Add AI Vocals to Instrumental Tracks

Upload an instrumental, paste the lyrics you want sung, and an AI singer performs those words over your music. What comes back is a new track with the vocal and the backing together, and every run returns two versions, so you can keep the one that sits best on the beat. You bring the backing and the words; the melody and the delivery are generated. It is the mirror image of Add Instrumental, which builds a band under a vocal you already have.

Two things are required: the instrumental audio file you upload, and the lyrics you want sung. The lyrics box arrives with an example in it that you overwrite, and if you clear it the job is refused rather than sent with blank words — nothing is invented behind your back on a track you paid for. Four more controls sit in the studio's Advanced panel and are optional, each pre-filled so a run works untouched — a style description for how the vocal should sound, negative tags for what to avoid, a title, and the vocal gender. Add Vocals takes an audio upload: you are providing the instrumental the singer performs over, not selecting a library track. The cleaner and steadier that instrumental is, the steadier the result.

20 credits per run, and each run returns two sung versions. Credits are taken when the job starts and go back to your balance if the run fails. Every engine in the studio picker costs the same.

The engine picker in the Add Vocals studio lists only the engines this job can be sent to, and it opens on a default, so you can run without choosing one. Older generations are not held back as a premium tier; this job is simply not available on them. Every engine in the picker costs the same 20 credits, so pick by sound and compare the two versions each run returns.

Negative tags tell the AI what to keep out of the performance, and the endpoint behind Add Vocals will not accept a blank one — so a value is always sent with your job. You do not have to type it: the studio pre-fills the box, and a safe default is substituted if you clear it, so a run never fails for an empty field. It is still the field most worth a moment of your time, because a vocal has many more ways to go wrong than a backing track does. Practical negative tags include "off-key, robotic, monotone, autotuned, shouty" to steer toward a natural, musical delivery. Your positive style description sets the target and the negative tags fence off the failure modes, which is often the difference between a take that feels human and one that feels synthetic.

Yes. The vocal gender control sits on Auto, where the engine picks; set it to Male or Female and that setting is sent with the job to steer the voice. The style line steers it too, so set both: "female, breathy, intimate" reads very differently from "female, belted, powerful". If a result sits in an awkward range for your instrumental, change the gender setting or the style words and run again.

Write the lyrics as you would sing them — line breaks matter, because they hint at phrasing, and shorter, singable lines usually deliver better than dense paragraphs. For the style, be concrete about the voice and the emotion: "male, soulful, restrained verses, big open chorus" gives the model a clearer target than "good singing". If a section feels rushed or crammed, trim words from that line; if it feels empty, add a short phrase. Treat the lyrics and style as things you refine between runs rather than getting perfect on the first attempt.

MP3, WAV, OGG, M4A, FLAC, AAC, up to 200MB and 8 minutes, one file per run. Your own recordings and demos work. Some commercial tracks may be refused. For the steadiest result, upload a clear, well-mixed, instrumental-only file that holds its tempo. The whole file is sent as it is, so a scratch vocal or bleed left on it goes with it.

Your own recordings and demos work. Some commercial tracks may be refused. A bought or leased beat can still be refused, because the check compares your file with existing recordings, not with your license for it. If a file matches an existing recording, it is refused and the 20 credits go back to your balance. The commercial rights on the result cover what the tool generates; they do not clear someone else's recording.

They solve different problems. Generate Music writes the whole song — instrumental and vocal — from a text prompt, so you do not control the exact backing. Add Vocals starts from the instrumental you upload and writes only the topline over it, which is what you want when you have already made a beat you like and just need it sung. Use this table to pick: Aspect · Add Vocals · Generate Music. You provide · An instrumental you upload · A text prompt only. AI generates · Only the vocal, over your track · The whole song, backing and vocal. Versions per run · 2 · 2. Cost · 20 credits · 20 credits.

It is not instant. The job runs in the background and the two sung versions appear in your Music Factory library when they are ready. You do not need to keep the tab open, and you can start a second run with rewritten lyrics while the first is still processing.

It is a normal track in your Music Factory library: preview it, download it, or open it in Separate Vocals to get the sung part and the backing as separate files — 20 credits for a vocal/instrumental split, 30 credits for a full stem split. If one line needs different words, change it in the lyrics box and run again; each run returns two fresh versions.

Yes. All Music Factory outputs include commercial usage rights, and that includes an instrumental you have topped with an AI vocal. The rights cover the material the tool generates: if the instrumental you upload contains someone else's recording, adding vocals to it does not give you rights to that original. Credits are consumed at the moment you start the job.

Turning an instrumental you already have into a song with words you wrote. Producers use it to hear a topline over a finished beat before booking a singer; songwriters use it to check whether a lyric scans once it is sung, which is not the same as whether it reads well; creators use it for a sung intro or a jingle over a music bed they made. Because you choose the backing and write every word, it suits songs that need specific lyrics over specific music: a birthday song with a name in it, a channel intro, or a hook for a beat that stalled without one.

Timing can vary between the two versions, and when a vocal sits off the beat the cause is most often the upload. A loose, drifting or heavily processed instrumental, or a busy mix with a scratch vocal still on it, makes the beat harder to follow. The fixes are practical: upload a cleaner, instrumental-only version that is steady in tempo; make sure the file is not clipped or very quiet; and keep your lyric lines singable rather than crammed, since dense lines force the singer to rush syllables. If one line still feels rushed, trimming a couple of words from it usually settles it.

Not from an Add Vocals track. A reusable singer can be saved only from a track made with Generate Music or Extend. For a one-off song, keep the same gender setting and the same style words each run to keep the delivery close. For a voice you can reuse, make a track you like with Generate Music, save it as a singer, and use that singer when you generate or extend.

Not in Add Vocals. This tool takes your instrumental, your lyrics, a style line, negative tags, a title and a male or female setting, and it has no singer picker. A voice you reuse across songs is saved as a singer from a track made with Generate Music or Extend, and used when you generate or extend. A track made with Add Vocals can't be saved as a singer.

It depends on how much the exact backing matters. If you already have an instrumental you like — one you produced — Add Vocals is the right tool, because it starts from that backing and writes only a topline over it. If the instrumental would come from Generate Music anyway, generating the song with vocals there is the simpler route: one run produces the backing and the vocal together. Either way each run returns two versions, so you can compare before you commit.