How to Add a Talking AI Character in the Corner of a Video, in Any Language

To add a talking AI character in the corner of a video, turn on Speaking character in ReelMint's Mint Studio, pick the character (or any photo), type what they say or attach your own audio, and choose the corner. After the video is made, the character's face is lip-synced to the line, read in their own voice, and placed in that corner for as long as they speak, over the video's own sound.

We build ReelMint. A narrator in the corner is the format of explainers, reaction clips and story channels, and it is how we get speech in languages video models handle badly. This post covers the steps, the options, how it is priced and where it falls short.

Why a narrator is made separately from the video

Most video models cannot speak every language: Arabic, for example, comes out refused or mangled. A lip-sync model does not speak at all; it moves a face to whatever audio it is given. So the speaking character is its own small production, run after the video exists:

1. the line is read in the character's voice, their cloned or assigned voice, in any language that voice reads, or your own audio track is used as it is;
2. a talking-head model animates the character's picture to that audio;
3. the head is cut to a circle or a square, ringed, placed in the corner you chose for exactly as long as it speaks, and the voice is mixed over the video with the video's own sound turned down beneath it.

Step 1: make the video as usual

The speaking character works on text to video, image to video and text to anime in Mint Studio. Write the prompt, pick the model and the length as you normally would (how to turn text into a video). Make the video at least as long as the line, because the character only talks while there is video under it. The studio can add scenes that continue your prompt to make room; the price on Generate updates.

Step 2: open "Speaking character"

It is an optional step in the composer: "a face in the corner reading a line in their voice". Choose who speaks:

• A character from your Character Studio. It needs a picture; if it has none yet, generate or upload one on the Characters page first. See how to create reusable characters.
• A photo from your library, which overrides the character's picture.

Step 3: choose what they say

• Type the line. Up to 1,500 characters, about two minutes of speech. It is read in the character's own voice; with no character voice, a default voice reads it, and you can pick another. Set the language, or leave it on auto.
• Or attach audio from your library, such as your own recording or a cloned-voice read made in the voice studio (how to clone your voice). The face is synced to that track instead.

The line is checked by the content screen along with the prompt.

Step 4: place it

• Corner: bottom right (default), bottom left, top right or top left.
• Size: small, medium (default) or large.
• Shape: circle (default) or square.
• Start time: how many seconds into the video the speaker appears.

Step 5: generate and read the price

The speaking character is quoted as its own line on top of the video's price, and the button shows the total. The length of the speech is fixed when you press Generate, from your audio track's real length or, for typed text, an estimate of about 13 characters a second plus a breath, so the charge matches the quote.

How it is priced: per second of speech, plus a small charge for reading typed text aloud. Lines of 15 seconds or less are filmed on InfiniTalk at 480p, which is the cheaper path; the corner is a third of the frame, so 480p is more than it shows. Longer lines are filmed on Kling AI Avatar, at more than three times the per-second price. If you want to keep the cost down, keep the narrator's line short. The narrator is charged once, when the talking head is produced; if it fails, the video ships without the corner and that part of the hold comes back.

Long lines are filmed in pieces

Every lip-sync model caps how much audio one generation takes. A longer line is cut at its natural pauses, each piece is filmed, and the pieces are joined with the original audio laid over the whole, so the timing never slips. The price stays per second of audio.

Honest limits

• The line cannot be longer than the video. Lengthen the video or shorten the line.
• It is a head in a corner, not a full presenter. For a whole speaking video of one face, use talking photo instead.
• Pick a clear, front-facing picture. The lip-sync is only as good as the face it is given.
• A multi-scene film that is re-merged later reuses the head it already paid for; it is not filmed twice.

Plans start at Lite, $9.99 a month for 800 credits, and every plan runs every model.

Can I put a talking character in the corner of an AI video?

Yes. In Mint Studio, turn on "Speaking character", pick the character or a photo, type the line or attach audio, and choose the corner, size and shape.

Can the narrator speak Arabic or other languages?

Yes. The line is read in the character's voice in any language that voice reads, Arabic included, and the face is lip-synced to that audio.

How long can the narrator talk?

A typed line can be up to 1,500 characters, about two minutes, and never longer than the video itself.

Can I use my own voice for the narrator?

Yes. Attach your own recording from the library, or give the character your cloned voice.

Is the narrator charged if it fails?

No. If the talking head cannot be made, the video ships without the corner and the narrator's share of the hold is returned.