top of page

The Master Guide / 14 Finishing and delivery · Graphics, captions, versions and delivery

Exports and the delivery check

Every file is exported from its own cut, then measured by tools and checked by eye before it leaves.

Every file is exported from its own cut, then measured by tools and checked by eye before it leaves.

The four kinds of export

Profile

Use

Picture

Audio

Master

One per version, texted and textless

MOV ProRes 422 HQ, the sequence's size and rate (24p; 25p where the plan set a 25 fps master for filmed footage), BT.709

PCM 24-bit 48 kHz stereo; −16 LUFS for the online master

Social

Meta, TikTok, YouTube, X, LinkedIn

MP4 H.264 High, yuv420p, a closed GOP of half the frame rate, about 18 Mb/s, native 24 fps, BT.709

AAC 320 kb/s; −16 LUFS, or −14 for a short social ad; true peak aimed at −1.5 dBTP on AAC

TV

MENA TV

MOV ProRes 422 HQ 1920×1080 from the 25 fps TV cut; interlaced, upper field first (steps below, command 1); captions burned in; no freeze frames where the channel refuses them

PCM 24-bit on channels 1–2; −23.0 LUFS, −1 dBTP

Review

Previews only

H.264 CRF 18; the review label on rough cuts only

AAC; −16 LUFS

Key:

  • MOV and MP4 are file types. ProRes 422 HQ is Apple's high-quality editing format; H.264 High is the common delivery format, and yuv420p its standard colour sampling.

  • BT.709 is the colour standard of HD video.

  • A closed GOP (group of pictures) means each group of frames decodes on its own, without frames from the next group. Half the frame rate is a group of 12 frames at 24 fps.

  • CRF 18 (constant rate factor) sets H.264 by picture quality instead of by data rate; 18 looks close to the source. Mb/s and kb/s are megabits and kilobits a second.

  • PCM is uncompressed audio; AAC is compressed audio.

  • Interlaced, upper field first: each frame is stored as two half-frames (fields), and the one holding the top line comes first, as many TV channels require.

  • LUFS is the unit of loudness, and dBTP the unit of the true peak (see Sound design and mix → Loudness: −23 TV, −16 online, −14 short social).

  • MENA is the Middle East and North Africa.

Loudness targets:

  • −23 LUFS for TV, and −16 for the online master and every online version not named a short social ad [official guidance];

  • −14 for a short social ad (feed and story placements up to about 60 s), named per version in the version table (decided).

Measure each version on its own. How to reach the targets is in Sound design and mix → Loudness: −23 TV, −16 online, −14 short social, and stems are in Sound design and mix → Stems.

Clean files. Masters, social and TV files carry no review label and no slate. Never overwrite anything already delivered: a new file gets a new version name.

Channel fades (decided): when a channel's sheet requires a fade in and out, it goes only on that channel's export, never on the master or another version. Write it into that channel's row of the version table, and tell the delivery check that the faded head and tail are planned dark.

The delivery check, on every export. Each line fails the file:

Check

Fails when

With common tools

Format

Codec, profile, frame size, square pixels or frame rate are wrong

ffprobe, command 1

Length

The frame count isn't exactly duration × frame rate

ffprobe, command 2

Colour and fields

BT.709 tags missing; field order wrong (progressive, or upper field first for TV); audio not 48 kHz stereo

ffprobe, commands 1 and 3

Loudness

Integrated loudness or true peak off target (±0.2 LU for TV, ±0.5 online); on TV, maximum short-term loudness above −18.0 LUFS

ffmpeg, command 4; in the timeline, Premiere's Loudness Meter

Black and freeze

Any black or frozen frame not planned. A still end-card hold is declared as planned; a freeze inside a planned locked shot is checked by eye

ffmpeg, command 5

Digital zero

Silence under −90 dBFS for 0.1 s or more outside a planned silence

ffmpeg, command 6

Key: dBFS is decibels below full scale, the loudest level a digital file can hold. Digital zero is a total absence of signal, quieter than any real room.

The commands for each line are in Do it yourself below; run each on the exported file as written.

Read every listed range against the planned ones: the end-card hold, a channel's fades, a planned silence.

List every quiet stretch under −50 dBFS for review (command 5 lists them), each marked planned or not. A planned silence drops to air or room tone, never to digital zero, unless the treatment writes a full cut-away (see Sound design and mix → Planned silence and playback compression). Write "PPM not measured" in every TV report: PPM (a peak programme meter limit) is a figure some channels also ask for, and a loudness meter doesn't read it (see Sound design and mix → Loudness: −23 TV, −16 online, −14 short social).

Then by eye and by rule

  • the first and last frames and every shot's key frames at full size (hands, product, label text, on-screen text, captions), because a crop can change what shows;

  • Arabic text and captions letter by letter (see Graphics, captions, versions and delivery → Arabic typesetting and render checks);

  • text holds and the end-card hold against the agreed figures;

  • graded stills against the brand's values (see The grade and texture → Grade standards);

  • the music licence covers paid media: the songs were generated and downloaded on a paid plan;

  • platform rules at delivery: AI-content labels (TikTok requires one for realistic AI-generated content, and Meta has a disclosure control), and the safe areas for captions, logos and product, read from each platform's current guide;

  • file names and sizes within each platform's limits (TikTok: 500 MB).

Watch and listen to every master and version yourself before anything goes out.

The client standard for the finished film

What the finished film is judged on, from a fail to award level. Solid is your minimum. The grade has its own standard in The grade and texture → Grade standards.

Part

Fails

Solid

Award-level

Sound and mix

Effects over speech; music fighting the voice-over; a studio-dry voice in a wide shot; loudness wrong for the delivery

Speech is the anchor, and the music ducks under it; hits fall between the words; each voice is placed at its shot's distance; every export is on its target

Sound so well recorded and mixed that it and the edit read as one

Graphics and type

The wrong font; a generic stand-in for the brand's emblem; heavy shadows; text jumping from place to place; a 0.16 s rise; a flare behind text

The brand's type and hierarchy; each repeated element always in the same place; entrances of about 0.4–0.5 s; text timed to its moment; Arabic letter for letter

Type that feels part of the film's world, not slides

Demonstrations and effects

Glowing particles the brand forbids; light that is emitted instead of reflected; a mechanism the client cannot prove

The brand's rules followed; materials that reflect real sources; the mechanism's wording confirmed by the client

Effects so well integrated that the demonstration reads as truth

Ending and branding

An abrupt black-to-white change; the end card held too briefly

The agreed end-card hold, reached by a designed move

The ending lands on the music, with the product as the hero

Versions and delivery

Wrong length or frame count; black or frozen frames; missing captions; loudness off; the master changed to make a version; a delivery link overwritten

The approved master never changed; each version its own sequence; the main ratios generated natively; every file measured; every delivery logged on a new link

Each version feels designed for its platform

Editability

A rendered-only edit

An XML with tracks by role, earlier takes stacked and switched off, overlays as ProRes 4444 and music options muted, with the LUT and the handover note beside it

You can swap any shot, music option or look in seconds

Do it yourself

Run the delivery check (ffprobe, ffmpeg, Premiere, DaVinci Resolve)

Run each on the exported file as written (an analysis command ends in -f null -, which reads the whole file and writes nothing).

  1. Format and colour:

ffprobe -v error -select_streams v:0 -show_entries stream=codec_name,profile,pix_fmt,width,height,sample_aspect_ratio,r_frame_rate,field_order,color_primaries,color_transfer,color_space -of compact file.mov

Square pixels read sample_aspect_ratio=1:1, and BT.709 reads bt709 three times. field_order reads progressive on every file but TV. On a TV file, upper field first reads tt or tb (set it in Premiere's export settings, Video › Field Order: Upper First).

  1. Frame count:

ffprobe -v error -count_frames -select_streams v:0 -show_entries stream=nb_read_frames -of csv=p=0 file.mov

A 30-second film reads 720 at 24 fps, 750 at 25.

  1. Audio:

ffprobe -v error -select_streams a:0 -show_entries stream=sample_rate,channels -of compact file.mov

It reads sample_rate=48000|channels=2.

  1. Loudness:

ffmpeg -nostats -i file.mov -af ebur128=peak=true -f null -

The summary at the end gives the integrated loudness (I) and the true peak (Peak). The maximum short-term loudness is the highest S: value in the lines above it; to print it alone, add this to the command:

2>&1 | grep -o "S: *-\?[0-9.]*" | awk '{print $2}' | sort -g | tail -1

In the timeline, Premiere's Loudness Meter effect on the Master track shows integrated and short-term loudness; take the true peak from this ffmpeg line or another true-peak meter. DaVinci Resolve's Fairlight loudness meter also works.

  1. Black, freeze and quiet stretches:

ffmpeg -nostats -i file.mov -vf "blackdetect=d=0.1:pix_th=0.10,freezedetect=n=0.001:d=0.5" -af "silencedetect=n=-50dB:d=0.3" -f null -

It lists any black lasting a tenth of a second or longer, any freeze lasting half a second or longer, and every stretch quieter than −50 dBFS that lasts 0.3 s or longer. Set these durations yourself: the filters' defaults only report events of 2 s or more, so a short fault would pass.

  1. Digital zero:

ffmpeg -nostats -i file.mov -vn -af "silencedetect=noise=-90dB:d=0.1" -f null -

Sources

  • EBU R 128 and its supplement 1 (the TV target).

  • The AES streaming recommendation (the −16 online target).

  • Two audio practitioners for −14 on short social ads (no platform publishes a figure).

bottom of page