AI Lip Sync for Music Videos: What Works, Singing Shots and Costs
AI lip sync makes a person, character or animal in your video sing your song's actual words on camera. In Melodious you switch it on in the brief when your song has vocals, choose who sings, and up to four chorus close-ups are marked as singing shots. Each singing shot costs 2.2 times a normal shot to render, singing shots are made at 1080p at most, and you can test one line before paying for the full video. If a shot can't be lip-synced, it is rendered as a normal shot at the normal price.
Your first 30-second storyboard is free. Plans from $19/month.
What is AI lip sync in a music video?
The short answer: AI lip sync makes a person, character or animal in your video sing your song's actual words on camera. In Melodious you switch it on in the brief when your song has vocals, choose who sings, and up to four chorus close-ups are marked as singing shots. Each singing shot costs 2.2 times a normal shot to render, singing shots are made at 1080p at most, and you can test one line before paying for the full video. If a shot can't be lip-synced, it is rendered as a normal shot at the normal price.
Without lip sync, a performer's mouth moves with no relation to the vocal. With it, the mouth follows the singing. It is also the most expensive thing in the studio and the most likely to look wrong, so the flow around it is built to limit your exposure. Here is how it works, in order.
For where this sits in the whole process, how to make an AI music video is the overview.
How do you turn lip sync on?
In the brief, not in the prompt box. When the analysis shows your song has vocals, the brief offers a Lip sync switch: "Your song has vocals. Someone in the video can sing the lyrics on a few chorus close-ups." For an instrumental track it is never shown. It is off until you turn it on.
Turn it on and it asks Who sings?
- The subject of your idea. If your sentence was 'a dog in a tree', the dog.
- You. Add a photo; the subject of the video stays in the story.
- Someone else. A saved character, or a short description of a singer.
If your singer needs to look the same across the whole video, save them as a character first. Reusable AI characters for music videos shows how, or add your own photo of the singer in the brief.
Which shots sing?
The director marks up to four chorus shots as singing shots and writes them as front-facing close-ups of the singer. The board shows a Sings badge and the lyric line on each one. You can move a mark to another shot or turn it off.
Two design choices explain the limits:
- Chorus lines only. Verses stay off-mic. Dense verses are where mouth shapes are most likely to drift from the words, so the default spends the budget on the lines viewers remember.
- Close-ups, facing the camera. The lip-sync model redraws the frame around the singer's face. A wide shot has a face too small to animate, so a singing shot needs the mouth in view.
That second point is enforced. Before a singing shot is priced, its keyframe is checked for a visible mouth. If it is a close-up of eyes, or the singer is seen from behind, the shot is made and charged as a normal shot, and the board says so: redraw it as a close-up to lip-sync it. You are never charged the lip-sync price for a shot with no lips in it.
What does it cost?
A singing shot costs 2.2 times a normal shot's render price. At 720p:
| Shot length | Normal shot | Singing shot |
|---|---|---|
| 4 seconds | 74 credits | 162 credits |
| 8 seconds | 148 credits | 324 credits |
Keyframe drawing is priced as usual. The quote you see before you render, and on Test this line, already includes the multiplier, and the quote, the reservation and the charge are one number.
The reason is the model's cost: lip sync runs on a separate, more expensive model (ByteDance's OmniHuman 1.5) than a normal shot, so a singing shot is priced at 2.2 times.
Put that against a plan. Four 4-second singing shots at 720p come to 648 credits on their own, close to Starter's 750 monthly credits, before any other shot is rendered. A small number of short singing shots is the realistic use. Starter is $19 a month (750 credits), Artist is $97 (5,200) and Label is $147 (8,500); how much an AI music video costs has the wider picture.
Why 1080p at most?
The lip-sync model, OmniHuman 1.5, outputs 1080p. If you render at 4K, the singing shots are still made at 1080p, and they are charged at the 1080p rate, not 4K. The 4K option states this on its own label, so you are not billed 4K for pixels that do not exist. The rest of the video is made at 4K.
Can you test it first?
Yes, and you should. Test this line renders one singing shot alone so you can check the singer before paying for the whole video. It is priced like any singing shot, about three minutes of waiting, and the clip is reused in the full render rather than charged twice. The full render reuses it only if the test was really lip-synced; if the test fell back to a normal clip, the full render makes the shot again at the lip-sync price.
If a singing shot can't be made
It is rendered as a normal shot at the normal price, and the difference is returned to your credits. You never pay the multiplier for a clip that is not lip-synced.
Related reading: make a photo sing your song, using your own photo in an AI music video.
What makes AI lip sync work best?
From our first model test on 27 September 2026, one 5-second chorus line from one song, tried on a person, a stylised character and a dog:
- A clear, frontal close-up. The stylised character gave the clearest mouth shapes. The dog held steady and faced the camera in quality mode with a singing prompt, and turned away in fast mode.
- A short, punchy line. One chorus line is a good test: short enough to judge, long enough to show drift.
- One singer. The brief asks for one singer, which keeps the four shots consistent.
This is three clips, not a benchmark, and results vary by song and image. Treat Test this line as the real answer for your own track.
If you are still deciding whether lip sync is worth it, a performance video without it is a legitimate choice. How to choose a music video style is the broader decision.
Frequently asked questions
Can any character sing, or only people?
Any subject can be chosen as the singer, animals included. In the brief, 'Who sings?' offers the subject from your own idea (a dog can sing), you with a photo you add, or someone else, which can be a saved character or a short description. A shot only becomes a singing shot if the singer's face and mouth are actually visible in its keyframe.
How many shots can be lip-synced?
Up to four per render. Melodious picks chorus shots by default, and you can move a singing mark to a different shot or turn any off. Verses stay off-mic, and a fifth marked shot would be priced and rendered as a normal shot.
How much more does a lip-synced shot cost?
2.2 times a normal shot's render price. At 720p an 8-second singing shot is 324 credits against 148 for a normal one, and a 4-second one is 162 against 74. Keyframe prices are unchanged, and the price shown before you render already includes the multiplier.
Why is there a 1080p cap?
The lip-sync model produces 1080p at most. On a 4K render, singing shots are made at 1080p and charged at the 1080p rate rather than the 4K rate, and the 4K option says so. The rest of the video is made at the resolution you chose.
What happens if the singer's mouth isn't visible?
The shot is not lip-synced and is not charged the lip-sync price. Before pricing, each singing shot's picture is checked for a visible mouth; if there isn't one, it is rendered as a normal shot and the storyboard tells you to redraw it as a close-up if you want it to sing.
Test one line before you commit
Turn on lip sync in your brief, pick who sings, and test a single chorus line first.
Try lip sync