Katechon ArcTwo kinds of voice, two methods
| Voice | Method | Where we used it |
|---|---|---|
| On-screen dialogue | Written into the shot description; the video model generates speech and lip movement with the picture | Every drama episode, e.g. Blood Sun and High Note |
| Narration | Text-to-speech (ElevenLabs) placed on shots with no one speaking | The Wolf Mother, RatKid and the Red Balloon |
| A recurring off-screen voice | Text-to-speech in a post pass, mixed under the clip audio | The king's voice and a ritual chant in The Fall of Kalak |
Dialogue inside the shot
Native-audio video models can speak a short line with matching mouth movement. Keep each shot to one or two short lines, name the speaker clearly and keep them in frame. In The Emperor's Pleasure Court Seedance 2.0 Fast delivered dialogue essentially word for word. Long speeches split across shots; a clip of 4 to 15 seconds holds a sentence or two comfortably.
Narration on top
For RatKid and the Red Balloon we wrote 20 narrator lines and placed each on a shot with no dialogue, so the narrator never talks over a character. The Wolf Mother is told by an older witness; naming who tells the story gave the long film a frame that carried its time jumps.
Checklist
- Write the exact line in the shot, with the speaker named and visible.
- Leave shots that will carry narration free of dialogue.
- Generate narration as separate audio and mix it under or over the clip sound in the edit.
- Listen to every clip for voices you did not ask for; see why that happens in the related answer below.
- Caption from the final mix, not from the script.
Dedicated lip-sync tools that animate a still face to an audio file are a different job: talking heads and avatars. For drama, generating the line with the shot keeps acting, framing and voice together.
Where Katechon Arc fits
Arc writes the dialogue into each shot so Seedance generates speech with the picture, and generates narration and voice with ElevenLabs in the same project.
Arc has no long-form talking-avatar model and no tool to lip-sync an existing video to new audio; use a dedicated lip-sync tool for that.
People also ask
Can AI video generators do lip sync?
Models with native audio, such as Seedance 2.0, generate speech and matching mouth movement for short lines inside a shot. Syncing an existing video to new audio needs a separate lip-sync tool.