Tracking, Keying and Layout. Workflow Demo.

by - 2012/07/11 36 Comments Compositing, Production

Here’s a little timelapse of the general workflow for our keying shots.

First we track the camera and save that as a blendfile. From that file we generate a new file that is then used as base for masking and keying. The tracking markers can often be re-used as a way to mask out stuff from the footage.

The cleaned footage is then saved as 4k openEXR files with premultiplied alpha channel.
A third blendfile is created as a base for that layout, swapping the footage for the clean plates and setting up the final shot dimensions (1920×800). That file is then handed over to the person that is then doing the layout for it. The scene layout for this scene was originally done by Andy.
Usually after the main composite is done I fix some alpha-blending issues that cannot be solved in the pre-key outside the main composite.

The key that you see here is still a little but too harsh on the one side of the head, but for the demo I didn’t want to tweak it too long.

  1. Micah Denn says:

    that’s supposed to be 1920×1800 right?

  2. TMW says:

    Interesting work flow, do you only need two markers to track tripod motion as opposed to eight for hand held? Also is the OpenColourIO stuff available in the Tomato branch?


    • PhysicsGuy says:

      Indeed, for a tripod shot, one needs only two markers. The reason for this is that one can only solve for two things, namely the changes in the azimuthal and elevation angle of the camera.

      If you have a shot with perspective shift, one can derive much more information from the shot, such as the position of the camera in 3D space. This also means you need more trackers to find the solution.

  3. Karlis says:

    Look’s good.. but it’s not always enough just with seeing one frame, what if you want to see this in movement? Does it have some kind of RAM preview? Because it’s important to see noise problems or other stuff, that appears only in motion.

    • sebastian says:

      That’s true. But that’s why the prerendering of the clean-plates can be handy, because that’s our preview for the key. But still, often we only see keying/tracking/roto problems in the final renders, that’s when we have to revisit comp&key.
      RAM preview would be awesome, but not possible (yet).

      • Karlis says:

        I understand..

        But yes, RAM preview would allow not only compositing work be more productive, but also motion graphics stuff to be made easier..
        I really don’t understand much from programming, but essentialy it renders frames and cashes them in RAM.. doesn’t sound very hard.. But again, I don’t know it so deeply. ;)

        • sozap says:

          Then it will be like rendering the viewer, or rendering the full nodetree with muting some nodes. I think it won’t be that usefull. Maybe the best thing to have is a ram or disk cache that store some result of the nodes, in different frame of the animation. Then when you scroll in the animation the compositor can load cached data without recalculating the full tree. That, combined with a proxy mode when you can render half res or quarter res sequence, will surely help to work in a more realtime way. But that is a tough thing to code I guess.
          By the way , thanks a lot for the timelapses, and congratulation with your progress so far, that’s very impressive !

          • Karlis says:

            I don’t think there is a way to not recalculate every frame.. it’s not an accurate representation of result then. And it’s not possible to render everything in background.. and if you change something in nodtrees begining, that influences everything after that node. I think the simplest way here is actually the most logical. Of course after implementing that.. it’s pissible to optimise it.. so it renders really only whats changed and not everything.. or maybe it caches previous render on disk and if you undo the change, it gets reused again.. I think that is approach for the new After effects version, but I could be wrong..

      • guest says:

        A proxy/RAM preview like that would be absolutely EPIC!!!

        I hate waiting hours for something to finish just to find that in half of the frames the key is off, or the edges warble.

  4. Andy says:

    Thanks for the preview – hope to see some keying/masking/tracking tuts related to Mango from you if possible

  5. Dusty says:

    and….. how many of these keying shots do you have a week to do?
    My respect for you grows daily! And with new technology too!
    Man I just can’t get over how good the result looks!
    Thank you for sharing! Can’t wait for more!

  6. Joe says:

    Is the final film going to be delivered in a 2.40:1 aspect ratio? (1920×800 pixels) Because if so and you are going for the true anamorphic look, you have to keep in mind that anamorphic bokeh is not round but oval in shape, because anamorphic lenses squeeze everything on the plane of focus by 2x, which makes a 4:3 image 2.35:1 aspect ratio when unsqueezed. But everything off the plane of focus gets squeezed by a different factor because light isn’t converging at the sensor to form a focused image. Meaning that when the footage is unsqueezed, the bokeh is not circular but oval shaped because it was squeezed by a different factor when filming. Anamorphic bokeh looks like this:
    It’s also why anamorphic lens flares have those cool horizontal streaks!

  7. J. says:

    Most ”scope” films are shot with spherical (normal) lenses and cropped in post nowadays.

  8. stephen says:

    Hi Sebastian,

    I’d love to know what compositing/keying books & training vids you have on your shelves :-) I’m sure if the Blender Institute were to release a compositing/keying learning resource/book/DVD there would be a lot of takers.

    Keep up the good work, looking awesomeballs.

  9. Hi Sebastian,
    thank you for demonstrating your compositing workflow, it was quite inspiring

    By the way, how do you got this straight connector lines?
    I tryed to find that option in the preferences and in the Node Editor itself, but i didn’t find it.

  10. Sparse Voxel Octree says:

    Hi Sebastian, great stuff.

    Would be possible to use custom masks for non masking purposes? I mean:

    – I draw a square mask over a window in my scene
    – I blur it and mix it with my render, set on “screen”
    – It gives me a nice glow effect on my window without exporting and doing it in GIMP/Photoshop…

    Hope it’s clear, kind regards

    • Payton says:

      YOu can do that now with the current masking system, its not too hard. Masks can be esed for many things, not just rotoscoping and matte creation.

  11. Wray Bowling says:

    This is the first time I am seeing The Film Shit™. What a strange effect. Somehow, before it was applied at the end I thought that this didn’t look very good. Then when it was applied, it looked “good” to me. I thought I wasn’t such a sucker for such simple color corrections but maybe there’s something more magical going on that I don’t know about? My guess is that I’m just not at all used to seeing someone work with 32-bit image data from beginning to end in a composited video shot. (that is what’s happening, right?)

    Anyway, very cool sebastian. I think the above paragraph clearly shows that it blowing my mind a little.

    • J. says:

      Welcome to the world of color mapping/LUTs. :D
      Yep, even though the image from the camera is linear and perfectly balanced, the human eye doesn’t work that way and wants something natural/analogue/imperfect and that’s why good old film looks so pleasing to the eye.
      Humans are not linear and unambiguous machines and that’s why making art with computers is an art; it’s making the perfect imperfect and emotionally touching.

      • blendercomp says:

        well said man, almost poetic! :)
        This obvious fact escapes our attention very often

  12. blendercomp says:

    Hey Sebastian and Mango crew, this is beyond awesome! :)
    Btw, this is the very reason why I suggested that the tracking DVD be released *after* the movie is completed not before.
    I do sincerely hope that the BF produces tutorials about this or even a new integrated series covering tracking, masking, and compositing in (even) greater depth.

  13. AlekB says:

    I’m amazed at how awesome this is! Good keying stuff is here. Now I have two questions: How do you just drag the Node Frames into existence? And second how do you set it so that the Background changes to whatever node you have selected? Also, for the people who want in depth tutorials, Blender is gaining most of the features that commercial programs have. So reading stuff the The Art and Science of Digital Compositing will get you way into some of this awesome stuff! Of course, in depth tutorial series would still be fantastic!!!

    • sebastian says:

      That is the frame node. select nodes, hit ctrl+j and they will be framed. If you want to detach a node from a frame, hit alt+f. To get rid of the frame hit x. All that is possible in latest trunk. Either build yourself or go to for latest version.

  14. MyName says:

    Omg, I just went to this website and got redirected to a domain with the new name for this movie!

    *won’t spoil it, but thinks it’s a good name :)

  15. lewis says:

    How do you do the floating socket?
    Like the little circle that reroutes noodles?
    Can be seen at 5:38 where the cursor is.
    Reply appreciated.
    Also how do you the box background?

  16. chabane says:


  17. maghoxfr says:

    Will there be a new training DVD based on all the improvements made after mango project? I would love to have that.