render layers - different lights

[pete242]

for later compositing in animations, i often render each light (or group of lights) themself for a lot of frames or if i dont know, when the should appear (maybe these lights should exact switch, with the rhythm of the music in the background) for the whole animation.

i believe, that if you would

1. render each gi for the group of lights you wanted for the animation at first (so if you have a lot of machines, each can do it at the same time and you dont have to wait for)

and then 2. render all lights in different bitmaps (so you can hold a lot of informations - like reflection / refraction at this point - in cache, that do not rerendered again) you can save a huge amount of time, i believe.

but i`m not sure, if this will work... so you have to assign each gi-map to the "layer"/bitmap output/light... is this possible and does this be really faster? i think this could be really revolution for compositing...

[pete242]

...

i just thought a little bit more about and maybe this would be great combined with some objects... so that you can get a lot of images for one frame in one time calculating a picture.

so a lot of more possiblities for postwork... esp. in animations...

[tom schuelke]

Hi Pete242

thats a good thougt. A guy in my company also wanted "lightlayers", because he told, cinema 4d could do something comparable..

Sascha Geddart talked at last with vlado, who told this could be very difficult to do for the gi algorithm.

on the other hand you are right. Also if rendertimes in case of lightlayers would screw up dramatically, i could immagine, that overal project, and compositingtime, could make it worth it.

and the new 1.5 already has very many layers.

Nevertheless this seems to be something different.

Well Geddart knows better, what vlado told. maybe tomorrow more

tom

[Geddart]

Hey pete242, greetings tom

LightLayers would be really powerfull for the users - Changing intensity of certain lights (or groups of them) in realtime would drastically speed up the workflow for certain projects.
There are two ways to implement that in a GI Renderer. First way is to do it like you did it pete242 (other users might don't know that therefore the explanation). Just render out those lights you want to adjust later on including their Global Illumination as passes. (put every other lights off) You can render as many of them as you want. When you want to see the final picture again, just add those "lightlayers" back together with the "add" mode in your compositing tool of choice.

The second method would be to integrate this function directly in the core of the raytracer. However this approach is pretty expensive in rendertime. Vray would have to add a "Lightvalue" to every Lightray it traces. This would be needed to seperate the different Lights including their Global Illumination later on. Rendertimes would increase as well as memory consumption.

We thought about Lightlayers too when we made a big pdf with suggestions for vlado and peter. (We'll post this pdf as soon as a 1.5 (test)version comes out for everybody) We already integrated them in a slick looking UI and played through several examples how you could use them in Production. As much as i'd love to see Lightlayers in Vray - i can understand vlado and peter that they don't want to integrate Lightlayers directly in the core of Vray. However - one could write a rather simple max script that automates the first way for you. You just specify the different light groups you want to render and the script does all the rest for you (putting off the lights you don't need - render a picture with only the lights you need including their GI)
Unfortunately i can't maxscript. (yet) For an advanced scripter this should be no Problem however and i'd love to see something like this for Vray.

(Even if it takes 5 time longer than a normal rendering - it might still make the whole process for the user faster cause he can do more changes interactively in post later on.)

[pete242]

so it is ok, by doing this for myself manual. i thought, if you calculate directly once a frame with all layers at the same time, you can save a lot of time while calculating, because you have a lot of the same stuff to calculate again and agin. what i wanted, would be to increase the rendertime, so that i dont need to render 5 times, i thougt it would maybe take 3 times instead to calculate. i am not sure, if this will work in theory...

with maxscript, this wouldn't be that big problem, i guess, but my intention was to get off the long rendertimes. i working in architecture and there aou allways have no time but allot of changings..

[tom schuelke]

He Pete242

well maybe it realy was a good idea to make a script about that, to solve the problem. Only a interface, to make those lightlayers, and to get the specific rendered layers. I know that the 1.5 allready will have layers, that would be perfect for compositing of diverse light groups, if you render them seperate.

Ok

and now here´s Geddarts and mine suggetion. (For all those of you, that have tooo much time and the aime to improve their scripting skills )

maybe a smal utility, that brings out the needed layers (allready in 1.5 integrated) and saves them.

maybe a way to control it directly in the viewport of vray ?




so if you´ve got a network for renderjobs, and enough time at night, this could decrease dramaticaly your trial and error time for the lightning of your scene.

[pete242]

Hi Tom,

that looks good, what you have designed. for me would the most advantage of this be, to get shorter rendertimes... especially when creating hdtv-scenes about minutes rendering. so if you render this 5 times, you will have 5 times rendertime of about 10-15 min. per frame.. so if you have 25 frame / sec and 5 lightning settings... this will maybe could be to do in maybe 3 weeks for a 10 minutes hdtv-scene by 30 xeon-dual machines... in architecture, this would be definately too long.

my suggestion is to increase dramatically the rendertime... it is much easier to do the amount of lightning later in post, if you are doing animations - like geddart explained. you can easy control the light, maybe if the music changes or if you want to simulate clouds between the sun...

if you decide to change the music or you change later the text and want to point out something different, you can just change the lightning instead of rendering the whole animation...

[tom schuelke]

he pete

here´s a nice workaround untill renderlayers are build in (if they will be build in)


this solution builds up on the in 1.5 included renderlayers.

[trixian]

What you basically would need is to split only a "illumination pass" into sepparate files for each light (without the shadows and diffuse etc). Shadows could theoretically be included, but would give you less freedom if done. Now this illumination pass would have to include primary and secondary light (If not it would have to be split up as well and things start to get messy and files large). To view this it would have to be composited with the sepparate diffuse (and shadow) pass to be viewed. In my experience, using screen gives a much more natural blending than add. Add has always been prone to burnouts and coloration-warps as it increases the luminance in the composite. (I guess this is based on how the add layer is outputed though, as with alpha channels and premultiplication to corresponding rgb values).

[tom schuelke]

Hi trixian

you are right.

what you see in my screenshot isnt all. the both immages allready use the functions of the new internal build.

both light situationsj allready are split in only the diffus color and the raw global illumination, for this gives you the freedom, to change only the lightpart of the overal output.

otherwise, changing anything also would change the diffuscolor. not only the light, and this way (with the raw gi, a fantastic thing in the new buiild), you get the lightinteraction with the diffuscolors.

now, diffus color and raw gi are composed with multiply.

what still is a problem, are in compositing the overbright areas, that lead with in between settings to wrong results. What we would like to have here is a transfer modi based on the flotingpoint results, hdri, or open exr, like vray 1.5 will be able to store.

this way, inbetweensettings could blend perfectly, incorporating also the energylevels of light.

absolut flickerfree lightchanges in animations would get posssible, as long as nothing moves.


tom

[losbellos]

i think if u've got the scene, just render with raytrace, and set the lights as u like, few try and you'll be on it. Near future there will be some global material override or something like that, as seen in the hep file, that can really speed up this, and when u got it right just swicth back.
Actually, for such composition maybe you even dont need raytrace, just that viewport prewiev, how is it called, make preview or something like that, you see on it when the light switch.