Remake: How Spoken Lines and Music Are Read From the Clip, and What Ends Up in Your Episode

Quick answer: Every shot of the clip is listened to twice: first by a speech recogniser that answers ‘is anyone speaking, and what did they say’, then by the model that describes the shot, which is given the recognised words to work from. Lines are therefore quoted, not guessed from lip movement, and a shot where nobody speaks is written as silent. If the clip has music, the read notes its character and every clip you generate carries the same short music brief, the way ads do. The dialogue language is chosen on the upload screen and fixed once the read starts: lines are written in it, translated if the source speaks another. A line the recogniser could not make out is simply missing; compare with your clip and add it to the shot in the editor once the storyboard is generated.

Still from “Neon Debt”, an AI short drama generated end-to-end by SceneMixer (7 characters, 1 episode) — Sample: “Neon Debt” — Game Engine Realism, 1 episode, produced end-to-end by SceneMixer from the script · watch the episodes

After a read, open the episode in Step 3 and read its outline: each scene block lists its lines in speaker-colon form.

How lines are heard

Speech recognition runs on each shot’s own audio. What it hears is handed to the shot description as ‘these words were spoken here’; the describer then names the speaker from the faces and writes the line into the shot. Music under speech is handled; shouting over a crowd, heavy accents and very quiet lines are where it fails, and those are left out rather than invented. Lines then appear in the outline like a script’s: Write Dialogue So It Is Picked Up.

A storyboard segment: location, time and weather fields, then numbered shots with durations, asset chips and spoken lines — Each shot is listened to; what was heard is quoted into the read and ends up as lines in the outline.

Why quote, not guess

Left to itself, a vision model asked ‘what is said here’ will sometimes hear and sometimes not, on the same shot, and will invent plausible lines when unsure. Recognised words pinned into the description stop both; the same words are used every time the shot is described.

Music

The read asks once whether the clip has music and what kind. If it does, each segment you generate carries a one-line music brief matching it, so the generated segments have music of the same character; the source track itself is never used. Ambient sound is noted but not carried. Where the Music Comes From.

Dialogue language

Chosen on the upload screen, defaulting to your interface language (the Chinese site is always Chinese), and fixed once the read starts. Lines heard in another language are translated into it. It cannot be changed afterwards for a remake, which is why Step 1 does not offer the dropdown. After the Read.

When a line was missed or wrong

The outline itself cannot be edited, and the storyboard copies its lines word for word. Generate the storyboard, then open the segment where the line belongs and add or fix it there, with its speaker; a line given to the wrong person is fixed the same way. Editing is free. Generate the Storyboard.

In the clip	In the read	In your episode
A clear spoken line	Quoted, speaker named	A line in the outline
Nobody speaks	Shot marked silent	No line
Unclear speech	Left out	Add it in the segment by hand
Music	Character noted	Same-character music in every segment
Ambient sound	Noted	Not carried; the storyboard writes its own

FAQ

Why is a line missing?

The recogniser could not make it out. Compare with your clip and add it in the segment once the storyboard exists.

Are the source voices used?

No. Voices come from the cast's voice settings, as in any project.

Can the remake speak a different language than the source?

Yes; pick the dialogue language on the upload screen and the lines are translated at the read.

Is the source music reused?

No. Its character is described and the segments generate their own.

Why does a crowd shout get no line?

Group noise is treated as background sound, not dialogue.

Heard, quoted, written

What was said in the clip is what your episode says.

Start a remake

Updated September 27, 2026