That statement is true when everything is done right, not only tone mapping. There are a lot of phenomena hidden behind the hood. Caustics, light polarization, diffraction etc etc, to name a very few.
Hah I started writing response before you and then you write 90perc. of what I wanted ! Absolutely agree with the above.
I would argue that "ugly" lighting in CGI is easier to pull of photorealism, best example being flesh light. High contrast, small & directional, creating super strong highlights, placing focus on shapes instead of surfaces.
And vice versa, high-end studio photography sets look very CGI like, similarly like retouched people look almost plastic.
This is where I would caution against jumping to conclusion there is something this has to do with "tonemapping" necessarily. Your cell-phone is not doing any advanced tonemapping that wood look better than Reinhard. Every digital sensor captures data linearly, those linear data when extracted in that way look exactly like Corona with everything at 0 (and HC=1). Yet it looks photoreal because of the amount of detail.
The opposite is also true to some degree, that very good tonemapping & grading can make simple & lazy CGI set look photoreal...but only in low-res and only for a moment.
So both of those aspects are important, something Lolec and few others brought out. It can't be tied to tonemapping.
Also...the mythical "DSLR-like" tonemapping. It's not imho what people think it is. What your camera does, it only applies gamma curve onto linear data, then raw convertor applies S-Curve (either its own, or the one described in raw file, which is usually based on JPEG post-production by camera maker) and their own bit of tonemapping which you have available in the settings on front page of ACR for example. At that point, camera hasn't done anything, the Sony sensor (whether sold to you by Sony, Nikkon, Fuji, etc..) has only provided linear data. No tonemapping whatsoever. They do attach associated s-curve into the file (which the raw converter only interprets! the curve is only identical if you save directly to jpeg from camera).
So if we want "DSLR-like" tonemapping, there would have to be way to use any raw converter tools. But there's a catch, they don't expect the amount of information provided by CGI. Raw formats are barely 16bit at best, already with gamma curve.
If you force-open 32bit file in Adobe Camera Raw, it will do random equilizing, which can look almost bizzare depending on dynamic range of your scene. This is the only "magic" it does to photos as well... but the algorithm obviously gets very confused by the kind of data it gets with rendering.
You can do the old trick of zero-ing everything in older process setting, then you can use the default state of rendering and only apply tonemapping that AdobeCameraRaw offers. It only affects highlights.
So photorealism really isn't that simple to define as there is a whole lot of intricacies.
Yeah, very much this.
There is interesting research done by Paul Devebec & Co (or John Hable? I am not sure, I came across it on FilmicWorlds which is his blog) where ultra-realistic models (like the most high-end Hollywood VFX stuff we can do) still looked wrong to our eyes. Then they flipped them above to fool our brain finally. The evolutionary amount of sensory recognition is so well trained, that we perceive almost minutae differences between real humans and CGI humans. Even if we are talking super rendered, super lit, super post-produced photo-scans with perfectly shaped-in emotion.
I believe this effect manifests to much higher degree in architecture as well as we would like to admit. After all, how much time humans spend inside some kind of architecture? Same with nature. It always looks somewhat wrong. Opposed example would be cars, recent innovation, rather simple in shapes. I think almost nobody can tell difference between CGI or real car... even not exactly great renderings.
Every time people congratulate something for looking very photorealistic in forum, there is some crutch that strongly helps it. Super massive DOF, odd angle focusing on detail, etc. Lot of time these projects have 4-5 very "real" looking images...until the last one, with flat lighting, wide angle showing the whole room that client actually requested. And it doesn't look real at all. With no hack to help it, the same grading & tonemapping suddenly doesn't help at all.
Can most artists be sure when the issue can be down to tonemapping & grading, and when it's the detail? I would say this ratio is lot different than people think.