Get Blender graphics on 35mm film
on August 24th, 2010, by Ton
We’ve delivered 2 minutes of dpx files, straight from Blender sequencer, to the film lab Cineco here in Amsterdam. The first digital projection in their grading room looked brilliant! (Gigantic new Barco projector). It’s now being moved to the Arri Laser to make 35 mm film, anamorphic 2.35:1 aspect. In a few days we’ll get a screening. Exciting!
Thanks to Blender’s linear workflow now, we should be able to give them perfect linear DPX files, for them to convert to wonderful film colors. However, it’s never as simple as it seems!
Here’s how we did it now:
– Files rendered in 2048 wide, 872 pixels high. This matches the standard for film, so no scaling is needed.
– DPX files (3 x 16 bits color) saved as linear – default in Blender with linear workflow enabled – but with a gamma 2.2 applied (code not committed yet). That is the thing that confuses me most… but according the Arri operator at Cineco it is “the standard”. Their LUT conversion (lookup table) assumes this to go to film color.
– At first we made the mistake to test DPX files saved from 3×8 bits sources, but these are not linear! So DPX ended up in the color space of the 3×8 bits graphics.
I also had to fix the header in DPX files to not denote a gamma anymore, for some reason that gamma value doesn’t denote what the data is, but what the files should be corrected for in displays (confusing!).
People with DPX and film pipeline experience are welcome to feedback here. A dpx file and the corresponding jpg can be downloaded below.
http://download.blender.org/durian/2975.jpg
http://download.blender.org/durian/2975.dpx
(BTW: the graphic looks a bit dark, but was in the screening room totally brilliant! Dark tones survive much better in projection than on screens. Do we have to make a special graded version for LCD and TVs)?
-Ton-
August 24th, 2010 at 11:43 am
Man, I wish I was there.
August 24th, 2010 at 12:02 pm
As far as I’m aware. If the source renders were ungraded and to be graded at the film lab then deliver log space DPX, but if the renders have been graded in house then gamma 2.2 DPX, worst delivery would have been linear space.
August 24th, 2010 at 12:14 pm
Yellow: I realize that grading – since you do that on your screens – is color space depending too, but for a general first color and contrast pass, our high quality Eizo displays should be pretty reliable. We also do a second check on Apple cinema displays.
For this we read linear float buffers in Sequencer, these become sRGB color-space corrected float buffers, graded for good color and contrast, and then get inverse corrected back to linear again.
August 24th, 2010 at 12:22 pm
Hi Ton . Linear workflow is one heck of a confusing crap but nevertheless helps in perfect post processing of the image . The reason behind 2.2 gamma is that when u apply a 2.2 gamma to the image, it compensates for the monitors/ projectors non linear gamma . So when they do the color grading / LUT conversions they can apply a inverse gamma . ie: .455 and linearize the image for any medium . well . Now i am too confused .. hehe. 🙁 . Here are some links for ur research on linear workflow . hope it helps .
http://www.creativecrash.com/3dsmax/tutorials/rendering/vray/c/linear-workflow-in-3dsmax-and-vray
http://mymentalray.com/wiki/index.php/Gamma
BTW . Cant wait to see the movie 🙂 .
regards
shrinidhi
August 24th, 2010 at 12:36 pm
cool, i wish i could see colours shown in screening room
August 24th, 2010 at 1:12 pm
I would think a room has more surrounding light than a cinema has.
The blacks can fall back. I could not test it on my TV as I now have LED tv.
On my cinema display I cannot see any difference between the jpg and the dpx displayed by Photoshop. And changing the gamma does not improve the image.
A little curve to accentuate the tones in the black does help the image somewhat. But yeah, the shiny cinema display does mirrors a lot of the room in the image, that’s probably better when it moves.
August 24th, 2010 at 1:14 pm
The jpg looked dark on my computer screen, but on my TV, it looked fine. 🙂
If I wrapped my head around the dpx thing right, did you effectively apply a gamma of 2.2 to your originally linear output and encode that onto the dpx as if “linear” at 16-bits with a header not telling the decoding device to correct for it, but with the LUT assuming the input as gamma 2.2? Why would that be called “linear” anyway?
August 24th, 2010 at 1:29 pm
The jpeg shows up fin on my display.
August 24th, 2010 at 1:41 pm
Hi Ton, I heard about ARRILASERs beeing aligned for 1828 or 1920 square pixels, but never for 2048. 2048×872 is DCI only and needs scaling to 1828×1566 film recording for anamorphic projection.
The 2048×1566 setting is for Super35mm only, also exposing the sound area. S35 is rather unusual in Europe and for DPX recording.
The DPX header gamma historically came from the 1st generation Sun Grafics Workstations and SONY (luminance boosted) grading monitors, KODAK used for its digital Cineon workflow.
August 24th, 2010 at 1:58 pm
D: yes, Cineco told me that they expect linear data with gamma applied. Weird, eh?
Pixelpartner: for ARRI resolution, this also gives me contradicting info. One operator there said 2048 was perfect, the other wanted 1920… what i want to prevent is them digitally rescaling our graphics, which will make it more blurry.
If the native 35mm width for arri is 1828 wide, we should have rendered it as such too? (Or of course use 4k graphics to scale down, which will survive well).
August 24th, 2010 at 2:15 pm
From the Arri Laser 2k manual pdf:
Nominal spot size is 12.0 UM, image format 2048 x 1556 pixels.
I read this info as if this is the native resolution, that goes into the system and gets processed without digital resizing?
August 24th, 2010 at 2:17 pm
lately at my work we also rendered files to 2048 x 872 (from 4k Red One files) to make it anamorphic 2.35:1. I don’t know how it works after developing (if projections need to be done with anamorphic adapter on projector). That was the request form laboratory.
August 24th, 2010 at 4:39 pm
@Ton:
About your last question:
I would suggest to make a perfect REC709 (HDCAM SR) master in their grading suite for the most accurate results. (as far as the budget enables)
This also how Hollwood movies end up on Blu-ray.
About scaling and resolutions for the Arri laser.
You’re scaling anyway, because you vertically stretching the picture for anamorphic.
If the Arri records 2K anamorhic, it’s indeed 1828×1566.
But in Hollywood 2048 wide images, (that’s just the standard for a 2K DI) are just scaled to this resolution, which works fine. 😛
I suggest to be realistic, if there are very minimal scaling errors it’s highly unlikely you’ll see them on screen. (copy from IN to releaseprint, shaky projector, optical stuff etc.)
I really appreciate your wonderful care and love for this fantastic film, but stay down to earth! 😉
August 24th, 2010 at 5:04 pm
@junol:
Wasn’t it better the render directly anamorphic from the 4K Redcode files, in order the have a higher vertical resolution?
August 24th, 2010 at 6:03 pm
I hesitate to suggest this but you could try asking over on the Cinematography Mailing List. http://www.cinematography.net
This is a highend professional DP list and so care needs to be taken when approaching.
August 24th, 2010 at 6:22 pm
Maybe I’m showing my ignorance here, but I thought digital projectors were the future. Just think of Avatar. Why bother with film when it can get all scratched up and degrade?
August 24th, 2010 at 7:34 pm
Because not all theaters have a DCI installation? 😛
And especially not the smaller screens and arthouse cinemas.
A DCP version will also come, undoubtedly, but it’s now important to get the 35mm print (which the festival requires) done. 😉
Btw, you better have a good 35mm print than showing this on a simple beamer.
August 24th, 2010 at 8:40 pm
Not a dpx pro, but a few days ago I stumbled on something similar on some extended tonemapping / linear workflow research:
It would make sense, if the data wouldn’t be saved in rgb space,
since rgb gamma produces hue / saturation shiftings. so maybe, there is some luminosity based degamma-ing going on inside their pipeline, and they do ask for “linear” dpx, in case someone delivers it in a different colorspace.
interesting to hear whats behind this.
August 24th, 2010 at 9:53 pm
Hi,
First sorry but i haven’t read all the answers, and all I have is guideline but you need to communicate with the lab.
1) dpx files in a typical film workflow are 10bit log, not 16bit lin.
2) you have log, lin and gamma (2.2) image but you can’t have a linear gamma 2.2 image.
3) typically when a film lab receive cgi (srgb)or hd video (rec.709) files to be shot on film, they apply a 3d lut that will try to keep the original color on film.
That’s probably what’s the lab is doing.
4) About the resolution, 2048×1556 was the standard for the final resolution of S35mm (without the sound) film scanned. The academic aperture (so 4/3 with sound) was 1828×1556. Now, nearly every piece of hardware will have a different resolution for scanning, and even with different arrilaser model. The later hardware are able to scan academic aperture with 2048 pixel wide, and the arrilaser should do it.
5) 2048×872 is the standard resolution for “techniscope” wich is a cinemascope ratio but without anamorphose, which would be 1828×1556 but with a 2:1 ratio at projection. If you are after a 2.35:1 ratio and you didn’t rendered the image out of blender with a anamorhose, the 2048×872 should be the way to go.
Hope this will help.
If you understand french you can visit my website: http://www.gradingroom.org and go to the “docs” page you will find a document named “les workflows hybrides”, which quite precisly descibe this.
hope you will be happy of the film color.
cheers,
antoine.
August 24th, 2010 at 10:00 pm
sorry i made a few mistake:
4) About the resolution, 2048×1556 was the standard for the final resolution of S35mm (without the sound) film scanned. The academic aperture (so 4/3 with sound) was 1828×1556. Now, nearly every piece of hardware will have a different resolution for scanning but will most of the time output a 2k file. The later hardware are able to scan academic aperture with 2048 pixel wide.
Now for the shoot, all the labs i used take 2048 wide fileshe. the arrilaser in the swiss lab we use and even with different arrilaser model. , and the arrilaser should do it.
voilà,
antoine.
August 24th, 2010 at 10:01 pm
sorry my keyboard is really sucking:
conclusion the arrilaser should take 2k files.
August 24th, 2010 at 11:25 pm
@Antoine:
On Reduser I read that 16 bit linear gamma 2.2 is fine for CGI projects.
Why would 10 bit log better?
Are you sure they’re REC709/sRGB space for as input for the grading suite?
August 24th, 2010 at 11:42 pm
Sorry for the typos, my English isn’t as bad as you might think. 😉
August 25th, 2010 at 1:28 am
linear and gamma is a freaking headache. 10bit log WOULDN’T be better but a lot of post houses just like to work with those kinds of files. i worked on a project where i could have delivered 16bitperchannel openexr or even 32bitperchannel HDR files but they couldn’t take them.
i don’t think gamma 2.2 with a LUT that then applies an INVERSE of the gamma curve to get 1.0 gamma is better either but evidently, that’s what the lab is set to do. you’re right ton, it would TOTALLY be simpler if they could could take linear with gamma 1.0 and not convert it with an LUT but i guess the majority of their clients bring in 2.2 so they’re set up for that. so it’s purely a step that is made to accommodate the transfer house.
August 25th, 2010 at 2:13 am
But… thanks to Cineco Blender gets more advanced! 😉
August 25th, 2010 at 2:27 am
Ton said: “(BTW: the graphic looks a bit dark, but was in the screening room totally brilliant! Dark tones survive much better in projection than on screens. Do we have to make a special graded version for LCD and TVs)?”
I beg you Ton – please put proper colour management on the map for Blender.
You have some talented minds involved with Blender that are well versed in the needs of colour management such as Xat.
You could likely implement the first non-invasive approach strictly in the nodal compositor and build outward from there by testing the code.
If you can do that, we can get proper mastery of the output (and entire pipeline) and solve this dilemma. Imagine proofing your work via ICCs / LUTs and grading via creative LUTs…
I beg thee!
August 25th, 2010 at 5:28 am
You should definitely read Antoine’s recommandation, they are the most correct for now in all aspects 🙂
@J.: about what you read on Reduser. The point with 10bit log is that it’s the industry standard for DPX *delivery* for labs. 16 bit linear is great for CGI, you’re right. But the color worfklow you use during your 3D / 2D work is usually different from the one you have when you deliver the data (whether it’s for a lab or a DVD, Bluray or even a H264 file for internet).
For exemple: you may work with half-float linear EXR or even a-la-nuke 32bit float in your vfx pipeline (because it’s great for 3D and 2D compositing) but deliver your movie in 10bitlog (which is a standard easily handled by every labs in the industry) to print on film.
@jin choung: you’re right, if the labs could directly handle linear EXR or linear 32bit float data, it would be even simpler. Using DPX’s 10bitlog is historical but convenient too because the file size is smaller and the color information available is perfectly suited for film (the quantity of information available in 10bitlog is sufficient to describe the color dynamic of film).
Just one clarification about DPX: they can either be 10bit files or 16bit files! And in 90% of cases the 10bit files are log! 16 bit files or sometimes “pure” linear but most of the time with a gamma applied. And in fact, Ton, when Cineco asks for 16bit linear with a 2.2 gamma it means 16bit images with a 2.2 gamma applied.
I bet that in fact their Arri Laser just can’t directly handle linear data. They need log files or gamma-based files because they have to specify a LUT. The LUT being: log (cineon or custom) or gamma-based (2.2 or rec709 / sRGB because these 2 last implies a gamma).
August 25th, 2010 at 5:29 am
Now, just a comment about what you said in your original post Ton. I think some of the confusion comes from the way Blender exports DPX today…
You say in your post that DPX images are saved in linear because it is the “default in Blender with linear workflow enabled”.
The point is that, even if you have an internal linear workflow (inside blender for 3D renders and 2D compositing), your DPX exports should really propose more choices:
* choice between 10bit and 16bit
* choice between linear, gamma or log color coding
with the following defaults:
* log when 10bits (the standard everywhere for film printing)
* gamma when 16bit
Why a gamma coding when 16bit? Because it is more common to see 16bit *integer* files with a gamma applied. In fact, the “true” linear files that you may encounter are indeed (most of the time) *floating* files like EXR (which is not really “16bit” as it is usually described: the correct term is “half-float” because this is what it is: the half of a 32bit float)!
August 25th, 2010 at 5:34 am
(Sorry for the multiple posts but I have it seems that there’s a size limit?)
The values available should be:
* linear: no values needed
* gamma: 2.2, 1.8 (or any value entered manually), Rec709, sRGB (these two lasts are almost pure 2.2 gamma but Rec709 and sRGB not only have a 2.2 gamma but are also color spaces which means that some color computing must occur to limit the gamut, maybe with littleCMS?)
* log: cineon (“Big” historical standard: black point at 95, white point at 685) or values entered manually (black point and white point are the minimum information necessary but beware (!) ideally, the best solution here would be to propose a true 1D LUT editor! The one in Autodesk Flame, page 1652 of the Flame 2010 user guide is a very good example!)
Ahlala, complex (but not that much, color graders deal with that every day. But it’s true that they are good and smart people :)) and technical subject but very interesting indeed! And as Troy James Sobotka sayed: a comprehensive colour managemen system misses in Blender and having would be a big big plus. But more than this: these tools are needed not to be “Professionnal” (hell yeah, with a big P) but to have a comprehensive toolset to work with colors. These missing tools are somehow the basics in fact :-/ (no offense!)
Good luck guys! Delivery is always a brainstorm of techniques with headaches but good communication with your lab is the key to success!
August 25th, 2010 at 5:38 am
Just a link to illustrate the limitation of the gamut for special gamma curves like Rec709 or sRGB: see the gamut of sRGB in the LAB space on the right illustration of http://en.wikipedia.org/wiki/SRGB_color_space
August 25th, 2010 at 5:46 am
Thanks, MaximD! 😀 Very informative.
Maybe it will help clear some things up.
August 25th, 2010 at 6:02 am
About Flame’s 1D LUT editor: it’s page 1652 of the User Guide available at bit.ly/flame2010userguide
It’s just an example though, the idea is to provide a way for user to specify its own LUT. Providing a simple tool / way to create them is important in this case. In The Foundry’s Nuke there is a simple “Log2Lin” node which handles the conversion in both sides: log2lin and lin2log
For example, in the case of a DPX export for lab in 10bitlog, a lin2log operation would be used.
One correction from what I said earlier: for the log parameter in blender export, the minimal values needed are:
* black reference (sorry, “black reference” is more precise than “black point”)
* white reference
* gamma (but beware this notion of “gamma” has nothing to do with the standard 2.2 gamma applied elsewhere! In a log coding, this gamma describes the *film* gamma)
But other interesting values may be available to be more comprehensive like the max value for black (usually 0) and white (usually 1023) and the notion of soft-clip for values bigger than the white reference (very interesting to maintain highlight information). These comprehensive settings are those available in Flame.
August 25th, 2010 at 6:04 am
Typo: “or the log parameter in blender export, the minimal values needed are”
-> the minimal values *should be*
August 25th, 2010 at 12:31 pm
@J
it was request from lab and I thing it was discussed with production or DP (maybe it was cheaper 😉 ), and why that resolution: the Antoine’s reply is I thing the answer.
greets
August 25th, 2010 at 12:33 pm
Hi Everyone,
Thanks for all the info, that’s really helpful! I certainly will ensure that Blender gets the appropriate settings for exporting to DPX, so you can match the client’s or the lab’s preferences.
Linear Workflow is still a newbie in Blender, and will definitely need further work to get the best artistic control over the final pictures. It’s on the agenda!
August 25th, 2010 at 12:33 pm
A few questions from the colour blind like me… 🙂
”
* black reference
* white reference
* gamma
”
That’s the “Levels” in photoshop?
And the LUT is an array made from the “Curves” in photoshop to remap given values to new values?
Is this correct?
When is the industry going to create/release 16bit monitors?
Blu-ray seems to be able to store 10bit ( AVCHD ). Is it worth the effort or is this more a thing for conversion to film?
August 25th, 2010 at 1:43 pm
Normal Blu-rays are just 8 bit 16-235. 😉
@Ton: Will the DVD/Blu-ray grading also be done at Cineco?
And who will make the Digital Cinema version? (DCM and DCP)
August 25th, 2010 at 3:55 pm
J: Cineco does no actual grading, they will output the dpx straight into their pipeline with their “standard LUT”. We will get a 2 minute preview soon there, to check on how our work survives.
For DVD we’ll just use the sRGB corrected output of our Sequencer.
Joeri: I also don’t fully grasp the black/white references. I’ve found on the web good discussions on it though, it seems film has no real 100% black, it’s always very dark gray. Also dvd and bluray show no actual black in films, check it!
August 25th, 2010 at 10:45 pm
Hello Again,
Nice to read all the comments (i had time today).
Well in my cgi pipeline experience (Maya, C4D, blender, Lightwave, XSI) we always output 16bit float (called half-float) buffers (or passes) EXR files, the only exception is the zdepth buffer that we always render in 32bit float.
DPX is fine, but EXR is the better. First not all software will import your 16bit DPX files, the standard is really 10bit.
DPX is best suited for master delivery to film for example, in 10bit be it log or gamma encoded, but not linear (you will waste data).
The lab will add a 3d lut that will convert sRGB/rec.709 color space into world of film in order to try to best match the color from your digital files.
I don’t think you should provide the lab with log files, because 1d lut lin2log will not handle well color, the 3d lut from the lab should be way better.
If you like some color management in blender, start to look to nuke, it has a very simple way to do it: you inform the software what kind of files you have (linear, log, rec709, or sRGB) and then it will convert it to linear for you to do some compositing, and in the viewer settings you set in what color space you wish to see it, most of the time it will be in sRGB (that’s most probably the color space of your computer monitor). So as you work on linear data, you watch sRGB data which is designed to be viewed on your monitor.
cheers,
antoine.
August 25th, 2010 at 10:49 pm
To joeir: a lut is more or less a mapping table, with trio (rgb) value that will get remapped to other trio value.
Now you have 1d lut and 3d lut, the latest will remap in more complex way and is the only way to get quite accurate color value for film simulation when grading.
August 25th, 2010 at 11:19 pm
Why exactly would it be better to bring EXR to the lab instead of DPX 16 2.2?
Just curious…
August 26th, 2010 at 12:59 am
I’m having this weird gamma problem with MentalRay and Real Lights.
For some reason I seem to need to gamma correct my textures before going to render in EXR buffers. But it’s highly confussing at what point I’m looking at a correct image. I guess I’m lucky I only need to render movies for video and web.
I can understand the “no black” in films. Its also better to have no white in print. Always use at least 4% black so the white areas on your paper will have some ink and but bleak out. I guess for film its also better to have some grains so there is no cut-off border between almost black and black.
August 26th, 2010 at 2:46 am
Antoine’s recommandation about Nuke’s way to deal with color management is a very very good advice: Nuke’s system is clear and limpid and is definitely powerful!
August 26th, 2010 at 3:06 am
If you want to try Nuke’s color management system, you can download the PLE (Personal Learning Edition) here: http://www.thefoundry.co.uk/products/nuke/ple/
No limitations except some funky color patterns on all images and the impossibility to use external plugins.
August 26th, 2010 at 8:37 am
To J. :
Sorry for the confusion. No I don’t advice to bring EXR to the lab (except if you are working with ILM lab ;-).
I am saying that the workflow I know is EXR out of the 3d software, then compose the different buffers in a compositing package (like Toxik or Nuke, or the blender composite) and then output DPX files as the digital master for digtial cinema, shoot to film or hdcam video tape.
cheers,
antoine.
August 26th, 2010 at 10:23 am
To Antoine:
Do you know any good tutorials on how to setup Maya to get a good preview render and still render to 16bit output?
August 26th, 2010 at 6:34 pm
To Joeri:
Well actually I’m not a 3d artist, i’m a digital compositor and a junior colorist, so i am not expert of maya nor any 3d package.
But i would think that if you do some preview render, composited with all buffers inside Maya renderer and outputed in sRGB color space, and then when you are happy do the renders in EXR 16bit half float and compose the buffers in Nuke, Toxik (well now it is call Composite and is part of Maya) or any other compositing package.
The EXR out of blender should be linear data (no gamma) so when you open the images it will look dark, you need to keep them linear while composing the buffers but view that in sRGB color space, like Nuke would do it nearly automatically, but Toxik can also do that just in a bit less intuitive way.
When you are finished with your composite convert the result to sRGB and render out DPX 10bit (which will have a 2.2 gamma, the one fro, sRGB)
Sorry for the rough explanation but color is quite a complex subject and hard to explain writing here.
Hope it will still help you a bit.
cheers,
antoine.
August 26th, 2010 at 8:47 pm
It’s a pity that your site is only in French, which I’m not able to read… 🙁
August 27th, 2010 at 12:38 am
to J.:
Well the documents i wrote in french were actualy quite a bit of a harcel to write, because for me this subject is easier to write in english as a lot of word come from english culture, like workflow (which in strict french would be translate as “flux de production”) are really natural in english for me and not in french.
My point is I will write these documents in english, but I really don’t know when. Sorry. If i remember correctly these documents took me at least one month to write; it is a synthesis of quite a lot of books and web articles.
The least I can do for now is link you to the most important things (BTW references that are in my document):
http://www.acvl.org/digital_intermediates/dicompanion/index.html
http://www.theasc.com/magazine/jan05/conundrum/index.html
http://www.theasc.com/magazine/april05/conundrum2/index.html
And Charles Poynton (specially for gamma question)
And of course the Kodak website for question about film, it is a gold mine of information!
These 3 things will give you a huge lot of information, and if you want more, just get my document called “workflows hybrides” and go to the last page where you get the bibliography, and read all that is in english
Hope these will helps 😉
cheers,
antoine.
August 27th, 2010 at 2:10 pm
I think what J. meant to say was that its a pity hes doesnt understand French.
J. will bablefish convert the site?
August 27th, 2010 at 9:27 pm
No, I mean exactly what I wrote.
Most labs in non-English countries also have an option to view their site in English. (if, for instance, an American production is shooting in your country and they want to use your lab for dev. and dailies etc. you know… 😉 )
Putting his site into Google Translate or Bablefish doesn’t work, since it’s entirely in Flash. 🙁
August 27th, 2010 at 11:19 pm
To J.:
Well I am not provinding directly prints, so we are not a lab, but we colaborate with labs, depending on the project needs.
Yeah the website is in flash, but the information is in the documents which are pdf. But don’t try to translate them with some automatic translator, I don’t think it will give a good result.
You have to understand that it is not a company, this is a 2 man association that just started. The first service we provide is grading not printing. We also are a non profit association and we first target local (swiss, belgium france) short and low budget feature film.
I am sorry that we can’t answer your needs for now. However if you need some consulting for a particular project, I have no problem to communicate in english.
cheers,
antoine.
August 27th, 2010 at 11:42 pm
I don’t have any needs, I’m just interested to read such things. (altough it would be great to get a job in that business, I’ve to admit)
September 3rd, 2010 at 12:36 am
Congratulations. Those renders indeed look brilliant!
September 9th, 2010 at 7:09 am
like the staff very much , wait for see , and thank at commit the good gives everybody , thank you very much ,