Week 11: Production, Week Two
Finishing an episode — assembling the queue's shots into one cut, sound and captions, loudness standards, and the delivery spec a platform actually expects.
17 May 2027
The queue produces shots; it does not produce an episode. Assembly is where every shot that cleared the quality gate gets cut into one sequence, and this lecture is entirely about the three things that turn a folder of approved clips into a piece someone can actually watch: the assembly itself, the sound and captions carried through it, and the delivery spec that gets it onto a platform without the platform mangling it on the way in.
The gate has already run#
A shot that passed the quality gate does not get re-judged during assembly on its own merits; assembly only decides order, pacing, and where a cut lands relative to a beat. Reopening a gate decision at this stage means the gate wasn’t doing its job earlier, and it’s usually faster to flag that for next time than to relitigate one shot now.
The Cut is exactly this decision at a small enough scale to try by hand: four gated clips, reordered, nothing about any single shot re-judged in the process.
Sound and captions, carried through#
Dialogue clarity and captions are not a separate pass bolted on at the end; they’re properties every shot in the queue was already supposed to carry, and assembly’s job is making sure nothing got lost matching cuts against each other. A caption that’s present but out of sync, or a dialogue line that reads clearly in isolation but gets buried once music is layered under it, is an assembly failure even though no single shot caused it alone.
Loudness has a published number#
Loudness gets measured against a published broadcast standard rather than judged by ear, because “sounds about right” varies between speakers, rooms, and the person listening. Matching a target loudness level means an episode won’t clip on a phone speaker or vanish next to whatever plays before or after it in a feed — a technical requirement this course treats the same way it treats a licence clause: non-negotiable, and checked, not assumed.
Delivery is a spec, not a save button#
Exporting a finished cut means matching whatever a platform’s delivery spec actually asks for — resolution, frame rate, container format, caption format — rather than exporting on default settings and hoping it’s close enough. A cut that looks correct on your own machine and fails a platform’s delivery check has not shipped; it’s failed at the very last step of a nine-week process, which is the most expensive place in the whole pipeline to fail.
This is the spec the pilot is delivered against, and it has two kinds of row in it. The container, frame rate, bitrate and vertical-frame rows are published by the platform that enforces them. The audio rows are not: −23.0 LUFS and −1 dBTP come from EBU R 128, which is a broadcast recommendation, and none of the three platform specs below asks for a loudness target at all. That part of the spec is this course’s own rule, chosen rather than found — because a level measured against ITU-R BS.1770 gives the same answer in any room, and “sounds about right” gives a different one to every listener. A number a marker can measure is worth writing down; an impression is not.
| Requirement | The spec |
|---|---|
| Programme loudness | Normalised to −23.0 LUFS, with a ±1.0 LU tolerance where the target is not practically achievable |
| True peak | Not above −1 dBTP |
| Meter | Compliant with ITU-R BS.1770, currently BS.1770-5 |
| Container and codecs, YouTube | MP4 with the moov atom first; H.264 High Profile, closed GOP, 4:2:0; AAC-LC at 48 kHz |
| Frame rate, YouTube | Uploaded at the rate it was shot — 24, 25, 30, 48, 50 or 60 |
| Bitrate, YouTube 1080p SDR | 8 Mbps at 24–30 fps, 12 Mbps at 48–60 fps |
| Vertical frame, TikTok ads spec | 9:16 at 540 × 960 or larger; .mp4, .mov, .mpeg, .3gp or .avi; under 500 MB; at least 516 kbps; up to 10 minutes |
| Vertical frame, Douyin upload API | Under 128 MB and under 15 minutes; mp4 or webm recommended; 720p or better, vertical |
Source: EBU R 128 v5.0 — https://tech.ebu.ch/publications/r128
Source: ITU-R BS.1770 — https://www.itu.int/rec/R-REC-BS.1770
Source: YouTube, Recommended upload encoding settings — https://support.google.com/youtube/answer/1722171
Source: TikTok Ads Manager, Video Ads Specifications — https://ads.tiktok.com/help/article/video-ads-specifications
Source: Douyin open platform, upload-video API documentation — https://open.douyin.com/platform/resource/docs/openapi/video-management/douyin/create/upload/
Before class#
Bring the three shots that cleared the quality gate together with any notes on where their sound or captions still need work before assembly.
This week’s exercise#
Assemble the season’s episode: cut the gated shots into one sequence, check loudness against the published standard, confirm captions are present and in sync, and export to the platform’s delivery spec. This is the pilot’s own production pass, so bring the finished pilot itself to Wednesday’s Dailies.
Reading#
EBU R 128: Loudness normalisation
European Broadcasting Union2023
Programme Loudness Level shall be normalised to a Target Level of −23.0 LUFS
W3C Web Accessibility Initiative
Captions are a text version of the speech and non-speech audio information needed to understand the content
YouTube Help: recommended upload encoding settings
YouTube
the player automatically adapts
抖音开放平台
视频文件大小不超过128M
FFmpeg
The following documentation is regenerated nightly, and corresponds to the newest FFmpeg revision.
Teaching team
- Vera Lin — Convener