Hi lucifer,
Obviously the .tx are not only convertible on Win, it seems from the scene these files are also not working on my Mac. Well, it crashed my app, after giving me a render error. So, no further data on that. I will report that.
The Open EXR content looks really like it was gamma based. It is not really recognized at all. Something is wrong. So I checked the export and it needs an sRGB to Linear inside of Substance if you like to have all the quality that Substance obviously has. If you go over (mouse wise) the image in Substance, in the left lower corner you can see the integer as well the float numbers then as in 0.028, which indicates something very differentiated. I never would go ahead and do the translation later, keep things organized. So, please a sRGB to Linear node inside of Substance. Perhaps check with the support. I guess the highest precision is given in the way I suggest here. I tested it of course.
In floating point formats, images are linear. Since I guess that the conversion is one time in Substance but often in Arnold, that would make sense anyway, to fix it from the start.
There is nothing really fancy on the ACES way, it is just a large color space and linear. I mentioned it, as since years they try to get something that unites all. Color spaces allow during production to go in any direction without being clipped. You just keep the quality high, with integer stuff, the worse can happen that one chancel gets clipped and with that any change will change the color, as one channel isn’t coming back, simply put.
Please note that the Substance, so far I can see is producing values between 0 and 100% for the"texture/image based part, but in float precision! Which would be sufficient to be stored at 16bit/channel, integer, if no alteration happens, as in change of color space. But any data channel, as in Normal, would not be well represented in this way. As I get not tired to say, sRGB or REC709 is a delivery, but not a production format. This is sadly not understood in large areas. The pixel values are converted based on the color space. If something is not inside of this colorspace it becomes clipped. There is no “healing” from that clipping. Something that is missed in many tutorials who suggest that “convert it back and fore” with no penalty, which is just wrong. We go straight to the BT2100 or REC 2020 generation of material, and HDR-footage will be possible, I do not talk about the tone-mapping that is often misunderstood as HDRI. Any limitations in the footage then presented, will look like mud besides the new options. Something I talk about since years, but then again, people come with: we do it since ... Well, I can only inform that things become visible that weren’t visible on the screens before. So what you see is not what will be seen later. I try to make things a little bit more future proof here, but obviously there is a lot of old “knowledge” that prevents a success of my doing here.
At least eh data channels in Multi-pass have lead to an more and more use of linear based formats, as in Open EXR. When ever one notice sRGB or REC 709 in a current workflow, the alarm bells should ring. Hence my suggestion to fix the sRGB in Substance, and not in Arnold.
The image below show the simple way, I guess a deeper look into “functions” would allow for a better representation of the values, while going “float” before processing anything else”, perhaps then we might see wider scales of values. But this is already a little bit out of the scope what we can discuss here, after all we connect to C4D or deliver from it. Please check out the Substance forum, to get more precise information. I hope I could illuminate the problem area here, so the work inside of C4D will gain the best quality. After all, Arnold seems to ignore at least my support ticket, as well 3rd party and most C4D shaders (certainly to stay native and fast, no complains here!)
All the best