alt Hacker NewsNetflix’s Media Production Suite
Comments
I was reading about the cinematography of _Collateral_, possibly the first large budget feature film to be shot digitally, and one of the issues back in 2004 when it was made was the amount of storage required for digital video and the risk of not being able to retrieve the images from the data stores:
> “We did massive testing with the hard drives, and everything was great, and then we had an experience where we shot, and when we sent in the material, they couldn’t get the information off the hard drive,” said Cameron. “So the studio went ballistic and was like, ‘There’s just no way we can we can let you guys do this.’”
> The compromise was the production would record to hard drives as well as SRW tape. And unlike today, verifying the digital footage was equally cumbersome and tension-filled.
> “We recorded everything two or three times on decks that we carried with us,” said Beebe. “So we were backing up, two or three times.”
https://www.indiewire.com/features/interviews/michael-mann-c...
> “… automation became imperative. The intricacies of color and framing management, along with deliverables, must be seamlessly controlled and effortlessly managed by the user, without the need for manual intervention. Therefore, we cannot lean into humans configuring JSON files behind the scenes to map camera formats into deliverables.”
I’d often thought (critically) about the lack of visual diversity in Netflix output - and this is something I often see stereotypical film-enjoyers complain about.
I’d never considered it as a consequence of Netflix’s sheer scale. It’s always really interesting when I discover that something I’d previously put down as an (unimaginably unimaginative) aesthetic choice might in fact be an operational choice. It makes me check myself!
It sounds an incredibly complex and clever system; I can’t help but feel that applying such a strong vertical to the more creative aspects of film and tv production - such as colour grading - will ultimately prove short-sighted.
15 years ago, the first start up I worked for provided APIs for music streaming in India. One of the founders who managed all infra was in US, and so the servers (bare metal) were in LA. I still find it amusing, that it was cheaper (and faster) just to fly to India, buy bunch of portable hard drives, upload the media, fly back to US and upload the data to the file server, than uploading the media directly from India to the US server. Obviously only applies when data is in order of TBs. Later saw the same thing with AWS Snowball and Snowmobile.
>Netflix has been spinning up ingest centers around the world, where drives can be dropped off, and within a matter of hours, all original camera files are uploaded into the Netflix ecosystem.
Netflix is looking to sell shovels to as many people as possible, it seems.
With the asset sizes they are talking about (hundreds of terabytes), how does it make it feasible to do this over the wire? Even with 1Gbps connection, it will take ~10 days to upload a single 100TB original camera file. And there could be several.
More on Netflix's Remote Workstation setup for artists https://aws.amazon.com/solutions/case-studies/netflix-workst...
I can see this evolving into something like AWS — a platform that offers high-end production tools to anyone willing to pay. That would democratize access to cutting-edge tech, effectively solving the tooling problem. But it still wouldn’t address the real bottleneck: compelling storytelling.
The part that struck me most was how much manual, error-prone work is still common in the industry. Still, I wonder how portable this is outside Netflix. It sounds like a very vertically integrated solution.
What languages are used? The screenshot of the desktop app looks native.
Isn't this what Frame.io does just without the markup tools? They have had camera to cloud for a while.
I wonder if Netflix will offer MPS as a service in the future for shows that are not on Netflix.
Having a dozen different VFX departments using different file transfer methods like FTP seems like a nightmare. But then I realized that the banks do this, and probably worse.
There's one that uses Gmail to exchange documents (not financial, but important nonetheless) and uses the read receipt to determine if it has ingested the data. Replaying ingestion is marking unread.
The tech is cool, but it seems like the main result of having such a pipeline is that Netflix has been able to produce an incredible amount of low-effort schlock that mostly lacks soul and artistic merit.
I read through that whole article thinking,
It never occurred to me until I reached the end that this wasn't a "enjoy this tool we made" post, but instead a "look how awesome we are" post. :-/