Musicians use it for visualisers, lyric-video intros, cover teasers and fan content; brands use it for jingles and mascot songs. Here is how to get a performance that holds up for a full chorus.
Preparing your track for an AI singing avatar
The cleaner the vocal, the better the mouth follows it. If you have stems, upload the lead vocal alone or with the instrumental turned well down, then lay the full mix back over the rendered video in your editor. Heavy reverb, stacked harmonies and auto-tune artefacts make syllables harder to read.
Clips can run from 4 seconds to 5 minutes, so a full verse and chorus fits into one take. For social teasers, 15–30 seconds around the hook is usually the sweet spot.
Choosing the singer’s photo
Pick an image with the mouth clearly visible and some space around the head and shoulders. Stage photos with a microphone in front of the mouth reduce quality; a mic held lower, at chest height, works fine. Album-art illustrations, 3D characters and band photos all animate — a group shot can become a duet or a choir line.
Directing the performance
Describe movement and feeling, not mouth shapes:
- Ballad: “eyes closed on long notes, hand on chest in the chorus, slow orbit”.
- Upbeat pop: “bobs to the beat, smiles, points at the camera on the hook”.
- Rap: “confident, sharp head movements, hand gestures on the rhymes, handheld feel”.
- Duet: “left singer leads the verse, right singer joins on the chorus, static wide shot”.
Music rights and platforms
Generating video doesn’t change who owns the music. Use your own songs, licensed tracks or properly cleared covers, and follow each platform’s rules on synthetic media labels. Animating a real artist’s face without permission is not allowed under our terms.
Finishing the video in your editor
Once the render is done, lay the full mix under the clip and line it up with the vocal you uploaded — because both come from the same session, they align from the first frame. Add lyrics, a logo or a colour grade as you would with any performance footage.
For a full music video, render several angles from different photos of the same artist and cut between them on the beat. Each render uses the same audio, so every angle stays in sync.
Formats for social platforms
Vertical photos produce vertical clips for Reels, Shorts and TikTok; square images suit feed posts and Spotify Canvas-style loops; wide shots fit YouTube. Render a 15–30 second hook for social first, then the full song once the look is approved.
FAQ
Can an AI singing avatar handle fast rap?
Yes, with a clean vocal. Very dense, heavily processed vocals reduce accuracy.
Should I upload the full mix or the vocal only?
A vocal-forward file gives the best sync. Add the full mix back in your editor afterwards.
How long can the singing video be?
Up to 5 minutes in one take, minimum 4 seconds.
Can two people sing together?
Yes. Use a photo with both singers and describe who sings which part in the prompt.
Can I use a cartoon character?
Yes. Illustrated and 3D characters with a clear face sing convincingly.
Who owns the music rights?
The original rights holders. Only use songs you own or are licensed to use.
Can I use an instrumental-only section?
Without vocals there is nothing to sing, so the face will mostly stay calm and move to the rhythm. Keep instrumental gaps short.
What format should the song be?
MP3 or WAV between 4 seconds and 5 minutes, under 15 MB.
Will the avatar dance?
It moves the head, shoulders and upper body with the rhythm. Full choreography with footwork isn’t what the model is designed for.