Lanczos filter precision

This project is killing me!
Does anyone know what version of lanczos we have in Flame?
I’ve been asked this question:
Do you know if the Lanczos in Flame is Lanczos2 or Lanczos3 (Lanczos4 or Lanczos6 in Nuke parlance - they count the kernel lobes twice for some reason)
Anybody know the answer?

I suppose one of the devs can answer this question(?), but man, sympathizing for you Paul, sounds rough. Out of curiosity, who is asking this question? And how on Earth can this be a question worth asking or, for that matter, knowing the answer to? I suppose it’s something that I don’t know enough about or care enough about and perhaps it could cause some significant or noticeable issue down the line.

Edit: also, after now doing some research on Lanczos filtering, I know now that it is pronounced “Lanchose”- I’ve always said it and heard it pronounced as “Lankos”

Can’t really name names, but their preferred resize filter is simon , which we don’t have.

Have you tried Ringo filtering? #theAcademyAwardForBestFilteringGoesTo

Also, why can’t we know who these people are? Why would we want to work with these people without jacking up our prices based on their ridiculous filtering requests (which I am sure are not their only ridiculous requests). I don’t want to work with these folks. And I’m sure other people don’t want to either. Do we really think they’re searching Discourse forums for grievances over filtering?

Also, obviously you can’t just say f*** these folks. It’s a frustrating position in general. It’s a frustrating industry, that (I personally believe, and I know I’m not alone) is waiting for an overhaul. Whether that will come or not: unknown.

Britt, I’m not getting into a naming and shaming scenario on a public forum. Suffice to say it’s not a production company, but a company dealing with data pulls and the person I am dealing with is a tech person, so perfectly legitimate questions.
Also, they have never dealt with Flame before.


I hear you for sure. I tend to get a little bit too passionate about this. My understanding of vfx has lead me to a dog eat dog world where I lean into hating “the man,” but sometimes also it’s just like “is it Lanczos 4 or Lanczos 6”. My bad. I went full revolution, burn it down mode.

However, I want to point out the weirdness of going from “this project is killing me” to “perfectly legitimate questions.”


I don’t know who Paul is working for, but Disney/Marvel and Netflix have cartoonishly pedantic specs that call for things like this. It drives me bonkers because I’m sure they think they’re doing the “brown M&M’s” thing, but unlike Van Halen who allegedly did this to ensure that the electrical and safety work was also fastidious, nobody is going to die if the resize kernel is wrong. No viewer is even going to know. It’s a very blanket way to pretend you care about quality control while not seeing the quality of the image at all.

Nothing any of us will ever make is so crucial that a difference in softness invisible to the eye should be a factor. But it is, because it’s easy to load up a difference matte and go “it’s different”.

As to why people work for them: money.

1 Like

I took a screenshot of what Netflix sends to it QC vendors. There used to be a website, but i can’t get to it without an account now (I’m talking some type of QC vendor account, that I do not have). But this is the checklist. When stuff gets kicked back, it’s probably the client sending it back. Netflix as a streaming service is very worried that you send something that makes it seem like something is wrong with Netflix. They rank it in terms of numbers (like a number 3 is not important and they tell it to the client and say, hey check this out, it’s up to you to fix or not. Number 1 is if there is some stuttering problem or something and it HAS to get fixed. Sorry I couldn’t get to the website again, ugh, but to take the mystery out of the whole thing here their list at Netflix. This list doesnt represent the ranking system. But you can know that if you get a note about grain or a stabilize or whatever, it’s something someone wants fixed, it’s probably not some Netflix exec naw.

I too have fallen foul of Netflix’s QC idiosyncrasies . I had a bunch of phone inserts shots rejected on the basis that the inserts were too saturated and too red. They were exactly what the client had asked for and there were no technical problems with the shots, to me, it seemed more like creative intervention on the part of Netflix.
I it would appear they trump any creative input from the director.


Interesting! I do a lot of Netflix stuff. It seems to depend on the genre/ director.

1 Like

Everyone who has delivered to NF has fallen foul of their QC. In fact last year one of their (very helpful) technical team cheerfully told me, “I’ve never known anything pass our first QC”. :face_with_raised_eyebrow:

@paul_round Although I know they do query grades and other creative decisions, is it possible they were referring to the PSE spec… everyone knows about the fast luma flashing, but both 1702-1 and 1702-2 test for “luminance flashes, extended flashing and saturated red flashes”.

I can proudly say I’ve had a single episode pass without fail, although this was at the beginning of covid so I definitely feel this had something to do with it.
*I’d also say that it should have failed as there were parts which should have been flagged to have been checked

Love the reference to brown M&S. I read Dave Lee Roths autobiography too. Makes one nostalgic for the more cowboy days of post.

Anyhoo. Have we answered the Lanczos question? The manual is a bit obtuse.

@fredwarren @YannLaforest can you help?

Hi Paul!

The Lanczos filter in Flame is adjustable through the “Precision” numeric value when used through the Resize node for example. The “Precision” sets the Lanczos alpha which in turn determines the amount of lobes of the function. To have Lanczos4, you would set a “Precision” of 4. When importing/exporting a clip however, the Lanczos filter is locked at 5.25.