3
$\begingroup$

I'm doing another lightsaber movie and I just realized something that makes me pretty sad.

My workflow is such that I edit, then export reference footage, animate the lightsabers, and then render out the lightsabers and clashes by themselves that way I do not lose more quality in the original footage and I have more flexibility. However, I realized that the images I'm seeing in the compositor aren't quite the same as what comes out of the VSE when I set the clip to the Add blend mode:

Here is my node setup: enter image description here

This makes it so I can see how it should look with the footage and render out the lightsaber layer before it's added to the footage:enter image description here

So then I take the rendered out lightsaber layer and add it back in the VSE:enter image description here

But much to my dismay, rendering out stills from the compositor and VSE look very different (I personally prefer the compositor's look).

With the Compositor: enter image description here With the VSE: enter image description here

To see the difference easier, right click to view in a new tab and then tab between the two (I didn't want to loose any more image quality by using a .gif).

So what is the difference between these two addition functions and how can I make it so the VSE add blend mode has the same result as the compositor? Clamping the add nodes and turning off the adjustment layer does not influence this effect.

Thanks!

UPDATE: Rendering to .exr surprisingly does not fix the issue. Further testing also shows that the issue is most likely in the adding method.

$\endgroup$
  • $\begingroup$ What format are you using to render the FX elements? $\endgroup$ – cegaton Jul 18 '17 at 2:56
  • $\begingroup$ @cegaton Quicktime .mov Thanks $\endgroup$ – Anson Savage Jul 18 '17 at 3:38
  • $\begingroup$ You are then comparing apples and oranges... The information on the original scene and that on the quicktime files is quite different. It is not the add mode. $\endgroup$ – cegaton Jul 18 '17 at 3:47
  • $\begingroup$ @Okay, makes sense, but kind of scary to think about. I wonder which formats come out different than what the compositor shows. So, would it make sense to render to the same format as the footage? $\endgroup$ – Anson Savage Jul 18 '17 at 13:31
2
$\begingroup$

Most likely the difference is not how blender does the add operation, but the fact that you are dealing with two very different kinds of images.

"I personally prefer the compositor's look"

In the compositor you have the original rendered layers, 32bit float linear, scene referred information.

If you compress that into a video format then you've transformed your images into something quite different: 8 bit display referred images that have been converted using the view transform (set in the color management section), and on top of that they have been compressed using some codec that basically trows away a lot of information in order to have smaller files and creates a new set of problems...

What is the solution then?

"I do not want to loose more quality in the original footage"

You don't want to loose quality on the VFX created in blender either! like the quantization and saturation errors you show on the last image.

So instead of rendering to a video format, save your elements to OpenEXR files. You will preserve all of the information from your original render and it will give you even more flexibility for your final compositing.

Also, since you are dealing with emissive elements with no opacity, you need a format that can do proper associated alpha. EXR will allow you to use proper transparency for the emissions and glows.

$\endgroup$
  • $\begingroup$ Awesome! I have one question though. If OpenEXR saves everything as it is supposed to look, what should I save the final video file to if I don't want to get as many weird errors? I don't think YouTube uploads image sequences... $\endgroup$ – Anson Savage Jul 18 '17 at 14:16
  • 1
    $\begingroup$ And make sure you have a large hard drive. EXR files are larger than PNGs (because they encode more information). $\endgroup$ – Mutant Bob Jul 18 '17 at 14:16
  • 1
    $\begingroup$ Youtube likes mp4 files and mpeg transport streams. For a little more detail I have written a short paper outlining some philosophy and techniques: web.purplefrog.com/~thoth/blender/python-cookbook/… $\endgroup$ – Mutant Bob Jul 18 '17 at 14:18
  • 2
    $\begingroup$ @AnsonSavage the idea is that compression should occur at the end of the pipeline, not before. So combine the images and render to the format you want. At the end of the day it will be completely destroyed by the heavy compression in youtube. Buy you will have a very nice original that you can show to your friends $\endgroup$ – cegaton Jul 18 '17 at 14:19
  • 1
    $\begingroup$ Adds are indeed done using different code paths. Hence, use the VSE for offline. Render your sequences using the compositor for full online quality. As usual, don't use the VSE for any form of pixel manipulation, including blends, dissolves, overs, etc. $\endgroup$ – troy_s Jul 19 '17 at 19:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.