Video & audio

Vertical talking-head video

A 40–90 second vertical script with captions, cut rhythm and a mix that passes a measured check.

Paid $15 list, $12 until 2026-12-31 · one skill, yours for good. The outline, the steps and the file open after purchase.

Background

One-sentence principle: give the result first, then the reason; change the picture every 5 seconds, one subtitle sentence per screen, speak at a talking pace.

Why it is built this way

Three numbers get fixed before a line is written, and they are chained: 1080×1920 with no letterboxing in the master, 45 to 90 seconds per piece, and a word ceiling of duration times 4.2 to 4.8 characters per second, so 45 seconds lands at 190 to 215 characters. Writing first and fitting the voiceover afterwards leaves the TTS one move, speeding up, and 5.5 to 6.4 characters per second reads as a newscaster. Cutting words costs a paragraph; speeding up costs the human feel of the whole piece.

The parts a script can measure go to check_delivery.py, and the rest is left to judgment on purpose. Speaking rate, one-line subtitles of 14 characters or fewer, a picture change at least every 5 seconds with no gap over 8, -16 LUFS at a true peak of -1.5 dBTP, 30 fps at 6 Mbps or more: those come back as FAIL with exit code 1. Content quality and human feel do not, which is why the counter-example table and the five-layer ceiling sit in the same document.

Where it bites

• Six scenes holding 30 seconds A three-minute piece with six change points reads as a slide deck, because the viewer counts new pictures and not runtime. The counter-example table asks for 2 to 4 images per scene at 10 to 12 seconds each, and check_delivery.py flags a longest gap over 8 seconds as a PPT.
• Landscape master cropped vertical Recording in 16:9 and publishing in 9:16 leaves the usable frame at about half the screen, so the opening second is wasted. The rule is to master in 9:16 and keep landscape for secondary distribution; the local engine only has a 16:9 template, so the vertical one gets added before rendering, never after.
• Colour blocks stretched over five seconds Colour blocks do satisfy the five-second change counter and still look empty. Picture cards come from Seedream through gen_visuals.py, 6 to 10 images per piece at a reference price of 0.2 CNY each, locked to 1080×1920 because the model defaults to 3:2, with the image count reported before generating.
• A sixth overlay layer The kit is five items: product identity card, UI screenshot card, two-line headline, subtitle capsule 120 px above the bottom safe area, and a CTA sticker. Five is the ceiling. A sixth layer always ends up covering the face, so the rule is to remove one first.

When to use it

Make 40–90 second vertical talking-head voiceover videos (live presenter / digital-human inset / all-graphics all work): fix aspect ratio, duration and word count → six-part structure → capsule subtitles → change the picture every 5 seconds → five-layer overlay kit → audio mix → quantified acceptance with scripts. Use when the user says 'make a talking-head video', 'turn this script into a vertical talking-head piece', 'a 45-second knowledge talking-head', 'digital-human talking-head', 'this talking-head looks too much like a PPT', 'how fast should I speak in a talking-head', or when producing a vertical final cut with the local video/ pipeline. Not for pure editing (clip-skills / video-editing), pure voiceover (local-audio), or long landscape tutorials.