Differences between operating on data before and after the LUT
In this example, we use a normal film emulation 3D-LUT taking 10-bit Cineon Log data to P3 RGB. If you operate on the data before the LUT, the color corrector is changing the Cineon Log data which is then put through the LUT to determine how it will display. If you are color correcting after the LUT, you are operating on gamma 1.0/2.6 P3 data and the look on the screen would be whatever the colorist desires with the changes. The controls will act differently since the metric of the data is different.
Operating on data after the LUT will break the "calibration"
Also, the look via after the LUT, in this case, will not translate to the film out process since the film emulation “calibration” afforded by the LUT is broken. By film emulation “calibration”, I mean that the image (Cineon Log) going through the 3D-LUT displays the look (exact match of the digital image on a digital projector to the film image on a film projector) that would be apparent if the Cineon Log DPX code values went to a recorder and then printed and displayed on a film projector.
When grading after this LUT, any changes will be apparent on the screen and if only a digital release is required, that’s fine. However, the Log DPX values did not change and if they are sent to the recorder, the film image would not look like the graded after the LUT image.
From P3 back to Cineon
If, however, the graded P3 image was then put through a P3 to Cineon DPX LUT then the DPX values sent to the recorder will reflect the grading changes. That is why when using film emulation LUT with the anticipation of going to film, the grading is done on the DPX data.
Any grading after the LUT can also produce colors that are not attainable on film, grading before assures that the color on the screen will match the film out.
Use the control settings that matches the data type
Both Primary adjustments and Log controls will work on ANY type of data. The only caution is that the RESULTS will be dramatically different. This is the domain of the color corrector.
If it is set up to control lift, gain, and gamma, then it will be most effective on linear images (either “true” linear or gamma linear where the “true” linear data has the inverse of the display gamma applied. This will result in a linear light output on the display.
If the color corrector is set up to control Log data, then it will be most effective on Log images. The overall brightness of the image is then controlled by simply adding or subtracting a constant on each channel, analogous but not identical to lift. The “gamma” of the image is controlled by a multiplier by the color corrector, since the Log data is already in a gamma metric, multiplying by a constant will change the overall gamma of the image, analogous to the gain control.
The gamma control operating on a Log data image actually preserves the overall gamma between black and white but repartitions it between the highlights and the shadows differently. Numerically, applying a gamma control greater than 1.0 will actually lower the gamma in the shadows (and darken them) and raise the gamma in the highlights (and brighten then). Gamma values less than 1.0 does the opposite. What we consider normal highlight and shadow control becomes a little different than the lift, gamma, and gain model but fairly easy to implement in hardware or software.
Bottom line is that it is best to use the control settings of the color corrector that matches the data type of the image. Using other settings may get you to a desirable look, but the colorist must be aware of the consequences. A colorist is an artist and may use any tools available as long as they are fully understood.