Dialogue is where generated films most often break the spell. A face can drift a little and the viewer forgives it; a voice that changes mid-film, or a mouth that moves against the words, ends the illusion instantly. We have found that three decisions, made before any audio is generated, keep dialogue believable, and that one safety rule keeps the film alive when the sync provider does not cooperate.

Choose the voice before the face

Pick the voice in the character bible so every line of that character uses it. Switching voices mid-film is the audio equivalent of a face drifting, and it is more noticeable, because the ear compares voices across minutes while the eye compares faces across seconds. In Story to film the director gives each cast member a voice profile, gender, age and a tone in words, and the studio matches it against the voice catalogue while avoiding giving two characters the same voice. The narrator is your choice from the list in the wizard, or an automatic default. Give the narrator a voice that differs from every character in gender or age, so the ear separates the layers without effort.

Two layers, one rule

A story film has a narrator and it has characters, and they must not say the same thing. The narrator tells; the characters speak. The director keeps your dialogue verbatim, assigns each line to a cast member on screen, and writes narration as short bridges between events. If a scene has a line, its narration is short or absent. If the story has a line, the narrator does not paraphrase it. A film where every scene has both layers sounds like a documentary about itself.

Short lines, clear consonants

Lip sync works on visible mouth shapes. Lines with hard consonants and pauses sync better than long murmured sentences. Rewrite for the mouth, not just for the ear. Lines are capped at one sentence per scene; longer speeches are split across scenes. Narration is budgeted at about nine characters per second, so a ten-second scene gets one clear sentence. Avoid numbers and abbreviations in narration; speech models read them inconsistently, and a written-out form is safer. Proper names should appear in the character's written look so the model has seen them before it has to say them.

Which sync tier

The standard tier syncs a clip to generated speech and is fine for medium and wide shots. The pro tier takes your own audio and is worth it for close-ups, where the viewer watches the mouth. Skip sync entirely on wide shots where the mouth is not visible; the line can play in the audio track with no loss, and you save the operation.

Bring your own recording

If you have an actor, record clean audio and upload it as an asset. Pro sync uses it directly, and the performance is yours rather than synthesized. Record in a quiet room, keep the take to the line, and leave a short silence at each end. The sync step needs words, not room tone.

When sync is unavailable

Lip sync depends on a video provider, and providers fail: quota, outage, a removed endpoint. We learned this during a test in which the provider ran out of balance halfway through a film. The rule we adopted: the film is still finished. The scene keeps its animation and the line is voiced with the character's voice in the audio track. You lose the mouth movement, not the words, and the failed sync operation is refunded to the wallet. An audio failure never discards a finished clip.

Placement and mixing

Narration clips are placed at the start of their own scenes on the timeline, not concatenated into one long file, so the narrator never drifts against the cuts. Lines go where the character speaks, which for a synced clip is the start of the clip. A short gap between narration and line is better than overlap. Music sits at a fixed lower level under everything and dips under speech automatically, so you do not mix by ear.

Changing a voice later

  • Narrator: chosen in the wizard before the film is created.
  • A character's line: open the scene card, edit the spoken text or pick another voice, regenerate the scene. Only that scene is charged.
  • A character's default voice: change it in the bible, and every future scene uses it.

Frequently asked

Can the narrator be a character? Yes, in first-person stories. The narrator voice is then that character's voice, and the two-layer rule still applies: narration describes, lines act.

How do I know the speech matches the script? Listen once with the picture off. Mispronounced names and a narrator that outruns its scene become obvious when the image is not there to mask them.

Does a language the voice catalogue covers poorly still work? Yes. Cast the narrator from the best available voice for the language and keep character lines short; weak voices are most exposed on long narration.