Lyrics written for singing
One line per phrase, section tags such as [Verse 1] and [Chorus], and no line long enough to force the delivery to rush.
Three routes · voice creation is free
AI singing: three ways to get vocals on a track
You can let the model sing with its own voice, sing with a voice profile built from a short sample of yours, or cover a song you already have. All three run in the browser, and each generation returns two takes.
0.6 CNY per generation for 2 tracks · 1.2 CNY when singing with your own voice · downloads unlimited
Pick the route that matches what you want
| What you want | How | Cost |
|---|---|---|
| A song with vocals, any voice | Write a description or lyrics; the model sings it | 0.6 CNY per generation (2 tracks) |
| The vocal to sound like you | Create a voice profile, then generate with it | Voice creation free; 1.2 CNY per generation |
| An existing track with a different voice | Upload the audio and cover it, or reuse a track you already made | 0.6 CNY per generation |
What the voice sample needs
- 10 seconds or more, and cleaner is better than longer - no backing music, no reverb, no room noise.
- Natural delivery: talking or singing both work; reading single words does not.
- Under 10 MB, in a common format (mp3, wav, m4a).
- Yours, or licensed - a voice profile is a reusable asset, so the rights question is not optional.
Three things that decide how good it sounds
A style prompt that names a voice
Say male or female, half-sung or rapped, breathy or powerful. Vague prompts get a generic performance.
Patience with drift
On long tracks the voice slowly drifts - judge similarity from the first 60-90 seconds, and keep the take whose chorus lands.
Guides worth reading next
Voice cloning step by step
How much audio you need, and what each step produces.
Lyrics format
Section tags, line length and pronunciation problems.
AI cover songs
Keep the arrangement and change the voice.
FAQ
How do I get AI vocals on a song?
Two routes: generate the song with the model's own voice (write the lyrics or a description), or create a voice profile from a 10-second sample and sing with that voice instead.
Can it sing in my own voice?
Yes. Upload a clean 10-second-plus vocal sample, create the voice, then generate with it. Creating a voice is free; singing with it costs 1.2 CNY per run and returns two tracks.
What does a generation cost?
0.6 CNY per generation returns two tracks with the model's voice; 1.2 CNY per generation uses your own voice profile. Downloads are unlimited either way.
Can I use someone else's voice?
Only with permission. Cloning a voice you do not own or have rights to carries legal risk, and the responsibility sits with whoever uploaded it.
Start with one song
Sign-up takes an email address. Failed generations are refunded, and downloads are not metered.