The most expensive part of a delivery spec is not knowing it existed. Every destination your finished work can go (a streamer, a broadcaster, a platform, a client’s archive) has rules about what the file must be, and the rules are scattered across partner portals, PDF standards, and help pages that assume you already know the vocabulary. This sheet is the one-page version: what each destination wants, the loudness targets in a single table, and where the fine print lives when a delivery contract lands on your desk. Bookmark it next to the frame rate sheet and the aspect ratio sheet; it’s the third leg of that stool.
Last updated: August 2026. No affiliate links on this page by design; it’s a reference. Specs below cite each destination’s own published documents, read early August 2026. Standards move slowly but they move; the source is named next to every number so you can re-check it the day it matters.
TLDR
If you just want the answer: broadcast Europe wants −23 LUFS, broadcast US wants −24 LKFS, Netflix wants −27 LKFS dialog-gated with −2 dBTP, and YouTube will turn your loud mix down to about −14 whether you like it or not. Masters are ProRes 422 HQ unless a contract says IMF or a broadcaster says XDCAM/AVC-Intra. Safe action is 93%, safe title is 90%. Everything else is detail, and the details are below.
The loudness table (the part to bookmark)
| Destination | Loudness target | True peak | Per |
|---|---|---|---|
| Broadcast EU | −23 LUFS (±0.5 LU) | −1 dBTP | EBU R128 |
| Broadcast US | −24 LKFS (anchor/dialogue) | −2 dBTP common practice | ATSC A/85 (CALM Act) |
| Netflix | −27 LKFS (±2 LU), dialog-gated | −2 dBTP | Netflix Sound Mix Spec v1.6 |
| YouTube | ~−14 LUFS (unpublished, down-only) | −1 dBTP convention | Observed normalization |
| Spotify | −14 LUFS | −1 dBTP (−2 if louder than −14) | Spotify’s published normalization |
| Cinema trailers (US) | ≤85 Leq(m) | n/a | TASA / ISO 21727 |
| Cinema features | No normalization; mixed at 85 dB SPL reference | n/a | ISDCF practice |
Three traps hiding in that table. First, the measurements aren’t interchangeable: Netflix’s number is dialog-gated (measures the dialogue, not the whole program), so a “−27” mix by full-program math can still bounce. Second, YouTube’s famous −14 is real but unwritten: the platform documents no LUFS target anywhere, it simply turns loud content down to roughly that level and never turns quiet content up. You can watch it happen per-video in Stats for Nerds under “content loudness.” Master to −14 for YouTube and you’re just choosing your own gain staging instead of letting Google choose it. Third, US broadcast’s −24 is anchored to dialogue as the measured element; music-heavy programs trip this one constantly.
Masters: what “deliver the master” means by context
Client/agency work: the mezzanine convention is ProRes 422 HQ (DNxHR HQX in Avid country), full resolution, with stems if audio was mixed. Hand over an H.264 as “the master” only when the file is terminal: web-only, nobody re-edits it, nobody re-grades it. The moment a client might conform, re-cut, or archive, ProRes is the professional default and everyone downstream will thank you.
Broadcast: still dominated by two file species: XDCAM HD422 (50 Mbps MPEG-2) and AVC-Intra 100, wrapped MXF OP1a, because that’s what station infrastructure validates. The delivery contract will name one. If you get to choose, AVC-Intra 100 is the newer of the two workhorses.
Netflix and the streamer tier: licensed content delivers as IMF (Interoperable Master Format), specifically SMPTE ST 2067-21 Application #2E packages: a folder of MXF track files plus XML playlists rather than one video file. The distinction between “branded” and “non-branded” content is packaging (the “A Netflix Series” credit), not codec. The public partner pages carry the headline requirements; the full spec bodies live in partner-portal PDFs, which is a polite way of saying: if you’ve been asked for IMF and you’ve never made one, this is the moment you either learn Resolve’s IMF export or budget for a mastering house. Neither answer is wrong; not knowing which you’re choosing is.
Platform uploads: who publishes real specs
The platforms split cleanly into two camps. YouTube and Vimeo publish engineering specs. YouTube wants MP4, H.264 High profile, with SDR bitrate guidance of 8 Mbps at 1080p (12 at 50/60fps) scaling to 35 to 45 Mbps at 4K, HDR higher, audio AAC at 384 kbps. Vimeo goes further and is the only one that formally accepts ProRes uploads, alongside H.264/H.265 at CRF 18 or better and constant frame rate required; their published bands run 10 to 20 Mbps for 1080p up to 30 to 60 for 4K.
Meta and TikTok publish container limits and let the transcoder decide. Instagram’s help pages give you formats and durations, not bitrates; TikTok caps files and resolutions by upload path and says nothing about encoding quality. The working practice: give them clean H.264 at healthy bitrates and accept that what viewers see is the platform’s re-encode, not your file. Our export settings guide covers the per-platform recipes in full, including the vertical-format safe zones, and the encoder presets explainer covers what the settings actually do; this sheet stays at the what-each-destination-wants level.
Safe areas, one paragraph
The current standard is SMPTE ST 2046-1: safe action is 93% of the frame, safe title is 90%, for 16:9. The old 90/80 numbers you may have learned are the analog-era convention; today’s title-safe equals 1961’s action-safe, which explains a generation of confused graphics templates. Vertical platforms have their own UI-safe zones (buttons and captions eat the edges), and those are documented per-platform in the aspect ratio cheat sheet.
Review copies (the delivery before the delivery)
Client review wants small, playable, and watermarked when it matters: H.264 at 8 to 12 Mbps for 1080p is the convention, or skip local files entirely and let a review platform host it (they re-encode anyway; the review tools guide compares them). The QC habit that saves careers: watch the exported file, the actual deliverable, on something other than the machine that made it. The timeline playing perfectly proves nothing about the export.
FAQ
The client asked for “broadcast quality.” What do they mean?
Usually nothing specific; it’s shorthand for “not janky.” Deliver ProRes 422 HQ at the shot resolution with a mix that doesn’t clip and you’ve exceeded the request. If they mean it literally, there’s a real spec document naming codecs and loudness, and your first question is “can you send the delivery spec?” Asking marks you as someone who’s done this before, not the opposite.
Do I really need to hit −23 or −24 exactly?
For broadcast, yes: automated QC measures it and rejects on tolerance, and in the US the CALM Act put law behind it. For the web, no: normalization means chasing exact numbers matters less than not clipping and not smashing your dynamics to win a loudness war the platform will undo anyway.
What loudness should I mix to if the work goes everywhere?
Mix once at the most conservative target you owe (broadcast if broadcast is on the list), keep stems, and derive the platform versions: a −14-ish web master is a gain ride away from a −23 broadcast mix, but the reverse trip costs you the dynamics you crushed. Downhill is easy; uphill is a remix.
Where does captioning fit?
Every broadcast and streamer delivery includes caption files (broadcast wants SCC in the US; streamers commonly take IMSC/TTML; platforms take SRT). The spec document names the format; the trap is timing your captions to a picture that then gets re-conformed. Captions are a post-lock deliverable for a reason.
In conclusion
Delivery is the stage where craft meets paperwork, and the paperwork wins every dispute. The numbers above cover the destinations most working editors ever deliver to; when a contract shows up with a spec sheet, the spec sheet outranks this page, me, and your preferences, in that order. This sheet gets re-verified when the standards bodies or platforms move. Spot a stale number? Comments, please… this page only works if it stays right.
