Convergence from my understanding just means that the successive iterations aren't changing the resultant image. It doesn't necessarily mean that images with high convergence don't have issues.I've read even your explanation several times, but still didn't fully understand what convergence is
Maybe you could rephrase it some other way?
EDIT: Okay, I think(?) I started to understand something about this word. If I set it to 95%, it would mean that if pixels in say iteration 999 and iteration 1000 are the same for 95%, rendering will stop. Something like that? And overall image quality might still be bad, but desired convergence is achieved. Eh?
But I got your point that it's a bad way to measure final image quality.
But what is required? Atm I just understand it as "if the scene looks lit well enough for my eyes and adequately to the place, it's good to go". But even if it is not, instead of adding more light with for example ghost lights, I can just bump up exposure value, or F/stop, or ISO, or shutter speed, etc, in tone mapping tab. I don't quite understand the difference.
The main trick with scene lighting is to make it as simple as possible for the rendering engine to process and arrive at a good result. HDRI's are extremely good for fast convergence as all of source light rays traced are directional and low intensity. Hence the final result is primarily a result of light bouncing off the props and straight into the camera. As soon as you add additional lights in the scene, the resultant light to the camera can be from multiple sources. Worst of all are high intensity lights, as when light happens to reach the camera from this source via a single bounce it will show up as a very bright pixel. The rendering engine also has to run a large number of iterations to evaluate what the final image result should be. Most of the lighting in my VN is from a HDRI and typically a maximum of 1 emissive sphere. This emissive sphere is equivalent to a ghost light, however I usually make it larger and lower intensity than a ghost light. Creating your own ghost light is incredibly simple and also gives you far better control. Also note that when you are using a HDRI for interior lighting you will need rotate the dome when the camera changes direction.
Sometimes this goes against what you are trying to achieve with that specific image and you need to have more complex lighting with a higher number of iterations. As an example a night time scene with lighting coming from street lights / cars etc. Also if you have a highly reflective surface you effectively double the light sources which causes problems for convergence. One of my characters also has emissive skin, which causes real problems rendering. Animation frames in which she is included take about 10 times longer to render as I have to bump up the iterations and each iteration takes longer.
There are other ways of lighting a scene for rapid convergence, such as 3-point lighting. The trick however is to make sure that light sources do not compete with each other. In the 3-point lighting system used for portraits, different portions of the face are lit by different lights. Similarly for HDRI + emissive you are better to have the emissive as a prop such as a light on a table or wall light. In the local vicinity it will dominate the lighting, however the HDRI will take over everywhere else.
You can play with the camera properties on the iRay camera, just as you do with any other camera. You just need to unlock the parameters. The only thing you need to be wary of with close ups is that the lighting panes related to the camera do not clip through objects. This will have the effect of internally lighting the object. To correct you just need to move out the offending plane.Okay, so. Here's my example. I have indoors scene, daytime. I have HDRI lightning, I use Iray Indoor camera and sometimes regular daz camera for POVs, since Iray Indoor camera can't have focal length less than 60mm for some reason (the latter requires some tweaking with lights, removing ceiling and/or walls, and mentioned tone mapping). My tone mapping settings are: Exposure value 12.75, F/stop 8, shutter speed ~100, ISO 100, other sliders untouched, if I remember correctly. I also blindly added Nominal Luminance in Filtering tab, just because I read about it somewhere here, set it to 1000 or 10000 (also, I don't see any difference at all with it). Also I have 1-3 simple sphere ghost lights around the character. They have emission value about 4000 to 20000 depending on the scene. I render to final resoultion which is 1080p 16:9.
I read about rendering in 4k and than downsampling to 1080 to get same quality or better, faster, but I usually do batch rendering overnight, so I thought - why bother.
Since before your reply I rendered with convergence reaching method, 95% convergence was reached after 1500-4000 iterations. If you are saying that I should reach good quality in less than 1000 and I don't - what does it mean, what should I do? The obvious answer for me for now is: add more light! But that the scene would look overly exposed. Do I need to add more light into the scene and then reduce stuff such as Exposure value etc? And it should lead to faster... c-converging... better converging? I really still don't understand the meaning of that word.
As far as you problems with convergence it is actually being caused by the ghost lights. By having several competing with each other you keep getting different results at the camera. If you want it to converge rapidly, just use the HDRI and no emissives. If you want a highlight on the character then use a single 'ghost light' but make it larger with lower intensity. If you double the diameter of the ghost light and drop the intensity by 4 you get the exact same amount of light, however the scene will converge more rapidly.
For each room you need to set up the lighting to be consistent, even when you change the angle. What I would suggest is that you use a single emissive sphere in the center room (about 1m diameter toward ceiling) in conjunction with a HDRI and iRay interior camera. As you change the rotation (side to side) of the camera you need to also change the HDRI dome rotation. If you want hard shadows, bump up the emissive sphere intensity and lower the environment intensity. Conversely if you want a softer look, wind up the environment intensity and lower sphere intensity. Also note that the HDRI you use should be one designed for scene lighting rather than a background. I tend to use one particular HDRI frequently as it has a combination of brown at the base, blue in upper portion and a hard white light as well. Hence when lighting a character it's like having the brown coming from the wooden floors, blue from the walls / roof and the white from the main interior light.I actually do that, again with ghost lights, but I thought that's, like, unprofessional? Because since it's a game where you click through the images, and if in 50 images in the same room the lightning changes 10 times - it's kinda strage if not annoying.
Sorry for a wall of text and thank you guys for already quite detailed answers!