Strange Cycles Light Behavior (have to blow light intensity to light scene)

Hi, friends

Today I decided to figure out the the reasons of a little bit strange (in my opinion) lighting calculation in Cycles.
We all know, that Cycles is physically based unbiased render. So as I understand it calculates light transfer according to some physical laws, which in the end gives us more realistic results.

I also hope that Cycles calculates all lighting linearly, according to linear color space workflow.

So during a long time working with Cycles I noticed, that I always have to overbrighten the lightsources to lit the scene with appropriate amount of light.

Look at this image
https://docs.google.com/file/d/0ByHRy_pSYAhgd1hQZy1Pcl9tX2M/edit?usp=docslist_api

Notice how overblown the lightsources is. I had to gain the lightsources strength 3x times to lit the cloth on the background. Notice, how unrealistic (in my opinion) tell image looks. The lightsources is overbrighten, and there were also have this overbrighten light spot on the floor under the lightsources.
Now look at this image
https://docs.google.com/file/d/0ByHRy_pSYAhgSGdFeWJCb2Z2ejg/edit?usp=docslist_api

This image and the previous one looks identical exapt the lightsources. It is not overbrighten now. In this image I left the strength of the lightsources at one, but I multipled all the colors in the materials by 3. In this case everything looks realistic exapt the overbrighten light spots under the lightsources.

Now, look at the third image
https://docs.google.com/file/d/0ByHRy_pSYAhgZE1lVTBjNEpJR2s/edit?usp=docslist_api

Notice, how realistic (in my opinion) it looks in terms of light behavior. Here I did the same operations with the colors, but I also added the Light Falloff operation to the lightsources with the smooth 0.2.

And looking at this 3 images, I’m asking myself, why it is so? Why I have to overbrighten the lightsources, or gain the materials colors extremely to get the appropriate lighting in the scene.

It is very important to me, because in this situation I have to use much more lightsources to lit the scene, I also receive much more contrast results in this case, and it is much harder to get realistic results without these tricks. All these symptoms saying to me, that the deal is in the linear workflow in Cycles, and looks like something is not working properly in it. Or maybe I don’t understand something.

P.S. There is no indirect light in the scene. Only direct.

Here’s a challenge for you. Grab a digital camera (your phone, DSLR, whatever) and go into a dark room that’s lit only by a lamp. Take a picture of this lamp, and try to correctly expose the lamp itself and a nearby object. You’ll get the same “problem” you see in the first render you posted. A standard 8bit image has much less dynamic range than our eyes do, so to be able to see objects and not have their illumination source blown out requires some response curve trickery in either the image, or the lights themselves. (if your phone has an HDR photo mode, it can do this automatically and might be able to expose the lamp and the object correctly in the test).

If your prefer the look of your third render, by all means go with it. “If it looks right, it is right”. Whether you want something to look like as you’d see/draw it, or what it would like if you photographed it, depends on your goals. That’s why you have the tools to get either look.

And on the subject of linear workflow, it’s working correctly assuming you have an sRGB display and didn’t change anything from the defaults.

Part of this has to do with bit-depth Cycles calculates RGB in, which J_the_Ninja explains to some extent. This is why the light-path node controlling an emission shader mix is something to make use of. It’s easy to have light going directly to the camera blow out the simulated sensor, while the rest of the light cast into the scene is not enough for general illumination. (Also problems with aliasing on light sources at or greater than 1.0 as the camera sees them. It still drives the interpolated values where AA should occur to the high end, and jaggies show up. Perhaps something to do with averaging rather than mean, and some aspect of when value clamping occurs in the calculation.)

It seems like you arrived at an alternate solution. A mixture of both probably doesn’t hurt. As long as it looks good.

And yes, getting good light in Cycles does sometimes involve a bit of trickery. Even more fun getting it to behave when going through glass or translucent materials while still wanting caustics in the scene. Not just lights themselves, but particular materials have to have nodes setup in order to behave nicely when they interact with light. (They’re not quite “out of the box” in working as you’d expect.)

Thanks to your replies

Can you explain deeply what combination of lightpath node, mix, and emission do you mean?

You can fix lighting with color management->curves or in compositing nodes editor (tonemap node, custom tonemapping nodes).