Voice as one continuous track, pictures on top, and the one-frame length check
The assembly rule for talking films: cut the audio from the same pieces as the picture, snapped to frames, then verify film length equals voice length. Drift over one frame is a bug, not a quirk.
VOICE TRACK ASSEMBLY, the checklist 1. Build the voice track first: every kept sentence, in order, cut from the source audio at frame boundaries (quantise in and out points to the frame rate: multiples of 1/30 s at 30 fps). 2. Lay the picture on top of it: the speaker's own shot where lips are visible, cutaways in the silent windows. The picture never decides where a sentence starts. 3. Keep the join inside silence: cut points at the ends of lines, trimmed of filler; mid-line only when the gap is silent. 4. After the render, measure, do not watch: `ffprobe -v error -show_entries stream=codec_type,duration -of csv=p=0 film.mp4` and compare the video and audio stream durations to the planned length. 5. Drift larger than one frame (1/30 s at 30 fps, about 33 ms) anywhere is a bug. Find it: a clip with a different sample rate, a variable-frame-rate phone recording, a cut not on a frame boundary. 6. Check the delivered file itself, at the seams, on at least one real camera file, never only on synthetic test sources.
In Claude Code, paste the prompt as a message, or run /video-edit which does this end to end.
Steps
- Resample everything to one sample rate (48 kHz) before assembly. A 96 kHz recording joined to a 48 kHz card drifted the audio clock 1.39 s at the seam and passed every synthetic test.
- Snap every cut to a frame: round in/out times to the frame rate before cutting the audio, so the audio piece and the picture piece are the same length.
- Compare planned length, video stream duration and audio stream duration after every render. Three numbers, one line, every time.
Tips
- Cutting the audio from the picture pieces is what a timeline editor does for you; a script-driven render has to do it on purpose, and that is where the drift comes from.
- A film that is half a second shorter than its voice track plays as a cut-off last word on the platform, and nobody tells you.
| works with | Claude Code: Runs as a skill or a prompt in the terminal; can read your files and write the result. · ffmpeg: A terminal command you run yourself. · CapCut / any editor: A manual recipe for any editor. |
| skill | npx skills add fortunto2/solo-factory/video-edit · source |
| cost | Free. No API, no key. |
| status | Unverified: written from the technique, not yet run end to end by us. |
| license | CC BY 4.0 for the text here; skills are MIT. |
| updated | 2026-11-01 |
| for agents | voice-as-one-continuous-track-length-check.md |
Related
Write the cut as a script, not a timeline: the JSON storyboard prompt
An agent prompt that writes the edit of real footage as a small JSON file: which sentences, which cover shots, what the cold open is. A complaint becomes a one-line edit, and the render repeats.
Read the film as text before rendering: the last check that costs nothing
The cut, printed as its lines in order, read by someone who never saw the raw footage. Catches the question whose answer was cut, the name said before the person, the 'this' that points at nothing.
The cutaway rule: what is said, same place and time, never over moving lips
Three constraints for every cutaway in a talking film, and the agent prompt that applies them to a transcript. Kills what viewers read as broken: a picture from elsewhere, a face out of sync.
The critic pass: repeats and sagging scenes first, mid-word joins second
A two-pass review prompt for a subagent or a second session: pass one on structure (repeats, sagging scenes), pass two on joins (mid-word cuts, audio seams). One read finds one kind of problem.
A contact sheet of every cutaway with its line under it
The one-page check that catches off-topic cutaways before anyone watches: every cutaway frame in a grid, its sentence under it. Measured once: 12 of 54 automatic cutaways were off topic.
The day as one text file: an agent reads the speech and proposes the story
The step before any timeline: the day's speech in shooting order, numbered by clip and sentence, read by an agent like a scriptwriter. It proposes story, scenes, hook and cutaways; you argue in text.