AI Music Generator: Songs That Come Back as Separate Stems

What does BINAI's AI music generator make?

BINAI's AI music generator writes a song, a jingle or a background track from a description in plain words, and gives it back as separate stems. You describe the style, the mood, the subject and who sings. You do not need to read music or know what a stem is. What comes back is a multitrack project: vocals, drums, bass and the rest of the instruments, each on its own track in a mixer.

That is the difference from most AI song tools, which hand you one mixed file. If the drums are too loud under your voiceover, a single file leaves you stuck. With stems you turn the drums down, or the vocals off, and keep the rest.

How does BINAI produce separate tracks?

Every text-to-music model available today, including MiniMax Music, Lyria and ElevenLabs Music, returns one mixed stereo file. None of them returns stems. So BINAI does it in two steps:

1. Generate the song as a mix, with the model you picked.
2. Take it apart with a source separator. A fast pass delivers quick stems first, and the editor opens on those so you can start mixing. A slower, higher-quality pass then replaces them in place.

The better stems are a hybrid: vocals and drums from one separator, bass from another, and "other" calculated so that the four tracks still add up to the original song. With every track on, the editor plays the mix the model wrote, not a rebuilt version of it.

Which models can I choose?

• MiniMax 2.6, the default: full songs with natural vocals and your own lyrics, up to 5 minutes.
• Lyria 3 Pro, Google's model: rich arrangements, up to 3 minutes, the fastest.
• ElevenLabs Music: the length is exact, up to 5 minutes, which makes it the one for a background track that must end with the video.

On top of the description you can set a genre, up to four moods, a tempo between 50 and 200 BPM, the energy, a key, the vocals (none, female, male, duet or choir), the language of the lyrics, up to six instruments, the length and what the music is for: a song, background, a jingle or a podcast. BINAI turns all of that into the model's input directly, with no extra AI step in between, so the same settings always produce the same request.

What can I do in the mixer?

Each track has its own level, pan, mute and solo, plus a low, mid and high EQ, reverb and echo, a compressor, pitch shift up to 12 semitones either way at the same speed, a low cut, warmth, stereo width and a de-esser for harsh "s" sounds. You can record a take from your microphone, add an audio file as a new track, split a clip at the playhead and loop a section. The voice tools include autotune and enhance for a sung track. The multitrack editor runs in the web app at app.binai.it.

When you are done, download the mix as MP3 (320 kbps) or WAV (24-bit), or every track as its own MP3 in one ZIP, lined up from 0:00 so it opens correctly in any other editor. By default downloads are levelled to -14 LUFS, the loudness Spotify and YouTube play at. The downloaded file works under any video, including the ones you make in BINAI's editor.

How much does it cost?

A 3-minute song with its stems costs about 94 credits on Lyria 3 Pro, the cheapest model, separation included. MiniMax costs a little more, and ElevenLabs is priced by the second, so long tracks on it cost more. Plans include 1,000 (Solo), 2,400 (Pro), 4,300 (Business) or 9,000 (Studio) credits a month, shared with every other tool. Music is on every paid plan; training a model of your own voice needs Business or Studio. See pricing.

What are the honest limits?

Stem separation is very good but not perfect: a little of one instrument can bleed into another track, most often into "other". The length you ask for is exact only on ElevenLabs. And BINAI's content check cannot listen to music. For a song it measures the audio and reads the lyrics, and it gives no scores, because scoring a hook nobody heard would be misleading.

Can I use AI music on YouTube and in ads?

YouTube allows AI music, including in monetized videos, if the track does not copy an existing song or imitate a real artist, and if you disclose realistic AI music when it is the main focus of the video. Purely AI-generated music may have no copyright at all, which means you may not be able to stop others reusing it. Read can you use AI music on YouTube and who owns AI-generated content before you build a channel or an ad campaign on it.

BINAI offers no celebrity or artist voices. A voice change to your own voice is possible from the Business plan, and it starts with a recording of a consent phrase that confirms the voice is yours.