On Compression

It should be noted that the attenuation / crosstalk in Filmic is a rather brute force design for various design reasons.

One could argue that this forms a version of "gamut mapping", but I have an ideological issue with most conventional thought around gamut mapping.

Filmic, in the early going, was a very organic growth. It started with a problem forwarded by Bassam Kurdali, and grew rather chaotically to involve about thirty extremely experienced image makers. Team Argentina forming a large chunk, and spanning across to quite a few folks from other countries.

In the early going, I had done quite a bit of practical testing on chromaticity linear attenuation, "wider gamut" rendering, etc. All with less than satisfactory results from my vantage, and more importantly, it wasn't serving the audience whom became the target of the effort.

There was a reasonable chunk of folks that, while they were open to experimental ideas, were unable to change course due to pre-existing assets and general "in progress" work. As such, there was a push to make the "general contrast" feel similar to what folks were looking at on sRGB displays using the inverse EOTF encoding. Further still, on the subject of compressing of dynamic range, the subject of chromatic attenuation ("desaturation") was a very challenging topic.

I knew from the longer history prior that chromatic attenuation and crosstalk was a critical portion of what I was hoping to introduce with Filmic. The discovery of the massive complexity of this surface was apparent long before Filmic, and was made even more complex given pre-existing work. Some of the challenges included:

• Trying to not completely explode the previous look of the authored work. The chromatic attenuation had to vaguely resemble the inverse 2.2 power function encoded sRGB "look".
• Trying to not completely explode assets designed under the previous sRGB inverse 2.2 power function encoding "look".
• Trying not to have a bunch of angry Argies and French folks irate. Never piss off an Argie or a French person. Ever. The "aesthetic" of the inverse transfer 2.2 encoding "look" was something folks were simply not ready to depart from to a degree. If one tried, the aforementioned Angry Army of Argies and French folk would have your head on a spit rather quickly, with sufficient notice delivered via direct message or email.
• Maintain a degree of compatibility with the eclectic surface of tools in use.

So while I have some emotional scar tissue looking at where Filmic breaks down horrifically in terms of my current vantage, I also am willing to concede that it was simply an inappropriate time to address all known issues in one fell swoop, let alone issues that would become more apparent as understanding expanded.