation
Start from the best file you have, choose High quality for songs that matter, listen on headphones, and fix small leftovers in the mixer. Open the splitter
Two people can run the same song through the same AI and get different results. The difference is usually the input and a few settings. These ten tips help you get the cleanest tracks.
Use a high quality source
Lossless files such as WAV or FLAC keep detail that low bitrate MP3 files throw away. The model can only separate what is still in the file.
Avoid re-encoded copies
A song downloaded from a video site has usually been compressed more than once. If you own the CD or a purchased download, use that instead.
Pick High quality for songs that matter
The extra overlap smooths the joins between windows. Use Fast only for quick previews.
Create only the tracks you need
If you only want a karaoke version, choose No vocals. The AI still works on the whole song, but you skip building files you will never use.
Do not clip the input
An already distorted, very loud master is harder to separate. If you have a quieter version, try it.
Listen on headphones first
Small artefacts hide on laptop speakers. Headphones let you decide if a result is good enough before you export.
Use the mixer instead of fighting leftovers
If a trace of vocals remains, lower the other tracks a little or raise the bass. Small mixing changes often hide small artefacts.
Keep the tab open and the device awake
Processing runs inside the page. Locking the screen or switching apps on a phone can pause it.
Close heavy tabs
The model needs memory. Closing other tabs gives it room and avoids crashes, especially on phones.
Export WAV, convert later
WAV is lossless. If you need MP3, convert it once at the end instead of converting between formats several times.
When to switch to the app
If your phone struggles with the browser version, the Android app runs the same kind of model with native code, supports MP3 export and lets you record your voice over the music.