Skip to content

NETFLIX 2025-04-01 Tier 1

Read original ↗

Netflix — Globalizing Productions with Netflix's Media Production Suite

Summary

Netflix's Studio Engineering team describes the Media Production Suite (MPS) — a set of cloud-based filmmaker tools inside Content Hub that moves original camera + sound media into a centralised cloud asset library and automates the logistics between production, editorial, VFX, and picture-finishing vendors. The post is a program-level architecture overview — named systems, infrastructure shape, standards-driven automation strategy, and a Senna (Brazilian F1 series) case study — rather than a component deep-dive. Its reusable architectural content: a hybrid cloud (AWS + Netflix Open Connect CDN + global high-bandwidth ingest centres) sized for 200–700 TB-per-title media, a centralised cloud media library that replaces tape shipping + hard-drive hand-carrying across vendors, and a standards-driven automation thesis that leans on ACES / AMF / ASC MHL / ASC FDL / OTIO so the expensive per-title customisation burden of traditional post-production can be automated at scale.

Key takeaways

  1. Scale of media per title. "An average Netflix title produces around ~200 Terabytes of Original Camera Files (OCF), with outliers up to 700 Terabytes, not including any work-in-progress files, VFX assets, 3D assets, etc." This sets the design constraint: the ingest + library + distribution pipeline has to move hundreds of TB per title reliably and cheaply enough that the traditional LTO-tape + hand-carry workflow becomes replaceable. (Source: sources/2025-04-01-netflix-globalizing-productions-with-netflixs-media-production-suite)
  2. Hybrid-cloud infrastructure — AWS + Open Connect + ingest centres. "Netflix has invested in a hybrid infrastructure, a mix of cloud-based and physically distributed capabilities operating in multiple locations across the world and close to our productions to optimize user performance … Local storage and compute services are connected through the Netflix Open Connect network (Netflix Content Delivery Network) to the infrastructure of Amazon Web Services (AWS)." Canonical wiki instance of concepts/hybrid-cloud-media-ingest: AWS as the durable substrate, Open Connect as the high-bandwidth transit backbone, and per-region ingest centres as the edge where physical drives are plugged in and material is uploaded at high-speed.
  3. Centralised cloud media library replaces physical distribution. Traditional workflows require "shuttling hard drives on airplanes, creating LTO tapes, & managing physical shipments." MPS eliminates this: once media is in the cloud, every downstream tool (Dailies, VFX Pulls, Conform Pulls, Media Downloader, Remote Workstations) reads from the same library. "When utilizing MPS, we don't require LTO tapes to be written unless there are title-specific needs." This is the load-bearing architectural shift the whole suite depends on.
  4. Footage Ingest validation pipeline. "Once a drive has been plugged in and our Netflix Footage Ingest application is opened, a validation is run, ensuring all expected media from set is on the drive. After media has been uploaded and checksums are run validating media integrity, all media is inspected, metadata is extracted, and assets are created for viewing/sharing/downloading with playable proxies. All media is then automatically backed up to a second tier of cloud-based storage for the final archive." Six-stage pipeline: validate manifest → upload → checksum → inspect → extract metadata + build proxies → tier-2 archive.
  5. MPS is a suite of seven named tools, not a monolith. The post enumerates: Footage Ingest (drive → cloud), Media Library (search/preview/share/download), Dailies (QC + sound sync + colour + render + deliver to editorial), Remote Workstations (cloud editorial workstations + storage), VFX Pulls (automated transcoding + colour + framing delivery to VFX vendors), Conform Pulls (EDL-driven consolidate + trim + package for picture-finishing), Media Downloader (automated download once media is available). Each tool opts in independently — Footage Ingest is the gateway because every downstream tool reads from the same cloud library.
  6. Adoption numbers — program maturity indicator. "Over 350 titles have made use of at least one of the tools noted above. Input has been taken from all over the world while developing, with users ranging from UCAN … EMEA … SEA … LATAM … and APAC." Only scale number in the article beyond per-title TB — no QPS, no fleet size, no daily ingest totals.
  7. Standards-driven automation thesis. "We cannot lean into humans configuring JSON files behind the scenes to map camera formats into deliverables. By embracing open standards, we not only streamline these processes but also facilitate smoother collaboration across diverse markets … we've decided to lean heavily into standards like ACES, AMF, ASC MHL, ASC FDL, and OTIO." Each standard is mapped to a workflow concern: ACES + AMF → colour pipeline management; ASC MHL → file-management / verification (checksum + manifest); ASC FDL → framing interoperability; OTIO → timeline interchange. Open standards make the per-title customisation that's "often only offered to very high-end titles" automatable at Netflix scale. Canonical wiki instance of concepts/open-media-standards.
  8. Automation vs. per-title customisation economics. The article frames the core trade-off explicitly: mature post- production facilities hand-build "scripts and automation that flow between various artists and personnel within their facility … though the customization is time-consuming. E.g., Transcoding, or transcriptions that automatically run when files are dropped in a hot folder, with the expectation that certain sidecar metadata files will accompany them with a specific organizational structure." Netflix can't linearly scale that model across hundreds of titles + dozens of vendors per title, so the architectural bet is **standards + automation
  9. a shared cloud library in place of per-vendor scripted hot-folders**.
  10. VFX Pulls workflow concretely. With traditional LATAM infrastructure on Senna, "VFX Pulls would have been done manually … prone to human error and more importantly … too slow and would have resulted in different I/O methods for every vendor." With MPS: "the Assistant Editor was able to log into Content Hub, upload an EDL, and have their VFX Pulls automatically transcoded, color files consolidated and all media placed into a Google Drive style folder built directly in Content Hub (called Workspaces)." One standard I/O method replaces one-method-per-vendor — editorial and DI stop managing N different file-transfer flows.
  11. Perceptual conform via computer vision. "Since this early beta and thanks to learnings from many shows like Senna, advancements have been made in the system's ability to match back to source media for both Conform and VFX Pulls. Rather than requiring an exact match between EDL and source OCF, there are several variations of fuzzy matching that can take place, as well as a current investigation in using one of our perceptual matching algorithms, allowing for a perceptual conform using computer vision, instead of solely relying on metadata." Two-tier fallback: exact-metadata match → fuzzy-metadata match → (future) perceptual CV match. Useful as a wiki concept because it generalises — any pipeline that resolves references by identifier and can fall through to content-similarity probes ends up looking like this.
  12. Remote monitoring replaces phone calls. "Traditionally, if you wanted to check in with your post vendor on how things are going for each of these media management steps noted above … you would have to pick up the phone and call your vendor. For Senna, anyone who wanted visibility on progress, simply logged in to Content Hub and could see any activity in the Footage Ingest dashboard." The cloud library's observability surface is a secondary benefit: the shared dashboard is the single source of truth for every stakeholder, eliminating the long-latency out-of-band status check.
  13. Senna as worked global-collaboration example. "The series … was produced across Argentina, Uruguay, Brazil, and the United Kingdom. With editorial teams spread across Porto Alegre and Spain, and VFX studios collaborating across locations in Brazil, Canada, the United States, and India, all orchestrated by our subsidiary Scanline VFX." Demonstrates the hybrid-cloud design's intended deployment: geographically distributed production with cloud-centralised assets instead of physical drive logistics.

Architecture at a glance

Production set (drive)
    │
    ▼
Netflix Ingest Centre (global, high-speed internet)
    │   Footage Ingest app:
    │     1. Validate drive manifest
    │     2. Upload OCF + OSF
    │     3. Checksum validation
    │     4. Inspect + extract metadata
    │     5. Build playable proxies
    │     6. Tier-2 cloud archive
    ▼
Netflix Content Hub / Media Library  (centralised cloud asset store)
    │   gRPC / web UI search + preview + share + download
    │   Remote monitoring dashboard (activity stream)
    │
    ├─► Dailies   — QC + sound sync + colour + render → editorial
    ├─► Remote Workstations — cloud editorial + storage
    ├─► VFX Pulls — EDL-driven transcode + colour + framing → vendor Workspaces
    │                 fuzzy metadata match ▸ (future) perceptual CV match
    ├─► Conform Pulls — EDL-driven consolidate + trim + QC → picture-finishing
    └─► Media Downloader — automated pull when asset lands

Open standards driving automation:
  ACES + AMF (colour)  ·  ASC MHL (checksum/manifest)
  ASC FDL (framing)    ·  OTIO (timeline)

Transport fabric:
  Netflix Open Connect CDN  ↔  AWS (durable substrate)

Operational numbers

Metric Value Notes
OCF per Netflix title (average) ~200 TB Not counting WIP / VFX / 3D assets.
OCF per Netflix title (outliers) up to ~700 TB Sets the ingest + library scale target.
Titles using at least one MPS tool >350 Only adoption number disclosed.
User markets UCAN / EMEA / SEA / LATAM / APAC Global coverage; per-market counts undisclosed.
Senna shooting start June 2023 Early beta era for MPS.
Senna production geography Argentina / Uruguay / Brazil / UK Editorial in Porto Alegre + Spain; VFX in Brazil / Canada / US / India.
LTO tape creation with MPS "not required unless title-specific" Physical-media default off.

Systems + standards named

Netflix systems introduced here:

  • Media Production Suite (MPS) — umbrella system for all seven production tools.
  • Content Hub — parent product in which MPS lives; also hosts Workspaces (the Google-Drive-style shared folder for VFX Pulls) and the remote monitoring dashboard.
  • Footage Ingest — drive-plug- in application that validates + uploads + checksums + extracts metadata + backs up OCF / OSF. The gateway tool.
  • Open Connect — Netflix's CDN, carrying ingest-centre ↔ AWS traffic for the media pipeline (previously known on the wiki as Netflix's streaming CDN; this is its first appearance in a non-streaming role).
  • Media Library — cloud-centralised search / preview / share / download surface. No dedicated page (MPS component).
  • Dailies, Remote Workstations, VFX Pulls, Conform Pulls, Media Downloader — other MPS components; documented in systems/netflix-media-production-suite.
  • Scanline VFX — Netflix subsidiary orchestrating VFX on Senna across four countries. Named, not deep-dived.

Open standards (canonical instances of concepts/open-media-standards):

  • ACES — Academy Color Encoding System; colour pipeline.
  • AMF — ACES Metadata File; pairs with ACES for colour-pipeline automation.
  • ASC MHL — Media Hash List; file-management + verification (checksum + manifest).
  • ASC FDL — Framing Decision List; framing interoperability across resolutions + lenses + safeties.
  • OTIO — OpenTimelineIO; timeline interchange between editorial / VFX / DI.

Caveats

  • Announcement / case-study voice, not a deep architectural retrospective. The post names the seven tools and the hybrid- cloud shape but does not publish: ingest-centre count or locations, per-ingest throughput, global library storage footprint, proxy-generation pipeline details, latency SLAs on the library, Open-Connect-to-AWS topology, or how Workspaces is implemented.
  • Tool maturity varies. "Each of the individual tools within MPS is at different states of maturity" — the post doesn't distinguish which are GA vs. beta, so claims about what's automated should be read as directional.
  • No failure-mode discussion. Nothing on partial-upload recovery semantics, checksum-mismatch handling, ingest-centre outages, or how Remote Workstations degrade when Open Connect backhaul congests.
  • Perceptual-match is future work. "a current investigation in using one of our perceptual matching algorithms" — no model architecture, no accuracy benchmark, no deployment status.
  • Adoption figure is cumulative + opt-in. ">350 titles have made use of at least one of the tools" counts any interaction with any MPS component — does not imply 350 full-MPS deployments or any per-tool adoption rate.
  • No cost or infrastructure numbers. The post frames ingest centres as "rolling out globally" but provides no count, capacity, or bandwidth figures; no comparison to the cost of the LTO + hard-drive workflow it replaces; no saved-hours or reduced-error operational metrics.
  • LATAM vs. high-end-market automation gap is stated as motivation, not quantified. The argument that automation via standards democratises workflows "often only offered to very high-end titles" is a design thesis — no measurement of how much previously-manual per-title work the automation eliminates.

Source

Last updated · 319 distilled / 1,201 read