Get a beat written around your vocal

How it works under the hood

This is audio-conditioned generation: the model takes your recording as input, not only a text prompt. ACE-Step 1.5, an open model under the Apache-2.0 licence, listens to the vocal and generates the parts that are missing, such as drums, bass, keys, percussion and synths, in the style you describe. Because it hears your phrasing, the instruments follow your timing and harmony instead of being laid over them afterwards.

The model returns the full mix, with voice and instruments together. Lyro also keeps your original vocal on its own track in the mixer. That lets you bring the voice forward, process it separately, or split the mix into stems if you want the instruments alone.

What you control

• The style: a genre, or your own description of drums, bass, chords and mood.
• The tempo and key, which Lyro detects and you can correct. Check any file with the free BPM and key finder.
• The vocal itself: the vocal tools include presets, pitch shift and time-stretch.
• The final balance, effects and loudness, in the mixer and mastering.

How to record a vocal that works

• Sing to a click or a simple guide loop in headphones, so the tempo stays steady.
• Record dry: no reverb, no echo, no music playing from speakers in the room.
• Stay the same distance from the microphone and avoid clipping, which is the harsh distortion you hear when the input is too loud.
• Leave a moment of silence before the first line, and start on the beat.
• Upload WAV if you can. A heavily compressed voice note gives the model less to work with.

Where the limits are

• Free-tempo singing, with no steady pulse, is hard to build a beat around. Expect a looser fit.
• Background music, room echo or noise in the recording confuses the detection and the model.
• Key detection can land on the relative key, and tempo on half or double the real value. Check both before you spend credits.
• In our small internal blind test ACE-Step 1.5 scored 3.00 out of 5 for production, below Lyria 3 Pro. Plan on a few attempts and some mixing.
• Long uploads take longer to analyse and generate.
• Phones and tablets have less memory than computers for large mixer projects.

Formats, credits and rights

A beat around your vocal costs one credit per started minute, with a minimum of 2 credits, and needs the Creator plan. Failed generations are refunded in credits automatically. WAV export is included on Creator, and downloads are never counted or capped.

Your recording stays yours: Lyro claims no rights in what you upload or in what you make. A vocal you wrote and sang is human-authored, which matters because the US Copyright Office says human-authored elements of a work can be protected while purely AI-generated material cannot. See music rights and the guide How to make a beat for your vocal.