4
$\begingroup$

As discussed in Colours from imported MP4 screencast video look dull in Blender, I recently stumbled across what I perceive to be an oddity with the new Filmic colour management: When working with imported 24bit images or video files (i.e. non-HDR) in the VSE and the colour management is set to Filmic, the colours in the output will not correspond to the input (see screenshots in question linked above).

This does not line up with how I thought Filmic works: My understanding so far was that (very simplified) it would scan the input for the maximum intensity and then scale everything down so that this maximum intensity becomes 1.0 (=pure white) - provided the input does contain intensities greater than 1. However, if I'm only working with 24bit input material, then this should never occur, shouldn't it? Or is Filmic not able to detect this limitation in the input material and always assumes a "buffer" on top of the 1.0 intensity?

To clarify: I already figured out how to get the desired output for the VSE (i.e. do not use Filmic). This question is exclusively about the Why of this solution.

$\endgroup$
  • $\begingroup$ I think that you may have to set the strips in the VSE to linear first as they will have a non-linear transform applied instead. $\endgroup$ – 3pointedit Jan 25 '18 at 12:02
  • $\begingroup$ @cegaton : Thanks for the clarification! Sounds a lot like an acceptable answer to me... <wink,wink/> $\endgroup$ – Oliver Giesen Jan 25 '18 at 16:45
  • $\begingroup$ @3pointedit unless the images are log encoded, the linear values will be to low to really make proper use of the filmic transforms. Only log encoded images would be properly used with filmic. $\endgroup$ – cegaton Jan 25 '18 at 17:21
  • $\begingroup$ @cegatron so the VSE should only assemble filmic images that have already been rendered? $\endgroup$ – 3pointedit Jan 25 '18 at 18:48
  • $\begingroup$ @3piontedit Scene referred images (for example a scene rendered in cycles), log encoded video or image dequenced or EXR files $\endgroup$ – cegaton Jan 25 '18 at 22:21
8
$\begingroup$

Your understanding is incorrect. Filmic blender does not scan the image to find high and low values. It does not normalize the image. There is a so-called "tutorial" out there that is spreading wrong information.

Filmic blender works with with fixed parameters. Data within those parameters gets mapped to display referred values. Unless you feed images that are in scene referred values and within the range used by filmic blender, it makes no sense to use filmic blender's transforms for the VSE.

Even if you where to linearize display referred images, you would need to bring the scene referred values to the proper range to use of filmic blender's transforms. Most images imported into blender get linearized to display linear values (0 to 1). Filmic blender works on values much larger than 1, white would be around 16.9 in scene referred values.

Your images would need to be linear EXR files (with scene referred values in the range used by filmic blender) or they would need to be log encoded, to be linearized using the proper transform to fall within the range used by filmic blender.

Please read the following link to learn more about filmic blender:

Render with a wider dynamic range in cycles to produce photorealistic looking images

$\endgroup$

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.