Maxwell Demonstrates Realtime Interactive Lighting

[Cerenkovman]

Holy crap! It may be that I'm just new to all this but that sounds really cool. Especially since I was under the impression that Maxwell was somehow substandard compared to renderers like MR and Vray. I like.

[Jozvex]

Not substandard! Just a different creature alltogether.

[mverta]

Maxwell's still in development, and its parent company, Next Limit, has taken a serious beating, PR-wise, for a host of reasons - some legit and some not. But at the engine's core is some serious potential, and this latest feature I think is a good example. Even traditional R-G-B channel-lighting for post-process control requires multiple passes, and doesn't carry things like caustic control or reflection with it without additional passes. This is the best implementation I've seen, to date.

_Mike

[tikal26]

Mike:
Youexplanation is pretty clear. Maybe they should of posted you explanation with the video so that there was no confusion, but anyway thanks for the explanation

[lexington_luthor]

Its interesting, and hopefully will be useful, but I would love to see how it compares to lpics in terms of features and performance.

http://graphics.cs.ucdavis.edu/~lefohn/work/lpics/lpics_siggraph.mp4

[UrbanFuturistic]

lpics, I think, is designed more for making assorted lighting decisions on test renders before rerendering, hence the frequent comparisons to offline rendering. It's not really something that includes refraction, caustics or any particularly advanced lighting effects... or even soft shadows by the looks of it.

I do have a question: do extra lights still incur a time penalty in Maxwell? I wonder because it would be spectacularly nifty to be able to light a scene with all the different lighting conditions at once and just switch them on and off.

[tikal26]

According to what they are saying in NL forum it does not slow it down or hinders it.

[Cronholio]

So this is really not much of a feature at this point, I mean all you can do is adjust the intensity of the various lights individually, and maybe eventually adjust some other parameters of the lights. In a system like LPICS or IPR Caching, you can adjust virtually every shading parameter of every object and light in the scene including swapping out shaders etirely, the only thing you can't do is move objects or the camera. You can move the lights, and add/remove objects and lights that exist in the cache, though. Your ipr image can be full res, and at final frame quality (yes caustics and shadows, etc, can be included but you will be limited as to how lights can be moved while using them, if they are not cached on disk). While not "realtime" the frame is rerendered in seconds or fractions of a second, and you can work with sequences of animation or different takes, so you could, say, light the same scene for different times of day/night in one go.

I'm not trying to bash on Maxwell because it is a work in progress, just pointing out that if you think this is the best implementation you've seen, you've probably not seen much.

[Jozvex]

Doesn't LPICS use a renderfarm of really good computers to do that though? I thought you needed a farm all helping at once.

In Maxwell you just render and then tweak all the unbiased GIness in realtime on one PC.

[Cronholio]

Doesn't LPICS use a renderfarm of really good computers to do that though? I thought you needed a farm all helping at once.

In Maxwell you just render and then tweak all the unbiased GIness in realtime on one PC.

In LPICS (which I should say I haven't used) and IPR Caching (which I do use) what they do is generate a cache, which could take a renderfarm depending on what you are putting into it. Once you have the cache you can tweak all you want and render in fractions of a second on a single workstation. Basicly they precompute as much of the scene as they are able while still allowing you to change important parameters for lighting and shading. If you watch the Maxwell video, it sure seems like the Maxwell guys are trying to do exactly the same thing but they aren't achieving the same results. They have those "Cooperative MXI files" which they merge and then load so they can relight the scene. With LPICs or IPR Caching you just have a single cache file full of scene information. Those Coopperative MXI files seem to be image caches, I'm sure they took a while, probably a very long time to generate given everything I've heard about Maxwell. But the fact they they seem to generate one for each light in the scene makes it look like they aren't really caches at all but simpler image formats which would limit what you can adjust in the scene. The demo shows exactly what can be done with any half decent compositing app by rendering a layer for each light.

[Spacelord]

Wow this is awesome !
Gee next Maxwell will be able to adjust
Depth of field in realtime !!!

It does seem pretty limited at the moment.
You obviously can't add lights but its a step in
the right direction.

good work Next limit. Wheres my XSI plugin ?

[tikal26]

Cronholio
Acoording to the explanation here and in the Maxwell forum you don't have to render layers. All that you have to do is rende one scene You do not need to do coperative rendering ( tha is supposed to be another feature).Also, i think that it is a good addition to a rendering software. I am wondering if the software you mention is a rendering software or it sits on top of the modeling, rendering software?

[olbo]

As stated, the video shows two different features:

1. The cooperative render function combines multiple render results (MXI files) of the same frame in one single MXI file, producing a new result with higher sampling level and less noise. This process does not require additional calculations and helps to reduce dramatically the render time, using several computers to render in parallel.

2. The multiple light edit feature allows adjusting the intensity of individual lights at any time. This feature does not require several passes or additional render time. Multiple light channels are available within a single MXI file for editing. This process can be done in realtime and each light can be modified independently. This is not a color mixing process, the intensity variations are energy based and are calculated in high dynamical range. Power emission can be scaled naturally.


So, you just render your image and tweak the light(s) as you wish while rendering
and/or at the end ... :p :D ;)

btw. lpics is a system to light your scene and tweek your shaders, it is not a render.

nb: I love the new feature of maxwell! :D :D :D



take care
Oleg

[Jozvex]

Yes tikal26 is right, the multiple MXIs is for the coop render but completely separate from the relighting (which works with the daylight stuff too I think).

The coop is like distrubuted rendering in that multiple computers work on the same render, except because M~W isn't a bucket renderer it doesn't DR in the usual way. As far as I know a relative in Spain (for example) could contribute time to rendering the scene, send you the MXI and then it improves the quality of your current image that you yourself are rendering.

[saftan]

Hey guys!
I just visited their homepage to check out the promo-video, and left with a couple of question marks..
Am I the only one who thinks the car in the video (where the camera is panning) is moving sideways???... it almost seems to float away from the camera!

Makes you wonder, why they chose that clip for the promo-video???

hmm...

[Cronholio]

The reason I assumed the MXI files were actually light layers is because he has five of them, and five lights in the scene, but I think I understand now...

OK so the MXI is a precomputed part of an image, right? So you do have to render *something* before you can use this relighting tool, the same as if you used LPICS or IPR caching or just used a compositor to relight a shot. In this Maxwell scenario you have to render at least a rough version of the image, then keep adding MXIs to it to refine it. Looking at the video, he's not relighting a render in progress which is what some people here seem to be suggesting. He has an image and adds the precomputed MXI files. Adding the precomputed MXIs improves the granularity of the resulting image, however, it still a long long way from being a final frame, and I'm sure they took considerable cycles to generate the MXI files. I don't really see how this can be considered an improvement over LPICs or IPR Caching or even rendering multiple layers and relighting in a compositor. In all those scenarios you can move, add and remove lights, in addition to manipulating their parameters and the parameters of the surface shaders. It's interesting in that you can apparently keep the image caches you generate and add to them for progressive refinement and the final output, but you can already do that in to some extent with the other methods people are using.

If you can actually move the lights in the shot and change shaders while using the existing image and MXI files I'll be impressed, but I'm inclined to believe that's not possible because of the nature of Maxwell as a renderer, and the fact that he's relighting a full image, not a cache; unless someone can demonstrate otherwise. A relighting tool isn't worth a whole lot to a lighter if they are limited to just manipulating existing light parameters in their current positions. That's kind of the whole point of these other relighting systems, the ability to make sweeping changes to the light setup and parameters of the lights and shaders. If you can't move the lights in Maxwell's "relighting" system, you are going to be stuck in the usual cycle of setting up lights, test render, revise setup, test render, on and on until you have a setup that works. Then you can move on to the relighting tool and generate additional MXI files. If that's the case, it probably won't save you much time if any.

[mverta]

No, you're wrong.

.MXI is just Maxwell's native HDR file format, not a special image cache or anything like that. The video shows 2 SEPARATE, UNRELATED processes. One process is the recombining of cooperatively rendered frames into a single one. The entirely separate, unrelated feature being shown is Maxwell's realtime interactive lighting function. You render; just render like any other time. You render your final image; no special settings, no accommodations, except that during or after the render you can completely change the contributions of any and all lights in the scenes, including their caustics, etc.

So for a typical workflow, you could set up a bunch of key and fill lights with arbitrary values, and just pick your lighting as the render goes, or afterward, as many times as you like.

You could render a house in the daytime, for example, and just by moving a couple of sliders, turn off the sun, turning on the interior house lights and have an entirely new image without re-rendering. You can do this during a render or afterwards. There is no impact on rendering time, or workflow, beyond the fact that it just saved you about a billion hours in tweaking.

_Mike

[olbo]

@Cronholio

Hmm, ... I have to disagree, it is mindblowing to have the possibility to change
the lightsettings. I can change the mood inside my rendering as I wish ...

I can change from daylight to night, turn lamps on and off ... it is amazing,
I don't have to rerender the picture! It saves me a lot of time, this is a fact.

btw. another fact is that you can change all this while rendering too. ;)

edit: ... Mike beat me to it. ;o)



take care
Oleg

[Per-Anders]

um... sorry, but am i missing something? to me it looked like he loaded in 5 files to generate a new composite, then just changed the layer opacity/exposure... that's not relighting, that's just changing layer opacity. you can do that with any apps output and photoshop. the only difference is here it's seperate gi emitters for layers rather than traditional lights (in most other apps), which is nice for sure, but still not relighting, there's no ability shown to change a lights position in realtime, it's size, falloff charactaristics etc.

[mverta]

Yes, you're entirely missing it, actually. The first step had nothing to do with the second. Do you guys even bother to read posts or do you just start typing? He's not loading 5 files to do the lighting thing. That's something entirely separate. Read my last post.


_Mike

[olbo]

@mdme_sadie

You haven't read the thread, do you ? :p :D ;)

Part of the official announcement:

As stated, the video shows two different features:

1. The cooperative render function combines multiple render results (MXI files) of the same frame in one single MXI file, producing a new result with higher sampling level and less noise. This process does not require additional calculations and helps to reduce dramatically the render time, using several computers to render in parallel.

2. The multiple light edit feature allows adjusting the intensity of individual lights at any time. This feature does not require several passes or additional render time. Multiple light channels are available within a single MXI file for editing. This process can be done in realtime and each light can be modified independently. This is not a color mixing process, the intensity variations are energy based and are calculated in high dynamical range. Power emission can be scaled naturally.

edit: ... haha, ... Mike you did it again. ;)



take care
Oleg

[Per-Anders]

um, i think i must be dumb, because i did read what you posted. but it just looks like what i said. the composite is part one asyou said. the second part looks like he's just manipulating the intensity of the composite layers (nice that it's allowing you to do it while the render is going, but nothing astounding in itself, any engine can render multipass in a single sweep), and there's nothing showing the lights moving around, just what appears to be the intensity of the light layers changing. so i must really be missing something here.

[ThomasAn.]

Yes, you're entirely missing it, actually. The first step had nothing to do with the second. Do you guys even bother to read posts or do you just start typing? He's not loading 5 files to do the lighting thing. That's something entirely separate. Read my last post.


_Mike

You guys should have split that video in two. A lot of this confusion would have been avoided. Now, you spend a lot of time just clarifying...

[olbo]

You guys should have split that video in two. A lot of this confusion would have been avoided. Now, you spend a lot of time just clarifying...

I guess you are right ... :rolleyes: ;)


take care
Oleg

[CupOWonton]

Hm, looks like its simply compositing whats already there, and you arent so much able to 'edit' the effects of a light completly, as much as youre only able to adjust their intensity and color, something rather do-able with compositing and layers. Not exactly new or innovative. It doesnt look like its going to magicly generate new caustics.

[ThomasAn.]

um, i think i must be dumb, because i did read what you posted. but it just looks like what i said. the composite is part one asyou said. the second part looks like he's just manipulating the intensity of the composite layers (nice that it's allowing you to do it while the render is going, but nothing astounding in itself, any engine can render multipass in a single sweep), and there's nothing showing the lights moving around, just what appears to be the intensity of the light layers changing. so i must really be missing something here.

It may look like it from the video, but it is purely coincidental. There could be 3 mxi files merged into one (they didn't have to be 5) and there could be 58 emitters in the scene. Merging was not a prerequisite for the interactive emitter adjustemt.

He could have opened just one of the mxi files (the most grainny one) and adjust the emitters of that grainny single file.

Whoever did that video was a little lazy and jammed both features in a single video file. So, now many people think that one feature was a prerequisite step for the other.

[Per-Anders]

well, the merging caches part certainly seems a good idea, though can this be used as part of a distributed rendering solution? i.e. you have 5 machines all set up rendering, then on a server machine you get an image updating as and when with the composite of the 5 nodes? now that would be a very useful with as intensive an engine as maxwell (and if so then well done for an elegant solution). or is it only a post treatment in this scenario?

[mverta]

Yes, you are still utterly not getting it.

I don't have any puppets; I'm not sure how to make this any simpler.

But I'll try: Forget the compositing; that's a different feature and UNRELATED TO THE LIGHTING DEMONSTRATION. It's not related to the lighting demonstration. The assembly of multiple images is not related to the lighting demonstration.

The lighting tool does not require compositing. The lighting demonstration does not require compositing. It is not compositing passes and controlling the "blend." You are not seeing pre-rendered passes being blended.

To clarify: realtime adjustment of lights in your scene does not require rendering passes of any kind. You don't have to render passes. You don't pre-render passes, you don't render caches, you don't composite, you don't pre-composite, you don't assemble images. You just render; only you can fully adjust your light intensities at any time. WITHOUT PASSES.

Did I mention this does not require compositing of any kind? This also does not require rendering passes. There is no compositing, nor blending of separate passes being employed.

Passes are not necessary. Assemblage of "multiple passes" is not necessary. No pre-rendering of any kind is necessary. You don't actually have to pre-render anything! And what's more: you don't have to pre-render any passes! Furthermore, no pre-rendering of passes is required, and no compositing takes place.

[olbo]

@Mike

Well, ... I'm almost getting it ... ;o) *SCNR*



take care
Oleg

[Cronholio]

mverta thanks for clearing that up.

I do have a question though. How exactly do you see this saving you a "billion" hours tweaking? You will still have to setup your lights, start a render, kill render, add/remove/reposition lights, start another render, kill, tweak shaders, start a render, kill.... until you have a good light/shader setup, and you will still have to manually tweak the light parameters while the image renders or after. Consider also that while you are doing this Maxwell takes considerably longer to render anything worthwhile than the average renderer. It might be nice to be able to tweak the lights while you are looking at the render in progress, but the manual effort you put into lighting the shot is the same, and the total time spent on the processor will likely be considerably longer than with any other renderer/method to achieve the same/better results.

[Per-Anders]

well as far as i can see it's what i'd call a pass, doesn't matter if it's internal to a file or a visible layer. i think we're facing a difference of language here. a pass doesn't have to be rendered at a seperate time, it can be rendered simultaneously, most engines can render all lights in one go, but give you access to each individual lights seperate light, shadow, specular etc... through the resulting multilayered file (be it tiff or psd or propprietary, or whatever), that you don't see it as what you consider a pass (i.e. a seperate render that happens on it's own), doesn't mean that it isn't a pass (as far as i'm concerned it's a pass, even if it's only done on a per fragment, sample rather than a whole image at a go).

[alexyork]

Yes, you are still utterly not getting it.

I don't have any puppets; I'm not sure how to make this any simpler.

But I'll try: Forget the compositing; that's a different feature and UNRELATED TO THE LIGHTING DEMONSTRATION. It's not related to the lighting demonstration. The assembly of multiple images is not related to the lighting demonstration.

The lighting tool does not require compositing. The lighting demonstration does not require compositing. It is not compositing passes and controlling the "blend." You are not seeing pre-rendered passes being blended.

To clarify: realtime adjustment of lights in your scene does not require rendering passes of any kind. You don't have to render passes. You don't pre-render passes, you don't render caches, you don't composite, you don't pre-composite, you don't assemble images. You just render; only you can fully adjust your light intensities at any time. WITHOUT PASSES.

Did I mention this does not require compositing of any kind? This also does not require rendering passes. There is no compositing, nor blending of separate passes being employed.

Passes are not necessary. Assemblage of "multiple passes" is not necessary. No pre-rendering of any kind is necessary. You don't actually have to pre-render anything! And what's more: you don't have to pre-render any passes! Furthermore, no pre-rendering of passes is required, and no compositing takes place.

:rolleyes: OK... i think we got it, Mike. People showing genuine interest in your product need explanation not patronising.

As for these features, they both look fantastic, certainly. It's about time coop rendering worked and this will surely bring a lot of users who abandoned Maxwell back. Speed is one of the main complaints of most MW users, so now that coop rendering is working this will open up the possibilities for proper full-res commercial renders.

The interactive lighting also looks great and will surely be of use. I can imagine that some artists might choose to render scenes with loads of lights and then use this tool to turn some of them on and off, effectively sculpting their lighting rig. When they're happy they can then go back and remove those emitters to keep the scene tidy and the renders optimised. This has great potential.

It's good to see NL attempting to fix old problems (coop) and introducing new features at the same time. In short, it's good we're finally seeing a balance at NL.

[rocarpen]

I'm showing this to my boss tomorrow. We're on VRay right now. I'd love to start experimenting with Maxwell a few hours a day.

I think I can clarify the two sides of the argument, too: a renderer like C4D can spit out your image with individual elements (including each light) each on their own layer. Tweak opacity and blending in Photoshop, and you've got a compositing process that seems similar to what we view in the video.

But Maxwell's solution is so much more. We seem to be talking about real-time interactive global illumination and caustics here. Adjust a light, and the scene reactes realistically. Obviously this is way beyond screwing with opacity levels on specular passes. I can totally change the look and feel of a scene in a snap.

As artists, we're still going to need to have a vision for our scenes (7:00AM summer morning, clear dawn sky, dew on the grass, etc). This tool I think is basically a sophisticated update of the 'render to PSD' workflow. I like.

[Cronholio]

um... sorry, but am i missing something? to me it looked like he loaded in 5 files to generate a new composite,

That's the problem I had. He loads these five MXI files then starts flipping trough the lights making it appear like there are 5 passes in one file. What I'm guessing now, based on the information that has been given, is that the MXIs are actually strips or parts of images similar to buckets, and the combiner is just stitching them together... Which I hope is not correct because it's be really silly to have to manually assemble every image you render on multiple machines, whether they are buckets, strips, caches, or layers ;p I kind of liked my original preception that it was some sort of cache you could progressively add data rendered on other machines to for refinement. I think they should abandon this current tool and work on that.

I'm sure Mike will be along in 10 minutes or so telling me how wrong I am :) Maybe he'll draw some pictures and post some charts and graphs or something.

[mverta]

My suspicion is that people will become spoiled by this feature almost instantly. But Maxwell is surrounded by a ton of myth, and very little field experience. The vast majority of nay-sayers have never even used it, let alone learned to master its unique approach.

What I can tell you is that Maxwell's material/lighting system behaves so predictably, thanks to its strict real-world model, that after awhile, you barely need to run render tests to see how a material is shaping up. I can get to the 90% point on a scene now, with textures and lights the way I want them, without ever having to render. And that's because in a given CG room, for example, modeled in real-world scale, I know approximately what size light(s) I would actually use to light it with, so I just plug those in. Real-world wattage/efficiency values for lights translate, as do camera ISO settings and f-stop. They actually mean things in Maxwell, and are calibrated so as to very closely replicate real-world behaviour. Add the extremely predictable material reflectance model and we're talking about a fraction of the time in setup. To say nothing of the fact that GI calculates with physical accuracy, flicker-free, in both network and cooperative mode, with the best AA I've ever seen. Add this realtime light adjustment on top and you've saved another huge chunk of time. And you're going to need it, because Maxwell's speed is the price for all this. And even that is getting better; the engine hasn't been optimized at all yet.

It'll be curious to see how things shape up.

_Mike

[Jozvex]

Oh dear I can't believe how badly this announcement turned out...

I want to point out in case people don't know, that Mike is a lead (probably the leadest?) A-Team tester on Maxwell. He's not interpreting the video at all, he works directly with the latest features and the developers. I'm an A-Team tester too but I'm not quite so super high up yet as to use that specific new tool hehe.

It really isn't a basic compositing system at all.

Example: I create a 3D house. I put 50 lights in it, and have a sun and sky. All the lights are turned off except for the sun and sky. I render the image 1 time and get a sunlit house. Now, from here until eternity I can turn on/off/adjust any of those 50 lights (and sun/sky) creating entirely new lighting with unbiased GI and caustics and reflects etc etc in realtime.

My boss says: "You know we could use a nightshot with light through the bedroom window instead."

I say: "Sure" And in realtime, turn off the sun, and switch on the bedroom light and adjust it's brightness to get the right effect. All from the 1 original image I rendered. New caustics, new GI, new reflections of different parts of the house etc.

[mverta]

... I suspect my bias was evident :P But people may not realize that I support Maxwell because of the results I get, not because I work for Next Limit; I don't. I just get to play with all the new toys sooner. :D

_Mike

[Coliba]

most engines can render all lights in one go, but give you access to each individual lights seperate light, shadow, specular etc... through the resulting multilayered file (be it tiff or psd or propprietary, or whatever),

So you're saying C4D can render a GI scene with each lights influence on the GI calculation separately, in layers, and when you turn on/off one of these layers, the GI calculation in all the other layers magically update as well? The entire scene lighting updates? For example one of the caustics that this light is creating, changes intensity depending on how you adjust the opacity of that lights layer in Photoshop? Or a shadow cast by this light (which is most likely in a shadow layer) changes when you change the opacity of this lights layer? Or that you can turn a GI daylight scene into a GI nightscene, by just turning off or on a bunch of layers?

[GARJones]

"This is not a color mixing process, the intensity variations are energy based and are calculated in high dynamical range. Power emission can be scaled naturally."

The thing that sets this apart, surely, is the fact that each light and indeed the whole scene is rendered in MXI format (pretty similar to HDRI). Doesn't this mean that unlike a normal multipass render, where tweaking the intesity of different lighting passes results in a loss of detail due to the narrowness of the RGB gamut that the scene is rendered in, Maxwell will allow you adjust the light in your scene without the worry that if you make any serious adjustments to any of your light-sources you're going to lose all the contrast.

And just on a quick afterthought, yes, you still have to position all your lights, but doesn't this feature mean that you don't have to spend any time messing around with their intensity or doing test renders to determine how strong the GI calculation should be? Sounds good to me.

[Jozvex]

but doesn't this feature mean that you don't have to spend any time messing around with their intensity or doing test renders to determine how strong the GI calculation should be? Sounds good to me.

Yes that's exactly right! Also, in real life of course light is energy and loses strength over distance etc. No matter how high the opacity is on a traditional key light pass, the light isn't going to reach further into other rooms of a house, illuminating new elements that didn't show up in the key light pass to begin with.

[Cronholio]

"This is not a color mixing process, the intensity variations are energy based and are calculated in high dynamical range. Power emission can be scaled naturally."

The thing that sets this apart, surely, is the fact that each light and indeed the whole scene is rendered in MXI format (pretty similar to HDRI). Doesn't this mean that unlike a normal multipass render, where tweaking the intesity of different lighting passes results in a loss of detail due to the narrowness of the RGB gamut that the scene is rendered in, Maxwell will allow you adjust the light in your scene without the worry that if you make any serious adjustments to any of your light-sources you're going to lose all the contrast.

And just on a quick afterthought, yes, you still have to position all your lights, but doesn't this feature mean that you don't have to spend any time messing around with their intensity or doing test renders to determine how strong the GI calculation should be? Sounds good to me.

Most professional renderers nowadays support at least one, often multiple full floating point formats, I think my renderer supports at least 5 or 6 32bpc formats, and several 16bpc formats. In addition, as Madame Sadie mentioned, you can shove multiple passes generated from whatever arbitrary light and shader variables you want into a single file. You don't have to fire off fifity renders for fifty passes and comp them later. Just click render and they all render simultaneously without incurring penalties.

The point I was making is, I think that the hard part of lighting to me anyway, isn't in the tweaking of light intensity and color, but in getting the appropriate ammount of light into the scene and casting light where light needs to be cast to direct the viewer. Never mind that you are extremely limited to the type of imagery you can create in Maxwell.. In Maxwell you can throw a bunch of 40 watt bulbs and flourescent lights in a scene and they work like those lights work in reality, great. But what do you do when the client says "I don't llike the harsh shadows I'm seeing under his eyes, we're losing the performace. Can you quickly relight this in comp?", You won't be able to if you didn't render with a light that can fill the area, and you have no layers to work with in comp.

[mverta]

How much work have you done in Maxwell?

_Mike

[Per-Anders]

So you're saying C4D can render a GI scene with each lights influence on the GI calculation separately, in layers, and when you turn on/off one of these layers, the GI calculation in all the other layers magically update as well? The entire scene lighting updates? For example one of the caustics that this light is creating, changes intensity depending on how you adjust the opacity of that lights layer in Photoshop? Or a shadow cast by this light (which is most likely in a shadow layer) changes when you change the opacity of this lights layer? Or that you can turn a GI daylight scene into a GI nightscene, by just turning off or on a bunch of layers?

no not gi and i'm not beliteling that at as it's great that m~r can do that as not much else can (mostly because of the way m~r's gi works).

however all the multipass lights in one go in non gi, then yes (as can many others, and i did originally point this out in my first post in the thread, i.e. apart from the gi). as far as I can see it's doing no more than that, but with it's lights that bounce.

maybe a better explanation would be as i'm seeing it : imagine rendering each light seperately with it's gi, then putting it into photoshop with each layer in linear dodge mode. changing the fill opacity or modifying the brightness of the layer would give you the same effect (try it if you don't believe me).

i know they say it's not a color mixing process, but perhaps it would be good if there were some side by side comparisons, between a layering in photoshop approach, versus their relighting approach, i.e. to really see the difference, the video is maybe just too compressed and low res to show any such subtelties.

maxwell is taking out the time and hastle of rendering seperately for the gi because it can, which is awesome but it's still not relighting (as far as I see it), you don't seem to have any seperate control over any parameter apart from the brightness, but relighting to me involves the movement of light sources, and changing of more than just brightness.

[holyzombiejesus]

Its awesome, theres no two ways about it. If I could post this on the Maxwell forum I would (soon though :D .) Its a shame alot of people seem to be misunderstanding the process but as soon as it gets released (or we get a few more video updates...hehe) it'l become clear. This'l save alot of people, ALOT of hours. Nice work guys. And cheers to Mike, Mr P.R :buttrock:

[pluMmet]

Hey mike- Lots of people read the first post then breeze thru the rest and start writing. Especialy when it gets long. Please edit your first post with the disclamer that there are two unrelated actions in the video.

-E

[Coliba]

maybe a better explanation would be as i'm seeing it : imagine rendering each light seperately with it's gi, then putting it into photoshop with each layer in linear dodge mode. changing the fill opacity or modifying the brightness of the layer would give you the same effect (try it if you don't believe me).


It wouldn't be the same, when you change a lights intensity it is not a linear falloff. You would get totally different results between just changing the opacity of one lights influence in a layer (no matter what blending mode you use) and actually changing the intensity of the light and thus completely changing the GI calculations (be they biased or not).

[mverta]

The FIRST line in my post is:

In this video, 2 separate features are being shown

If somebody's interested enough to be responding at all, they can wade through a couple of sentences and get an idea what they're responding to.

_Mike

[pluMmet]

The FIRST line in my post is:

In this video, 2 separate features are being shown

If somebody's interested enough to be responding at all, they can wade through a couple of sentences and get an idea what they're responding to.

_Mike

Weird reference but in Star Trek NG Data plays a guy in a game called Stratagamea. Data lost even though he played perfectly.


Obviously you wrote correctly. In light of it leading to a misunderstanding maybe another correct way of saying it would be helpful ;)

[azazel]

what kind of codec that video uses ? i can't play it....

[Opelfruits]

what kind of codec that video uses ? i can't play it....

http://www.techsmith.com/download/codecs.asp

[azazel]

thanks :) it works now :)

[Per-Anders]

It wouldn't be the same, when you change a lights intensity it is not a linear falloff. You would get totally different results between just changing the opacity of one lights influence in a layer (no matter what blending mode you use) and actually changing the intensity of the light and thus completely changing the GI calculations (be they biased or not).

correct based on an inverse square or cubic falloff, however if you check the video it doesn't look like this is the case here, it has the same visual qualities of add (or linear dodge in photoshop). so i'm assuming that the end pixel result is simply ((light1samples/totallight1samples)*brightmullight1)+((light2samples/totalliht2samples)*brightmullight2)+...
until shown otherwise.

[pluMmet]

As stated, the video shows two different features:

1. The cooperative render function combines multiple render results (MXI files) of the same frame in one single MXI file, producing a new result with higher sampling level and less noise. This process does not require additional calculations and helps to reduce dramatically the render time, using several computers to render in parallel.

2. The multiple light edit feature allows adjusting the intensity of individual lights at any time. This feature does not require several passes or additional render time. Multiple light channels are available within a single MXI file for editing. This process can be done in realtime and each light can be modified independently. This is not a color mixing process, the intensity variations are energy based and are calculated in high dynamical range. Power emission can be scaled naturally.

[fasteez]

its not <r,g,b,a> composition but the concept here is the same no but with a more large real value range ? or is it "finite element" grids of lightwaves merged into one space ?

[yinako]

uhm.. isn't this like IPR but maybe more options.

And he didn't really rotate the camera, so its not really realtime renderer like other realtime renderers...huh.?

[CupOWonton]

Oh dear I can't believe how badly this announcement turned out...

I want to point out in case people don't know, that Mike is a lead (probably the leadest?) A-Team tester on Maxwell. He's not interpreting the video at all, he works directly with the latest features and the developers. I'm an A-Team tester too but I'm not quite so super high up yet as to use that specific new tool hehe.

It really isn't a basic compositing system at all.

Example: I create a 3D house. I put 50 lights in it, and have a sun and sky. All the lights are turned off except for the sun and sky. I render the image 1 time and get a sunlit house. Now, from here until eternity I can turn on/off/adjust any of those 50 lights (and sun/sky) creating entirely new lighting with unbiased GI and caustics and reflects etc etc in realtime.

My boss says: "You know we could use a nightshot with light through the bedroom window instead."

I say: "Sure" And in realtime, turn off the sun, and switch on the bedroom light and adjust it's brightness to get the right effect. All from the 1 original image I rendered. New caustics, new GI, new reflections of different parts of the house etc.

Can you change the light, camera, or objects angle or position in realtime?
Because this still sounds like postwork composited layers.

[Opelfruits]

cupOwonton - this isnt a mirracle cure for realtime 3d, light intensity is all you can change in realtime, but this will save hours of re rendering and testing.

[CupOWonton]

cupOwonton - this isnt a mirracle cure for realtime 3d, light intensity is all you can change in realtime, but this will save hours of re rendering and testing.

For sertain things, sure. But theyre calling it "realtime" referring to 3d rendering, when its realy "Post Work."

[Opelfruits]

thats because you can alter the lighting in "realtime", its not post work at all, you are changing how the lights interect with the whole scene, i.e reflections, refractions, illumination, caustics. that is far from what you would do in post with another render engine

[ThomasMahler]

Can you change the light, camera, or objects angle or position in realtime?
Because this still sounds like postwork composited layers.

You gotta be kidding, if this stuff works as Jozvex described it, this would be pretty kickass. Think about how much time you'd save if you wouldn't have to render over and over again, just to change some light attributes.

[CupOWonton]

thats because you can alter the lighting in "realtime", its not post work at all, you are changing how the lights interect with the whole scene, i.e reflections, refractions, illumination, caustics. that is far from what you would do in post with another render engine
All youre doing though is changing intensity, something you can do with pretty much any other compositing program out there. You say its 'changing how it interacts with the whole scene' thats not true, youre just reducing or increasing values on an HDR scale for seaperate layers, thats just how it interacts with the image. They can call this 'realtime lighting' when it actualy involves turning or repositioning lights or even changing their shape.

[CupOWonton]

You gotta be kidding, if this stuff works as Jozvex described it, this would be pretty kickass. Think about how much time you'd save if you wouldn't have to render over and over again, just to change some light attributes.

After working with a GI setup for a wile, you pretty much know what settings to use in most scenerios.

With this process and maxwell - if I decide I need to change the angle of the light, and I havent already filled the entire room with 3000 lights so I can make sure Ive covered 1/1000th of my available options, I'd still end up needing to render a pass. If I have to change the size of the light, I'd need renders for that too. Changing the intensity is nice, true, I would like it if Max/Mray/Vray/Brazil split layers for each light, but that would obviously make for much larger saved image files, but its still not 'realtime lighting'.

[Opelfruits]

theres probably no point in trying to explain anymore, you obviously cant see past how it looks like render layers, and if you want to use render layers in whatever program you use then by all means please do this. But this has nothing to do with layers.

[playmesumch00ns]

I too am having trouble seeing how changing the brightness of the lights (which is all that is shown in the example video) is any different from rendering each light individually then adding all those renders in shake and playing with the values of each.

The end result would be exactly the same

[Opelfruits]

changing the intensity of lights in maxwell would also alter the intensity of the indirect illumination, this is impossible with your standard render layers

[P_T]

changing the intensity of lights in maxwell would also alter the intensity of the indirect illumination, this is impossible with your standard render layers

Not if it's HDR right?

[Opelfruits]

Not if it's HDR right?

sorry i dont understand

[playmesumch00ns]

changing the intensity of lights in maxwell would also alter the intensity of the indirect illumination, this is impossible with your standard render layers

Not if it's LDR either.

The intensity of indirect illumination depends only on the intensity of the light source (as long as the geometry of the scene stays constant).

[Opelfruits]

ok .

[mverta]

We'll be generating more examples pretty soon, which should help illustrate how this is entirely different from simple post-process tricks.

_Mike

[rocarpen]

We'll be generating more examples pretty soon, which should help illustrate how this is entirely different from simple post-process tricks.

_Mike


I can't believe people are having so much trouble wrapping their heads around it. People, changing the opacity on your specular pass is not the same thing as adjusting a light and having the indirect illumination, caustics, and reflections ALL react dynamically.

[playmesumch00ns]

But it IS the same thing if all you are changing is the light's colour or intensity, that's what we're trying to say.

[mverta]

No it isn't. Changing the light intensity changes the entire GI for the scene, caustics, etc. This does not happen in a simple layer/opacity scenario.

_Mike

[playmesumch00ns]

So you're saying if I have a light in a box, then double the light's intensity, the overall brightness of the image will be something other than double that of the original?

Or if I double the intensity of a light shining on a wine glass, do the the caustics under the glass not get twice as bright?

[MKeep]

You can easily test this out by copying frames from the quicktime video and pasting them into layers in photoshop. If you grab the frames where each light is isolated and the slider is at its highest intensity you can sort of play around with a low-res version of the scene and draw your own conclusions.

[CupOWonton]

So you're saying if I have a light in a box, then double the light's intensity, the overall brightness of the image will be something other than double that of the original?

Technicaly, since theres light bounces, yes. HOWEVER, the only real difference between this and a regular compositing setup put in the correct order, is that this works on HDR. So the light adjustment is a bit more acurate than a regular layer adjustment even if that layer was a composite of the light, its reflection, and the GI created by it.

Then again, Im not too familiar with say, Combustion, or Shake, so Im not sure if they are unable to work in HDR.

[mverta]

It's not a question of "brightness" it's the nature of the interaction we're talking about. Like I said, we'll be doing some additional examples to illustrate.

_Mike

[mverta]

you can sort of play around with a low-res version of the scene and draw your own conclusions.

One of which will be that you don't need to render passes to do it.

_Mike

[playmesumch00ns]

Technicaly, since theres light bounces, yes.

Will there be more or less? Where does the energy come from/go to?

[CupOWonton]

Will there be more or less? Where does the energy come from/go to?

Depending on the scene it varrys.

If say, you have a white plane light/emmitter standing vertical on a grey plane and nothing around, the light will bounce once, and dissapear. If the light bounces back in say, inside a perfectly white box, everything will be illuminated because perfect white would bounce infinately.
If it was a grey box, say perfect mid gray, the light would bounce using the standard falloff but the grey would be lighter than the grey of the plane in open space, because the light never bounced back onto the plane in open space, they hit once, and go off into infinity. Because of the intensity and falloff values, doubling the light output may more than double the resulting illumination if the light continues to bounce around off of light valued surfaces.

So, it realy all depends.

[fasteez]

its not <r,g,b,a> composition but the concept here is the same no but with a more large real value range ? or is it "finite element" grids of lightwaves merged into one space ?

sorry to repeat mike , but my question was serious .. maybe you missed it because of the "flooding"

[stew]

Because of the intensity and falloff values, doubling the light output may more than double the resulting illumination if the light continues to bounce around off of light valued surfaces.
I don't understand.
Say, I have a wall that reflects 80% of the incoming light. How are 80% of double the intensity not the same as 2x 80% of single intensity?

[CupOWonton]

Well the light coming in would be the first bounce hitting that wall. But,since this is inside, the wall bounces light onto other surfaces, which bounce light back onto the wall too. So depending on how much they reflect as well, the faloff of the light may be a bit different because its hitting different materials at a higher intensity.


Its hard for me to think, im kinda tired. Ill set up a test scene using a mid-grey cube and an area light attached to one open end set to 1.0 . Render once using direct only, render another using GI at 1.0, and then at 2.0, read the pixels at the edge of the light. I think MentalRay reads using floating point, which would show me a more acurate representation unless I can get HDR intenisty out of the color picker... Assuming Im right, the light+bounce area next to the light should have more than doubled, just by a little bit.

[sacslacker]

All I have to say is while these features are nice and all, Maxwell has been an utter mess up until now. Next Limit has treated its customers like hell and I would seriously think twice about purchasing this application. Just a word of warning that I wish I would have had.

Maxwell looks to be making progress, we'll see. I still say we are at least a year from having a product that is usable. By usable, I mean not having to deal with the serious limitations of working with studio or the plugins. Like actually having an undo. Like having the interface actually work. I could go on.

I know this will bring the wrath of the Maxwell lovers but as a very frustrated user, I feel it necessary to warn anyone who purchases this application. You're doing so knowing that its still in development and a production release seems very far away. This is not something that I knew when I bought the application many months ago when we were promised a release date. Since then the application has been rewritten with several cores now and I swear, it's absolutely irritating. If you buy it, know what your getting up front and dont come to those forums and complain. Theres a ton of that already, trust me.

[gigatron]

gah what codec is it encoded with?! It's not playing for me :S

[olbo]

gah what codec is it encoded with?! It's not playing for me :S

... klick (http://www.techsmith.com/download/codecs.asp) ;o)


@sacslacker

I have to agree with you ... but this feature is non the less mindblowing, ... at least for me.

:D :D ;)



take care
Oleg

[mverta]

I don't think this sort of statement should bring the wrath of "Maxwell lovers," though obviously not everyone feels the way you do, if there ARE Maxwell lovers :)

Your frustrations are shared by tons of people, myself included; people just have different ways of dealing with the frustrations. Personally, I think Maxwell is setting up to be a great underdog story, but it might end up otherwise.

One thing I do have a problem with, is when any and all threads related to Maxwell turn into flaming/praising wars. I started this thread to show off what I feel is a powerful feature, and I've appreciated up to now that the discussion has been about the feature, and not another philosophical debate This isn't the place for blind incessant praising of Maxwell, nor vitriolic hate mongering. As far as I can tell, there are a billion other threads for that stuff, already :)

_Mike

[sacslacker]

I'm with you, I dont mean to be hating on Maxwell. To be honest, I've invested in the software and hope its successful. It definitely has promise and these new features are darn sweet. I just wanted to throw in a warning because I didn't know any of the issues with Maxwell until I purchased it. Had I had a few warnings, I might have waited until the product was more mature and saved myself a bit o' grief.

With that said, I think we all appreciate your updates and explainations. You communicate better than NL does and communication has been one of the hugest problems with these guys. I hope you get some sort of compensation because your posts have instilled a bit more confidence that the product is truely progressing.

The new features are very nice and I'm glad to see some innovative stuff coming from NL. I just wish the basics were shipped before the fancy stuff but thats just me. Sorry to derail your thread a bit.

Back to the regularly scheduled program.

[visionmaster2]

great news ! very very good light system.

Thank you Mr mverta, for what you are doing for the maxwel users, here and on the maxwell forum !:beer:

[Per-Anders]

just to defend my argument that it looks like standard layering (before you *all* say i've lost my marbles... if i ever had em that is), here's a very quick vid i made using camstudio (shoudl play in quicktime on any pc, it's the default microsoft video 1 codec) and bodypaint (production bundle version) to layer up a basic 32bit file renderd with final render in add mode (equivalent to linear dodge in photoshop, though photoshop doesn't support layering in 32bit mode yet). to me it looks the same.

http://www.peranders.com/general/mp01.zip

what i should like to see, is a simple side by side images, not video example showing the difference between this, i.e. hdri layering in add mode, and m~r's "relighting" on a single image just to see the actual realworld difference between what's occuring with the energy distribution method and basic layering (maybe it's as simple as the relighting gives a different effect over the multiple bounces, or perhaps it results in a smoother result), i think the issue is that it's maybe not so clear in that video and it's perhaps just not giving it the best demonstration to really make it clear how different this is, if it is that is.

[opus13]

maybe it is just a layering/blending scheme. it sure wouldn't be the first time maxwell tried to show off something entirely unimpressive... then again, they are also famous for showing something new in the worst light possible, and need to redo it several times before it even looks like progress.

either way, its kind of hard to accept anything they say. :shrug:

[Per-Anders]

well i wouldn't go so far as that. the impressive part is really that as a very nice side effect of how it works it appears to be able to render all those lights and their gi effects into their own seperate layers in one gi pass (which is really a great saving, and might cause it to be in this respect as fast or faster than other solutions in certain situations). and the ability to blend the files together in order to get rid of grain is very nice too (though i hope that can be made to be more automated).

it's only that it's advertised in this thread as relighting, which to my mind conjours up images of things like lpics etc rather than just layering, and they've not called it layering, even though that's exactly what it looks like (at least to my eyes).

[CupOWonton]

To multiple reflections/refractions and caustics work correctly in Add mode?

[Per-Anders]

of course. everything that's illuminated is simply added and you can only see what's illuminated after all. we are just dealing with the raw basics of light here.

[CupOWonton]

OK... then... their compositing system realy is no different from anything else Ive seen then if thats the case. All it seems to do is store each light and additional caustics/gi and seaperate them out in the editor for tweaking.

A nice option without the need for multi passes. However, whats the file size of this compared to just a regular image file? And do the changes in the rendered image become reflected in the next render? would you be able to use it to tell MW to exclude lights you no longer need for say, an animation, and render it all out using the new tweaked settings?

[Coliba]

mdme_sadie, thanks for making that example. To understand the differences better, look at when you're changing the intensity of the large fill light. It starts blowing out all the objects in the room, while the back only becomes slightly brighter. If you had changed it's brightness value first, then re-rendered the scene, the whole scene would look different because the whole GI solution would change. Normally, if it is a good GI engine, when you increased the brightness so much of the fill light, the back should have been illuminated much more.

Here you are just changing the intensity of the pixels, so you are first of all not taking into account the inverse square falloff of realistic lights, and second, taking into account the GI algorithms of your renderer. You have no way of telling your layer: this is where the light is in the scene, this is it's shape.What you are showing basically is per light colormapping changes. This is not what is shown in the vid from Maxwell, wouldn't matter if you were editing hdr or ldr layers. It is much more than adding pixel values, it allows you to change the entire GI calculation (distribution of energy levels) during and after rendering.

[Per-Anders]

i disagree, that's really just down to the gi solution itself and the lights falloff (inv square) which naturally blows out closer. if i upped the gi bounces the fill would happen as expected.

after some tests i agree that the look would be different, however i'm still not seeing any actual difference visually between what i did and what was shown in the video. and that's why i'd want to see a side by side. the individual lights output as 32bit layers then blended with add/linear dodge and the internal render "relit" to teh same percentage levels (and then of couse a final render where the original emitters were set to match the percentage levels in the "relit" scene).

[ThomasAn.]

mdme_sadie,

Your method seems to be prety much equivalent. The only difference is that Maxwell does all this in only one pass and without having to go through a seperate layering setup in Photoshop. Also it is doing all this in H-HDRI (which is a higher range than HDR. They call it MXI).

[rocarpen]

mdme_sadie,

Your method seems to be prety much equivalent. The only difference is that Maxwell does all this in only one pass and without having to go through a seperate layering setup in Photoshop. Also it is doing all this in H-HDRI (which is a higher range than HDR. They call it MXI).

Plus real time global illumination, caustics, and reflectives.

[ThomasAn.]

Plus real time global illumination, caustics, and reflectives.

Well, the manual layering method as shown by mdme_sadie already includes GI, caustics, and reflectives.

[rocarpen]

Well, the manual layering method as shown by mdme_sadie already includes GI, caustics, and reflectives.


Like I've said, that's limited to adjusting opacity levels and blend modes in something like a Photoshop PSD or AfterEffects comp. From what I've seen, Maxwell goes beyond that.

[CupOWonton]

Like I've said, that's limited to adjusting opacity levels and blend modes in something like a Photoshop PSD or AfterEffects comp. From what I've seen, Maxwell goes beyond that.

From what we can see it's not calculating any GI or caustics in realtime. Its stored bit data of each light and the results they create (Diffuse/Specular/Reflection/Refraction/GI/Caustics/Dispersion). The ambient light has its own layer, and each individual light had its own. It simply works in an HDR format which allows for more ranged controll over the intensity. Its adjusting blend modes, and seems to cap each one to prevent over doing it. Im sure if you cranked one too high it'd just accentuate the already viewable noise.


I hope their next video to elaborate on this also includes doing this with DOF.

[Coliba]

The ambient light has its own layer,

What do you mean by ambient light? In Maxwell there's no such thing as ambient light.

Anyway, you can't compare it to just an overall brightening of the pixels. If you brighten all the pixels in an image say 25%, you won't get the same results as if you had increased your lights brightness with 25% and rendered again. Not in a GI situation.

Also, what about shadows? If you're rendering package separates the GI into one layer for the light, then another layer for the shadow, when you increase a lights brightness, depending on the situation more indirect light bounced around means certain shadows should not look as dark, or perhaps lighter at the edges etc.....

[stew]

Anyway, you can't compare it to just an overall brightening of the pixels. If you brighten all the pixels in an image say 25%, you won't get the same results as if you had increased your lights brightness with 25% and rendered again. Not in a GI situation.
Sure you would. Where and in which direction indirect light (be it diffuse or caustic) bounces, depends on the objects and light positions, not the intensity of the light. The only thing that changes with the intensity of the light is the intensity of the reflected light, and they correlate linearly. 25% more direct light means 25% more indirect light - if it weren't, we'd have to rewrite the laws of thermodynamics (where would that energy difference come from?).

[playmesumch00ns]

Exactly. How hard is that to understand? If you double the intensity of the light, the intensity of everything else gets twice as bright as well, because it couldn't work any other way.

You would only have to recalculate the GI if you moved the light, or changed some other property of the scene such that the light direction changed.

Mike has hinted that there is something else going on, and I'm willing to believe him since it seems that the mxi files store a lot of sample information as well, but the simple fact is that the example shown so far contains NOTHING that cannot be reproduced EXACTLY by rendering each light to an individual layer then toying with those layers in a floating-point compositing program.

[stew]

Yes, I can imagine quite a few interesting things that one could pull off in an MLT environment: Keeping a cache of "known good paths" could allow you to rerender only the affected rays when a surface shader changes, in many cases reusing the rays already traced (not sure how much memory that would take).

[Per-Anders]

also i'm not sure where this talk about shadows comes into it. m~r is a gi only engine, there are no shadows as such, only (just as with real life) the absence of light. caustics are futhermore just light effects (reflected and refracted light), reflections just show light reflected direct to eye. it's all the same. not even seperate layers in this scenario (though potentially it could be split out into caustics, reflections, refractions and diffuse through ray filtering, but certianly not shadows). there aren't even lights in m~r afaik, just emitters (same or similar to making a surface luminant in any other gi engine, though here they cast light rays, but there is no what you'd be used to in 3d as lights).

[ThomasAn.]

Plus real time global illumination, caustics, and reflectives.

Well, here is a test.
Two layers (two passes) in Vray.
Light adjusted independently in Paintshop.
Caustics and reflections are affected (as expected).

Sorry for the huge (uncompressed) file I didn't find the right codec at the moment.
(1.7Mb compressed)
http://rapidshare.de/files/13388295/Lights_IndependentControl_.7z.html


(For 7z: www.7-zip.org)

[denis3]

All I have to say about Maxwell is Buyer Beware.

If you decide to buy into Maxwell then remember the following:

1/ You are investing into Alpha software. They still haven't done a feature freeze (as this topic shows). There is no launch date promised - they originally said they would be releasing in October 2005. (and I think this was already a re-scheduled release date.)

2/ Some features already promised are still not working reliably. - currently the track record for solving these problems is worryingly poor. It seems NL would prefer trying out new ideas and features rather than iron out existing bugs.

3/ There are no guarantees that this recently publicised relighting feature will work for production work. i.e. if you look closely at the movie demo you will see a fragment of one of the light panels (top left) seems to appear within the other light 'passes'. Looks like an error to me. But then remember this is alpha software.

4/ There are no guarantees that Maxwell will work with your software pipeline and your specific project needs.

5/ There are many beautiful maxwell renders posted but mostly came from the original 1.22a version which they have canned in favour of another render core (so called rc1.5 ?) which does not match the look of the so-called 1.22a beta. This is very regretable.

6/ You will fall in love with it.


Essentially you are paying to help finance their R&D and enjoy the chaotic ride at the crest of the rendering wave.

[Opelfruits]

All I have to say about Maxwell is Buyer Beware.

If you decide to buy into Maxwell then remember the following:

1/ You are investing into Alpha software. They still haven't done a feature freeze (as this topic shows). There is no launch date promised - they originally said they would be releasing in October 2005. (and I think this was already a re-scheduled release date.)

2/ Some features already promised are still not working reliably. - currently the track record for solving these problems is worryingly poor. It seems NL would prefer trying out new ideas and features rather than iron out existing bugs.

3/ There are no guarantees that this recently publicised relighting feature will work for production work. i.e. if you look closely at the movie demo you will see a fragment of one of the light panels (top left) seems to appear within the other light 'passes'. Looks like an error to me. But then remember this is alpha software.

4/ There are no guarantees that Maxwell will work with your software pipeline and your specific project needs.

5/ There are many beautiful maxwell renders posted but mostly came from the original 1.22a version which they have canned in favour of another render core (so called rc1.5 ?) which does not match the look of the so-called 1.22a beta. This is very regretable.

6/ You will fall in love with it.


Essentially you are paying to help finance their R&D and enjoy the chaotic ride at the crest of the rendering wave.

its not quite as bad as you make out, RC5 is very useable and faster than the beta - see front page of cgtalk - keytoon desk

6 - yes this is true, it has a magical power over you that is not understandable to any human being

[UrbanFuturistic]

3/ There are no guarantees that this recently publicised relighting feature will work for production work. i.e. if you look closely at the movie demo you will see a fragment of one of the light panels (top left) seems to appear within the other light 'passes'. Looks like an error to me. But then remember this is alpha software.Actually, that's a reflection of the top right light, the wall between the top left light and the glass sphere is a mirror and there's actually only 3 lights visible in the scene.

[Caffeinemonkey]

hack!

What a hack!! this blows!


Patrick Thomas

[ThomasAn.]

2/ Some features already promised are still not working reliably. - currently the track record for solving these problems is worryingly poor. It seems NL would prefer trying out new ideas and features rather than iron out existing bugs.

The only thing really missing is the visibility of caustics as seen from *behind glass* panels and this also affects the clipmap behavior (it is working only after enormous amounts of time). This is not really Maxwell's fault, because it is a core limitation of the MLT algorithms. Basically the PhD's who researched the original spectral equations and MLT theory could not handle it satisfactorily ... and here we are demanding the software company to reinvent this research with a year. (They might get it to work though, because they have very bright engineers)

Maxwell is an awesome product and most importantly a great implementation. The user friendliness that it introduces is unparalleled and has not been seen before in a single rendering package. Maxwell is the way of the future. The main issue is that the management lacks experience (but its engineering aspect is quite advanced and viable).

[CupOWonton]

Maxwell is an awesome product and most importantly a great implementation. The user friendliness that it introduces is unparalleled and has not been seen before in a single rendering package. Maxwell is the way of the future. The main issue is that the management lacks experience (but its engineering aspect is quite advanced and viable).

In this case " unparalleled user friendlyness" means:

"being stuck with Studio for setting up lighting and materials rather than inside the modeling program"
" no availability of MatID layers,normal layers,object ID layers, or any standard compositing layers for that matter"
"..time.."

Anyone want to add to that?

Lets stick with facts, not opinions.
Now, what we're trying to figure out here is exactly how this is any different than composited Added layers of different lights on a diffuse background.

I think the main differences anyone could agree on so far is its HDR so you get a higher contrast ratio, and that its neat it saves each light + diffuse + caustics + GI on individual layers without the need for a multipass (I totaly agree that's a cool idea).

[ThomasAn.]

In this case " unparalleled user friendlyness" means:

"being stuck with Studio for setting up lighting and materials rather than inside the modeling program"
" no availability of MatID layers,normal layers,object ID layers, or any standard compositing layers for that matter"
"..time.."

Anyone want to add to that?

Lets stick with facts, not opinions.

What you just said is not really a fact, it is merely a rant of your own flavor (and childish at that). The alegation, or innuendo, that the use of studio is intended to depracate plugins, is entirely faulty and not supported by official statements. The plugins *WILL* be brought up in sync. The necessary modules needed to be blueprinted first at this stage of development ;before the plugins can tap into the new material features. Studio is an important component, but it is intended to be optional. (writing plugins for 18 different architectures doesn't happen overnight and the studio is a common programmatic denominator).

[stew]

I think the main differences anyone could agree on so far is its HDR so you get a higher contrast ratio, and that its neat it saves each light + diffuse + caustics + GI on individual layers without the need for a multipass (I totaly agree that's a cool idea).
Hm, that's nothing new. The RenderMan standard has arbitrary output variables for a long time now (which means that you can save anything into separate layers - light, normals, objects IDs, etc), which too can be saved at full floating point precision.

Oh, and I just checked - my ancient copy of C4D R7 also writes separate layers for almost any component ("Radiosity", specular, diffuse, caustics, etc) from one render pass into 16bit PSD files.

Don't get me wrong - I think Maxwell will be a valuable member of the renderer market, but it's evolutionary, not revolutionary. As far as I know, they are one of the first to deliver a production ready MLT implementation.

[tweeeker]

Phew... What a confused thread. All I can recommend is that some of you guys study some basic light transport before redefining the laws of physics in threads like this. As playmesomechOOns has already stated a whole bunch of times light is ADDITIVE and LINEAR.

Think about it this way - if your in a room with a light controled by a dimmer switch, as you gradually increase the brightness, the whole room gets brighter in a LINEAR fashion. Its not, for example, like only after turning the dial halfway the 'gi' starts to appear. Thats what would happen if it were non linear like some of you are suggesting, which would be bizarre.

As a result, theres nothing in that demo that can't be done in a floating point compositor such as shake or digital fusion.. in fact many people already do so on a daily basis.

I'm not knocking the feature at all, I think its pretty neat to be able to do all that stuff at the click of a button... but this thread should be a few posts long, not 8 pages.

T

[denis3]

Opelfruits - From what I've seen RC5 has a great deal of trouble dealing with emitter lighting (i.e no sunlight system - just emitters.) Dark/shadow areas seem to take forever (i.e days) to clear up. So, I wouldn't agree with you about RC5 being faster.

Opelfruits - Keytoon - agreed, their stuff is brilliantly animated and rendered.

odubtaig - thank you for pointing this out - I hadn't realised that this is what was happening in the sample movie.

ThomasAn - Personally I don't find Studio has unparallelled user friendliness. I cannot agree with this.

Where for instance is the undo? I find it hard to assign multiple materials to complex polygon selections within Studio especially when these polygon selections don't necessarily follow the original object topology.

ThomasAn - agreed MLT is the way of the future in terms of photo realistic rendering. Maxwell looks to be the leader at present in this field.



My worry is that NL might actually be engaged in a patent land grab, essentially using the headstart they currently hold to establish a number of patents within the field of MLT rendering. All great except of course they might leave users behind with a product that has limited functionality as a mainstream render tool.



I wonder how long it will take the rendering heavy hitters to come out with their implementations of MLT?

[Strang]

...I wonder how long it will take the rendering heavy hitters to come out with their implementations of MLT?...

mentalray has had the "path_material" for a while. it does path tracing on a per material basis instead of always on (i think)

vray has ppt (progressive path tracing)

[CupOWonton]

ThomasAn - Personally I don't find Studio has unparallelled user friendliness. I cannot agree with this.
Where for instance is the undo? I find it hard to assign multiple materials to complex polygon selections within Studio especially when these polygon selections don't necessarily follow the original object topology.Hahaha, no undo? Oh god, thats hillarious. One of the most basic functions nessisary for art programs or any program for that matter, and they didnt think to include it?

I wonder how long it will take the rendering heavy hitters to come out with their implementations of MLT?When it becomes practical for them and their users.

mentalray has had the "path_material" for a while. it does path tracing on a per material basis instead of always on (i think)
vray has ppt (progressive path tracing)

Neither of those are MLT however.
Metropolis Light Transport specificly calculates electromagnetic waves of light.(I think it would be cool to see if they could make GlowInTheDark materials that actualy react to UV light)
PPT uses a similar method by tracing pixels over and over, but its still not the same.

[andrewsweet]

Just like to add :-)

Hi Mike :-)

Well yes this feature looks great!!! and there's NO knocking it.

But it's sad really! Maxwell is without doubt a stunning render software BUT i'm afraid after using it for a year..... they have some basic issues to sort before showing these extra features

Lets get the plug-ins working first, then studio then the new features :-)

Not knocking !!! just afraid the whole thing will die away!

Andy

[Strang]

Neither of those are MLT however.
Metropolis Light Transport specificly calculates electromagnetic waves of light.(I think it would be cool to see if they could make GlowInTheDark materials that actualy react to UV light)
PPT uses a similar method by tracing pixels over and over, but its still not the same.

http://graphics.stanford.edu/papers/metro/

"... we generate a sequence of light transport paths by randomly mutating a single current path (e.g. adding a new vertex to the path). Each mutation is accepted or rejected with a carefully chosen probability, to ensure that paths are sampled according to the contribution they make to the ideal image. We then estimate this image by sampling many paths, and recording their locations on the image plane..."

so in laymen's MLT is randomizing paths of a ray many many times to produce the correct effect

although i very well could be wrong. i believe modifying the path by some amount is something they can already do. ppt is vray's implementation and path_material and dgs_material is mentalray's implementation.

here is a page on ppt...

http://www.spot3d.com/vray/help/VRayHelp150beta/tutorials_pathtracing.htm

here is quote from mr3.4 docs...

"...Global illumination effects can be rendered with path raytracing. On contrary to the distributed raytracing, for each ray hit, at most one ray is reflected or refracted when path raytracing is used. With sufficiently large trace depth limits, most of the indirect illumination effects are computed correctly. However, the results are very noisy (sound familiar?) and an extremely high number of samples is required for obtaining high quality images..."

i dont know if mentalray or vray actually have the exact MLT implentation that is outlined in the MLT paper but they sound pretty similar.

anyways i know how these debates turn out... so i am going to shut my hole :O

[ThomasAn.]

MLT and PPT are not the same.
Vray does have PPT, but it is not as optimized as Maxwell. As a matter of fact in my tests Vray PPT seems to resolve anywhere between 2x to 3x slower than Maxwell RC5.
You do not get Vray for its PPT feature... you get it for its GI speed and very good refractive caustics (particularly, Irradiance maps and LightCache).


Also, Maxwell is not pure MLT either... it is modified and customized (as they are trying to overcome limitations of the theory)... the same way that Ubundu is based on Debian but it is not Debian.

[ThomasAn.]

ThomasAn - Personally I don't find Studio has unparallelled user friendliness. I cannot agree with this.

Where for instance is the undo? I find it hard to assign multiple materials to complex polygon selections within Studio especially when these polygon selections don't necessarily follow the original object topology.

Well, there was no mention of Studio specifically in the earlier post. The comment was a blanket statement for the overall Maxwell rendering paradigm... the ability to assign ready made, preset, physically accurate materials ... and then hit render.
(preset materials used to be there during beta and should become available again in a more expanded way; in the form of complex IOR libraries).

Also, user friendliness is meant in the sense of least amount of parameter fiddling. You do not have to worry about reflective/refractive caustic settings, gloss subdivisions, Maxdepths, transparency cutoffs, Transparency levels, Ray Biases, Max tree depths, leaf sizes, Face Coeficients, DOF subdivisions, Aliasing Min/Max retes, Aliasing Filters, Aliasing area/size params, QMC samplers, adaptive amounts, noise thresholds, subdivision multipliers, Color thresholds, distance thresholds, normal thresholds, Interpolation types, calculation samples..... and so on and so forth .... tweak/render and balancing numerous abstruct, unitless variables .....

Maxwell, in its vision, has the least amount of setup time (the least amount of man hours). However, since MLT is computer intensive you get much more PC hours... but... even at that... Maxwell is the most optimized MLT renderer out there. There is no other spectral product (that I know) which can resolve noise as fast.

I think in general they are sitting on a great idea and they have solid technological foundation, but as mentioned, the management has been rather inexperienced and some mistakes have been made in the overall handling.

[Jozvex]

Opelfruits - From what I've seen RC5 has a great deal of trouble dealing with emitter lighting (i.e no sunlight system - just emitters.)

There are no problems with not being able to use sunlight in RC5. Sunlight, emitters, physical sky and skydome all work. It has also been proven numerous times that RC5 does have "the look" that 1.2.2a had and in fact it has even more of it because the light propagation has been really improved since beta.

And as they've stated, many things such as the sunlight-through-glass-when-seen-through-the-glass issue and others have already been solved for the RS2 engine. They just need to implement it into the currently released engine.

From an A-Team advantaged point of view, my outlook for Maxwell's future is :beer: :thumbsup: :buttrock:

[UrbanFuturistic]

I don't know what they're doing, but it ain't pure additive blending. I've done some simple tests.

This was done with YaFRay, which isn't going to be as accurate as Maxwell :p, but still does GI and caustics and it still outputs to Radiance HDR.

Compositing was done in Cinepaint which took a long time as it has a habit of falling over if you do anything esoteric like open a file or try to save. This was also why I had to get screen shots of the final results with GIMP and which is why image dimensions may vary slightly.

http://img137.imageshack.us/img137/4359/fullstrength6dw.th.png (http://img137.imageshack.us/my.php?image=fullstrength6dw.png)Lights at MAX
http://img117.imageshack.us/img117/5808/darker0mb.th.png (http://img117.imageshack.us/my.php?image=darker0mb.png)Single Pass, lower light levels: 0.3 and 0.7 energy levels
http://img114.imageshack.us/img114/9593/compositeopacityaltered9rm.th.png (http://img114.imageshack.us/my.php?image=compositeopacityaltered9rm.png)Composite of full strength passes, additive blends at 30% and 70% (caustics, especially in mirror)
http://img112.imageshack.us/img112/4111/compositelightaltered3xz.th.png (http://img112.imageshack.us/my.php?image=compositelightaltered3xz.png)Composite of 0.3 and 0.7 passes, additive blends at 100% (Shadow in front of mirror is darker)
http://img102.imageshack.us/img102/5642/hdrshopblend6ov.th.png (http://img102.imageshack.us/my.php?image=hdrshopblend6ov.png)Composite of HDR Shop scaled passes, additive blends at 100% (caustics again, just less so)

One thing people seem to have forgotten is that light might scale linearly, but an additive blend in postprocess doesn't. Take three images, do additive blends, change the order and the result is different. Not something you get when everything's done 'in camera'.

If anyone thinks they can do better (which wouldn't surprise me given my mad skillz :p) I'll be happy to email them the HDR files so long as their inbox can take the strain :)

[gjpetch]

One thing people seem to have forgotten is that light might scale linearly, but an additive blend in postprocess doesn't. Take three images, do additive blends, change the order and the result is different.

Eh? The order of an additive (screen) blend shouldn't matter. Think about the colour values as numbers, additive means to add, and 1+2+3 is the same as 3+2+1, or 2+1+3, or whatever.

[tweeeker]

I'd try and find out the math behind your compositors 'additive blend'. If the order makes a difference, then it's not really additive, which will explain why your getting incorrect results.

Just to clarify, judging by the shadows it looks as though you have 2 lights in that scene. Have you done a floating point render of each light individually? - cause thats what you'll need if you want to change the intensity ratios of the two lights in a compositor. Each of the light passes should contain the direct illumination, gi, reflections, caustics etc just for the that light.

T

[andrewsweet]

http://graphics.stanford.edu/papers/metro/ "...Global illumination effects can be rendered with path raytracing. On contrary to the distributed raytracing, for each ray hit, at most one ray is reflected or refracted when path raytracing is used. With sufficiently large trace depth limits, most of the indirect illumination effects are computed correctly. However, the results are very noisy (sound familiar?) and an extremely high number of samples is required for obtaining high quality images..."

i dont know if mentalray or vray actually have the exact MLT implentation that is outlined in the MLT paper but they sound pretty similar.

anyways i know how these debates turn out... so i am going to shut my hole :O

Path tracing in MentalRay! not bad eh!!

http://img123.imageshack.us/img123/2665/pathtestv2color6nv5un.jpg (http://imageshack.us)

[KRZ9000]

andrewsweet why dont you start to make clear what you want to achieve with your mentalray renders. its ridiculous and more then enought that you post them on the maxwell-forum. sofar noone thought they are maxwell-renders therefore i think you can stop now making an idiot out of yourself

[andrewsweet]

andrewsweet why dont you start to make clear what you want to achieve with your mentalray renders. its ridiculous and more then enought that you post them on the maxwell-forum. sofar noone thought they are maxwell-renders therefore i think you can stop now making an idiot out of yourself

Hmmm!

As I own both I'll do what I like thanks!! it's interesting how defensive people get over software! is this really a crime! and on top of this think Maxwell is great software bar the points listed above.

And for your information I receive stacks of emails asking for tips on both MR and MW and I'm always very happy to help people! but sadly there's the odd nut case! LOL

I have to thank Maxwell to be honest as is making me learn more about how powerful MentalRay is!

As for being an Idiot!! man that hurts! I've been doing this 10 years! and judging by your website gallery your in no position to start calling people idiots - which is not nice is it :-)


95% of the work on the Maxwell site is poor at best! so it's not hard to better this! really is it!

And you also said nobody on the Maxwell forums thought those MR renders were not Maxwell!? trace back over a few pages and your see some of there main guys making comments on how good they looked :thumbsup: ... you were saying!

Test for you then! what engine rendered this? and does it matter? NO

http://img131.imageshack.us/img131/9421/and1zw.jpg (http://imageshack.us)

Chill

Andy

[KRZ9000]

its the way you do it...you obviously tried to trick people with your pics and failed. your motivation why you do that is your bussiness. the fact that you continue to do so makes me think that you actually want to tell the forum something that is not yet clear. is it that you want to prove that mentalray can produce maxwell style quality? why not say what you think. behaviour like this doesnt make you very sympatic even though you tried to balance the atmospehere you cause lately.

[andrewsweet]

its the way you do it...you obviously tried to trick people with your pics and failed. your motivation why you do that is your bussiness. the fact that you continue to do so makes me think that you actually want to tell the forum something that is not yet clear. is it that you want to prove that mentalray can produce maxwell style quality? why not say what you think. behaviour like this doesnt make you very sympatic even though you tried to balance the atmospehere you cause lately.

Well maybe I work to many hours! I'll be more honest in future posts. I'll start I proper thread with detailed information on both setups! then maybe someone will benefit from it.

Sorry if this has caused any issues - it's not meant to!

Cheers

Andy

P.S> So what engine rendered that image? care to guess :p

[stew]

"... we generate a sequence of light transport paths by randomly mutating a single current path (e.g. adding a new vertex to the path). Each mutation is accepted or rejected with a carefully chosen probability, to ensure that paths are sampled according to the contribution they make to the ideal image. We then estimate this image by sampling many paths, and recording their locations on the image plane..."

so in laymen's MLT is randomizing paths of a ray many many times to produce the correct effect
MLT is much more complex than path tracing. Path tracing is "stupid" in that it traces (quasi)random paths. MLT, however is putting a lot of work into judging a path's contribution to the final image and mutating existing good paths to find similar paths.

In laymen's terms, imagine the two rooms with a small window between them. There's a candle in one of the rooms, you render the other. A path tracer will trace random paths, where only a tiny fraction of them actually goes through the window and hits the candle. A MLT renderer will start off with random paths, but once one of them hits the candle, it will start trying to find other paths that hit the candle too by modifying the one good path it already has.

A path tracer is almost trivial to implement, and the performance usually is embarrassing. Writing a MLT renderer is much more complex, but can generate much better images in less time.

The Stanford site you linked has comparison images of MLT and path tracing, same scene, same render time:
path trace (http://graphics.stanford.edu/papers/metro/fig5a.jpg)
MLT (http://graphics.stanford.edu/papers/metro/fig5b.jpg)

[KRZ9000]

hehe im bad in guessing.....since its that noisefree;) i would tend for mentalray since the dof looks unmaxwell to me but dont nail me down on this. its good to hear that you bring in your experience with mentalray in further more open discussions. in the end we all can only benefit from it.

[andrewsweet]

MLT is much more complex than path tracing. Path tracing is "stupid" in that it traces (quasi)random paths. MLT, however is putting a lot of work into judging a path's contribution to the final image and mutating existing good paths to find similar paths.

In laymen's terms, imagine the two rooms with a small window between them. There's a candle in one of the rooms, you render the other. A path tracer will trace random paths, where only a tiny fraction of them actually goes through the window and hits the candle. A MLT renderer will start off with random paths, but once one of them hits the candle, it will start trying to find other paths that hit the candle too by modifying the one good path it already has.

A path tracer is almost trivial to implement, and the performance usually is embarrassing. Writing a MLT renderer is much more complex, but can generate much better images in less time.

The Stanford site you linked has comparison images of MLT and path tracing, same scene, same render time:
path trace (http://graphics.stanford.edu/papers/metro/fig5a.jpg)
MLT (http://graphics.stanford.edu/papers/metro/fig5b.jpg)

All very interesting stuff!!

Are there any engines out there *bar MW assuming it does* that use the MLT method? I'd like to play.

Andy

[P_T]

hehe im bad in guessing.....since its that noisefree;) i would tend for mentalray since the dof looks unmaxwell to me but dont nail me down on this. its good to hear that you bring in your experience with mentalray in further more open discussions. in the end we all can only benefit from it.

Dude, says on the bottom right corner "MAXWELL LOW-RES TEST". Unless he's lying, I'd say it's a Maxwell render. :D

[stew]

All very interesting stuff!!
Are there any engines out there *bar MW assuming it does* that use the MLT method? I'd like to play.
Metropolight (http://www.3dvirtualight.com/mlt/)

[UrbanFuturistic]

Eh? The order of an additive (screen) blend shouldn't matter. Think about the colour values as numbers, additive means to add, and 1+2+3 is the same as 3+2+1, or 2+1+3, or whatever.Like I said, m@d sk1llz :thumbsup:

Anyway, here's another comparison using screen rather than addition where the layer ordering doesn't matter. It's closer, but still no cigar. There's still this weird little dark patch in the middle of the caustics.

Like I said before, in the likely event that you can do better, PM me your e-mail address and your preferred compression format and I'll send you the original files; As it stands, the .zip is over 1Mb in size and this forum doesn't accept the substantially smaller .tar.bz2 files.

http://img117.imageshack.us/img117/5808/darker0mb.th.png (http://img117.imageshack.us/my.php?image=darker0mb.png)Original Single Pass
http://img106.imageshack.us/img106/5247/compositeopacityscreen8ke.th.png (http://img106.imageshack.us/my.php?image=compositeopacityscreen8ke.png)Cinepaint Screen Composite
http://img156.imageshack.us/img156/6231/compositeopacityscreenps4oc.th.png (http://img156.imageshack.us/my.php?image=compositeopacityscreenps4oc.png)Photoshop 5 LE (don't laugh :p) Screen Composite (8bpp PNGs)

[playmesumch00ns]

I'm having trouble seeing what the difference is there.

Does Yafray calculate its photon distribution based on the power of the light sources?

[UrbanFuturistic]

Look closer :p

http://img110.imageshack.us/img110/4826/closeupsinglepass6tx.th.png (http://img110.imageshack.us/my.php?image=closeupsinglepass6tx.png)Original

http://img152.imageshack.us/img152/3840/closeupcinepaint5le.th.png (http://img152.imageshack.us/my.php?image=closeupcinepaint5le.png)Composite

So far as I can tell, it's a problem because even HDR can only save values within a 32Bit floating point range and the blue value gets stuck at 1 while red and green keep going up. If Maxwell is using 64Bit Doubles, then it can certainly store all the values within the range of HDRI and a great deal more meaning adjusting the blending in Maxwell will give much better results within the clamped range than HDR could ever give. This is particularly useful with elements like caustics where a lot of light is focussed on a songle point and is more likely than anything else to exceed the limits of the Radiance HDR format as can be seen here.

[tweeeker]

odubtaig - you have to use ADD not screen. screen is usually implemented as
1 - (1-a) * (1-b)
what you want is
a + b


also, the double precision will/might be needed to help with rounding errors and the like. Its certainly not required for dynamic range. Check out Paul Debevecs site, theres a paper on there about capturing the dynamic range of the sun, which fits easily into a 4 byte float format.

T

[playmesumch00ns]

I'd put the difference down to the renderer treating your images slightly differently. 32-bit float is more than enough to capture the detail.

[Chris-TC]

Maxwell will give much better results within the clamped range than HDR could ever give. This is particularly useful with elements like caustics where a lot of light is focussed on a songle point and is more likely than anything else to exceed the limits of the Radiance HDR format as can be seen here.

I'm sorry, but I really don't know how you arrived at these results.

Here is an extremely quick composite that shows that you certainly don't need Maxwell to do this kind of lighting adjustment in post.

(Image 1)
Left Light: Intensity 13000
Right Light: Intensity: 15000
http://img153.imageshack.us/img153/4193/adda7gt.jpg

(Image 2) -> rendered as EXR
Left Light: Intensity 3250 ( = 25 %)
Right Light: Intensity 0
http://img153.imageshack.us/img153/2919/addb7ye.jpg

(Image 3) -> rendered as EXR
Left Light: Intensity 0
Right Light: Intensity 3750 (= 25 %)
http://img153.imageshack.us/img153/5675/addc0rz.jpg

(Image 4) -> Image 2 and 3 composited together in post with exposure adjustment
http://img153.imageshack.us/img153/3287/addd6uh.jpg

(Image 5) -> to prove the dynamic range and flexibility you get, the left light overcranked in post
http://img153.imageshack.us/img153/8145/adde1ks.jpg

Please be aware that I kept the render settings low to get this done quickly.

[Coliba]

But you had to render 2 times for this example, so double the render time.

Ok, say you can render everything out in one pass, in layers. Wouldn't you need for every light in the scene to save:

1 layer for the lights illumination,
1 layer for the shadow
1 layer for the caustics
1 layer for the reflections
1 layer for the GI influence of the light (is this even possible? Isn't GI rendered on it's own layer, taking into account all the lights in the scene?)

So then in a GI situation using passes, you can't just increase the brightness of 1 lights layer, and hope that the scenes overall GI solution will get accurately updated as well, can you? I'm just wondering, maybe you can even with GI envolved, but I don't think so...not with a layer approach where everything is separated.

[CupOWonton]

Actualy no,Each layer that belongs to the light would include everything attached to that light as 1 layer.

Light+GI+Caustics+Reflection+Refraction, it would be 1 image of that per light.

The problem with increasing the light far beyond the base layer is that youll end up with some extreme contrast without enough samples everywhere. I can immagine that the noise would be very apparent doing this with a M~R image since youre normaly left with tons of tiny specs everywhere.

[tweeeker]

But you had to render 2 times for this example, so double the render time.


render times were never the issue here.. Besides production renderers such as prman and mental ray have mechanisms where an arbitary number of elements can be generated from a single render. So no, you wouldn't be doubling the rendertime.



So then in a GI situation using passes, you can't just increase the brightness of 1 lights layer, and hope that the scenes overall GI solution will get accurately updated as well, can you? I'm just wondering, maybe you can even with GI envolved, but I don't think so...not with a layer approach where everything is separated.

If you wanted to renderer all the elements you mention individually thats fine. You'd just have to scale all the elements for each light equally. So, if you wanted to halve the brightness of a light, you'd need to halve the direct, indirect, caustics, reflection, etc etc. If you start tweaking these individually then obviously your going to get a different result to the renderer... but of course that would be the whole point of rendering all those elements individually in the first place.

What you can't do, which is perhaps what's causing so much confusion, is have 2 direct lighting passes (i.e one for each light) but only 1 combined GI pass. In this situation there's obviously no way to know what each lights contribution to the GI will have been, so it's not possible to accurately 'scale' the gi.

[UrbanFuturistic]

I'm sorry, but I really don't know how you arrived at these results.I suck at this! I keep trying to tell people that :p I'm just trying to get people to actually demonstrate what can be done with or without Maxwell instead of just arguing about it :D

The basic litmus test, I would have thought, would be to render a single pass with lights at much less than maximum intensity, render seperate light layers at maximum intensity and try to get the composite to match the single pass as closely as possible. As it stands, I'm mostly good at realtime programming so my artistic skills aren't up to much, so if anyone wants to step up and give us a proper demonstration of what's doable *wink* *wink* *nudge* *nudge* *prod* *prod* *poke* *poke* that would be good :podubtaig - you have to use ADD not screen.I was using that, then someone told me it was screen, and I'm confused and don't know what I'm doing :sad: and it's cold, and there are wolves outside...

[tweeeker]

and Chris_TC - thanks for the great example.:)

[tweeeker]

I'm just trying to get people to actually demonstrate what can be done with or without Maxwell instead of just arguing about it :D

The basic litmus test, I would have thought, would be to render a single pass with lights at much less than maximum intensity, render seperate light layers at maximum intensity and try to get the composite to match the single pass as closely as possible.

...err ... isn't that almost exactly what Chris_TC demonstrated a few posts back?

[Chris-TC]

Thanks tweeeker.


The basic litmus test, I would have thought, would be to render a single pass with lights at much less than maximum intensity, render seperate light layers at maximum intensity and try to get the composite to match the single pass as closely as possible.

Well, I have done the opposite. Rendered the single pass at normal intensity and the separate passes with less intensity. But the principle is exactly the same and there's absolutely no need to repeat the test.

[Jozvex]

A new demonstration has been released:

www.jozvex.com/Maxwell/ml.rar

Still only emission intensity at this stage though.

(made by Tom of NL not me)

[KRZ9000]

this should make it abit more clear. and dont be afraid layer-lovers...nobody takes away your prman and im sure you do great things with it nobody else can do.

[stew]

Look closer :p

http://img110.imageshack.us/img110/4826/closeupsinglepass6tx.th.png (http://img110.imageshack.us/my.php?image=closeupsinglepass6tx.png)Original

http://img152.imageshack.us/img152/3840/closeupcinepaint5le.th.png (http://img152.imageshack.us/my.php?image=closeupcinepaint5le.png)Composite

So far as I can tell, it's a problem because even HDR can only save values within a 32Bit floating point range and the blue value gets stuck at 1 while red and green keep going up. If Maxwell is using 64Bit Doubles, then it can certainly store all the values within the range of HDRI and a great deal more meaning adjusting the blending in Maxwell will give much better results within the clamped range than HDR could ever give. This is particularly useful with elements like caustics where a lot of light is focussed on a songle point and is more likely than anything else to exceed the limits of the Radiance HDR format as can be seen here.
Actually, it looks to me more like Yafray (that's what you used, right?) is not using a deterministic method - that is, rendering the same image twice will not give you 100% identical results due to different (quasi) random numbers used, which influences where and how far photons go. I can also see different irradiance caching artifacts outside the caustic. If you render the exact same scene twice in one session, are the resulting images 100% identical?

[stew]

A new demonstration has been released:

www.jozvex.com/Maxwell/ml.rar

Still only emission intensity at this stage though.

(made by Tom of NL not me)
I'd be much more interested in seeing what the f-stop and shutter dials do - if these allow you to change raytraced depth of field and motion blur at full quality, then they have a killer feature that I haven't seen elsewhere.

[UrbanFuturistic]

Actually, it looks to me more like Yafray (that's what you used, right?) is not using a deterministic method - that is, rendering the same image twice will not give you 100% identical results due to different (quasi) random numbers used, which influences where and how far photons go.Nope, not 100% identical but not enough for me to be able to immediately account for the difference seen above.

I'd just quite like someone to get a more pro (and less flakey) renderer and do the same thing or similar, and not go in the opposite direction :p, so I, and others, can see for ourselves if it's something that happens in all renderers and is Maxwell really doing anything special in this respect?

[holyzombiejesus]

Just to let you know, tom just posted another SERIOUSLY cool video over on the Maxwell Forums.

http://www.maxwellrender.com/forum/viewtopic.php?t=12299

Cheers
HZJ

[P_T]

Just to let you know, tom just posted another SERIOUSLY cool video over on the Maxwell Forums.

http://www.maxwellrender.com/forum/viewtopic.php?t=12299

Cheers
HZJ

Looks like the same sort of stuff as the one posted at the beginning of this thread only with a variety of scenes.:shrug:

So far, from the demo and as demonstrated by a few members, I can only see time saving advantage by tweaking light intensity in the renderer instead of a compositing program, not really a "Realtime Interactive Lighting" feature as it's initially hyped up to be since you can't move any of the light's position, change the colour of the lights, adding gobo, etc.

[Chris-TC]

I'd just quite like someone to get a more pro (and less flakey) renderer and do the same thing or similar, and not go in the opposite direction :p,

Uh, what is it with you? :rolleyes:
Why do you think it would not work in the opposite direction? The EXRs I've rendered out have a dynamic range big enough to go in either direction. I could make them so dark that you don't see anything or make them so bright that the lights would look like nuclear explosions.

You said yourself that you're not good at this kind of thing. Well, then let me tell you that it doesn't matter at which intensity you render out. You can easily adjust the lights' intensity in post.
Do you see the caustics boost in the last image? Or the reflectivity boost towards the left edge of the floating object? This kind if thing wouldn't work wihtout using an HDR format. But if you do use one it apparently does. Just accept it alright.

P.S.: As for flakey renderers. The renderer was Mental Ray and it apparently rendered out the correct results in all cases.

[UrbanFuturistic]

Uh, what is it with you? :rolleyes:
Why do you think it would not work in the opposite direction?
That's the thing, I don't know if it would or not which is why I'm asking for a duplicate demonstration :p C'mon, is it really that hard to just crank the lights on full and do the composite? I mean, you're not also having to deal with the 'fun' of cinepaint are you?P.S.: As for flakey renderers. The renderer was Mental Ray and it apparently rendered out the correct results in all cases.I was referring to YaFRay when I said flakey ;) I have problems with it that I just don't get with the Realsoft3D renderer but then outputting to HDR from that is a wholly complex process and I'm not sure how to do it.

I just think, if you can output something the same, and i'm not asking for an exact scene replication, just that you start with higher light levels that produce white caustics in a clamped colourspace and crank down. I know it's something that may seem obvious to you but some of us don't work with MentalRay day in, day out :p

If nothing else, it'll be unarguable proof on one side or the other that'll finish the argument that had been going on for eight pages before I poked my nose in. Is it a problem with the HDR colourspace or is it, and I know this is much more likely, a problem with YaFRay's handling of colour values that just doesn't occur with non-flakey renderers like MentalRay?

That and it'll settle whether the extra colour precision from the MW 64Bit values are a boon or whether other renderers can handle it just fine thankyouverymuch.

[Chris-TC]

That and it'll settle whether the extra colour precision from the MW 64Bit values are a boon or whether other renderers can handle it just fine thankyouverymuch.

I knew it was a good idea to save the file.

So I re-rendered the two single light passes. The left light has 10x its original intensity, the right light has 20x its original intensity.
http://www.digipiction.de/var/add1.jpg
http://www.digipiction.de/var/add2.jpg

Here's the composited shot with correctly adjusted exposure:
http://www.digipiction.de/var/add3.jpg

And just to make a point: here's a live recording (6 frames/sec) of the exposure adjustment (Quicktime, 1.7 MB):
Quicktime Exposure Adjustment in Post (http://www.digipiction.de/var/dynamic.mov)

[UrbanFuturistic]

Thankyou, that's beautiful :thumbsup:

[plastic]

dunno if this has been said before, but the reason because the photoshop composed pics are darker is because photoshop doesnt seem to fully support high color ranges with layers, they are clipped.

[Lorecanth]

don't know if anybody else noticed in that latest vid by tom...but that was real time adjustment of lighting on a sub surface scattering material... thats impressive.

[eek]

can the lights be moved around in realtime with the render? reminds me of max's own interactive renderer.

eek

[CupOWonton]

don't know if anybody else noticed in that latest vid by tom...but that was real time adjustment of lighting on a sub surface scattering material... thats impressive.

I dont think youre paying attention, because none of this is remotely 'realtime' when it comes to 3d, its as realtime as photoshop.
Its saving images in what is apparently a custom HDR file with multiple layers that add ontop of eachother. The pixels were already there, he's just adjusting their intensity/influence on the image.
The big 'huzzah' i can see here is that it saves MW users time on an incredibly slow render. As long as they know what lightsources theyre using in a stillshot, they can wait .. hours to a day for their cleanest render *less time now thanks to their compositing setup with allows the import and merger of those fiels* , and then do post work on multiple layers without having to re-render it several times. If this were applied to say, Vray,FR,C4D,Brazil,MRay, or any of those, this would also greatly increase their render/edit time dramaticly as well.

I think this is something that a good amount of the companies should try to setup. Maybe with a multi image export in PSD or other file formats however so they can be edited in other programs as well. Like for animation. I would sure like to see an array of 1000Tga's on several layers each using seaperate lights on ADD so that the lighting can be edited for animation as well. Like flicking on a light midway through an animation.

Realy the idea of seaperating the lights and their influential properties to their own layers is good and sound, but, not realtime, and quite overhyped in calling it so.

To gain "realtime" NL may want to try setting up nodes that render the scene in under a second on hundreds of computers and instantly composite the data... But that would be kind of overdoing it i think. Then again, this is NL... seems like hype should be their middle name.


Now, back on earlier in the thread we got to see a video of a relighting system. Essentialy allowing animators and environmental editors to see what the final lighting will look like. It Saves normal and color channel data , along with depth from what I can see , and allows them to add in lights and see their overall effect on the final product before actualy hitting the final render again. Thats also something I'd like to see integrated into every piece of 3d software out there. Its obviously not acurate, but for basic lighting purposes, very nice.

[Coliba]

I knew it was a good idea to save the file.

So I re-rendered the two single light passes. The left light has 10x its original intensity, the right light has 20x its original intensity.
http://www.digipiction.de/var/add1.jpg
http://www.digipiction.de/var/add2.jpg



These look correct to you??

If you had such a strong lightsource, the light falling through the transparent object would just be a solid green with sharp edges? I really don't think so.....It just looks like the green part was plastered over the rest.

Would be more interesting to compare a GI scene and see how tweaking compares between this Maxwell implementation and hdr layers, particularly paying attention to shadows/caustics/reflections as lights are adjusted.

[Chris-TC]

These look correct to you??

If you had such a strong lightsource, the light falling through the transparent object would just be a solid green with sharp edges? I really don't think so.....It just looks like the green part was plastered over the rest.


Frankly, I'm starting to get pretty annoyed. I created this entire scene, including the material in maybe 6 minutes. I didn't want to spend a lot of time on it because I was trying to make a point.
Even if the material is not physically correct it doesn't matter at all because that was never the point.

The object has only about 0.4 transparency if I remember correctly. This is why neither it nor its shadow blows out completely at this light intensity.
Sure, go right ahead and take a highly overexposed photo of a mildly transparent, colored object and see if the shadow really blows out completely...


Would be more interesting to compare a GI scene and see how tweaking compares between this Maxwell implementation and hdr layers, particularly paying attention to shadows/caustics/reflections as lights are adjusted.

Yes, of course. At first I needed to go the opposite direction with the compositing, now that this has worked it needs to be GI.
Well, it would work just the same with GI, all you need to do is turn it on.

I'm starting to think that some of you are desparately looking for an excuse to call this Maxwell implementation a "break-through".

The only break-through I see is that you get the post adjusting functionality without increased rendertimes. And I admit, that is indeed pretty cool!

But for one thing, Maxwell rendertimes are high enough already, increasing them any further would be ridiculous. My two single passes rendered out at 640x480 in 70 seconds per image. That's a little over two minutes for both passes. Yes, I turned down the quality but I wonder how much of an image you'd see using Maxwell if you gave it two minutes of render time.

And second of all. Maxwell's proprietary format may be cool in that it has all of this post lighting functionality included. But does any compositing package actually read this data? Using multi-pass techniques, rendering out to HDR formats, you get all the power your compositing package gives you. In other words: you're not limited to just adjusting the light intensity - instead you could run all kinds of filters and adjustments.

[Chris-TC]

Sure, go right ahead and take a highly overexposed photo of a mildly transparent, colored object and see if the shadow really blows out completely...


I just thought about this, and it would be pretty hard to take an actual photo.

So here's my suggestion: why doesn't a Maxwell user create a similar scene? Include an extremely strong lightsource and a half-transparent, colored material and show me what the shadow ends up looking like. I'd really like to see if it turns completely white.

And don't worry, I'll be here all week, so you've got plenty of time to wait for the grain to clear :thumbsup:

[Jozvex]

Rendering now.....hehe

:thumbsup:

[Coliba]

Chris, thanks for the scene, I just wrote that because the result looks very strange and I'd like to test it with Maxwell. This feature they showed is not available to the public yet but hopefully soon.

I don't think it's accurate rendering an overexposed picture right now with Maxwell because that one might look correct, but when tweaking the lighting with this new feature, results may vary. We'll see.....

[Jozvex]

Yeah I don't have access to the nifty new feature yet but here's a super bright caustics render. The glass is 62% transmissive.

http://www.jozvex.com/Maxwell/CausticGlare.jpg

[Chris-TC]

Thanks for the render Jozvex! So there you go. The shadow is sharp and not completely blown out.

The amount of "blown-outness" of course also depends on the intensity of the light. If I turned my light into a nuclear explosion (right now it isn't because you can still see shading on the wall in the back), at some point the shadow would of course blow out completely, no matter how transparent the object is.

Like I said, my object isn't physically accurate because its specular reflectivity, diffuse and transparency values add up to more than 1 which returns incorrect results. It was a rushed scene but it served the purpose.

[Per-Anders]

technically the shadow is solid green there and not blown out in chris_tc's render for two simple reasons.

1) It appears he's not using any GI, there is no bounce light, therefore only what comes through the object is shown, it's a very basic scene and therefore there is no additional light to brighten the shadow from it's raw state.

2) It looks like the object is pure green (maybe very slightly yellow), it may never get fully white, why? well the equation is very simple shadow = light color * intersection object color. To make that more understandable as to why this means it will never get solid, just think of this as seperate red, green abd blue, now if we give it a value, say 10% red, 100% green, and 0% blue then that equation must result in no blue at any time, it doesn't matter how bright the blue component of the light is anything times zero is zero. It would eventually turn yellow though.

The caustics coming through would also eventually spread out to bleach out.

The argument holds true though regardless. Turn off all lights, turn on whatever effects you want, gi, caustics, reflection, refraciton, sss whatever, render, and you'll get black. Add a luminent source (a light) and render, you now have the sum effect of one light on your scene. turn that light off, add in a different luminent light and render, you now have the sum total of that lights effect on the scene. Add the two images togehter with pure add a+b and you have the result of turning both lights on in scene, both their caustics, gi effects, sss effects, reflections, refractions, everything that they illuminate directly or indirectly. Manipulating the exposure of the layers will give you the effect of maniulating the lights brightness, multiply layer with a color and you have the effect of tinting the light and so on.

The point being it really doesn't matter what's in the layer, or how physically accurate the original image/layer is, whether it's rendered in Maxwell or Renderman or Bryce or Blender/Yafray. Adding it together in this way will simulate the same effect you would achieve in your render engine by direct manipulation. The principle is the same, light is light, it's even how your monitor works.

[Coliba]

Just wondering though, if we need hdr images to do these kinds of pretty heavy adjustments, are there any renderers that render a "layered" hdr format in one go, for all the lights in the scene?

Or do you have to do a separate render of the entire scene for each light?

[Cronholio]

Just wondering though, if we need hdr images to do these kinds of pretty heavy adjustments, are there any renderers that render a "layered" hdr format in one go, for all the lights in the scene?

Or do you have to do a separate render of the entire scene for each light?

They are called Deep Rasters.

In Houdini you can, and I'm sure you can in PRMan. Any renderer that allows you to output arbitrary variables from shaders (Light, shadow, surface, whatever) will let you shove any information at all into the channels of an image. You aren't limited to RGBA and Z either. You can make an image that's RGBA in addition to containing point cache info, surface IDs, sidedness, normals in one cahnnel or 3 channels, UV information... You could make an image that shoved integers into every pixel that repeatedly spelled out your name when decoded if you wanted to. Any information you can generate can be put into an image channel.

[Per-Anders]

Just wondering though, if we need hdr images to do these kinds of pretty heavy adjustments, are there any renderers that render a "layered" hdr format in one go, for all the lights in the scene?

Or do you have to do a separate render of the entire scene for each light?

C4D AR should (haven't tested it, but TIFF and B3D support 32bit and layering, EXR only supports 16bit and layers, however it's not gammut locked like normal 16bit, psd only supports 16but gammut locked + layers), i'd imagine that PRMan woudl also output to any 32bit file format with layers as woudl anything else. MR shoudl also be capable, however may require seperate passes rather than making every layer in a single render pass. However M~R is the only one to currently (afaik) do this with GI in a single pass, which does make it a great timesaver feature, but it should be called light layers, or multipass rather than relighting.

[Jozvex]

And I think you could do it fairly easily using framebuffers (ie use FrancescaLuce's ctrl_buffer shader) to do it with Mental Ray.

[rocarpen]

C4D AR should (haven't tested it, but TIFF and B3D support 32bit and layering, EXR only supports 16bit and layers, however it's not gammut locked like normal 16bit, psd only supports 16but gammut locked + layers), i'd imagine that PRMan woudl also output to any 32bit file format with layers as woudl anything else. MR shoudl also be capable, however may require seperate passes rather than making every layer in a single render pass. However M~R is the only one to currently (afaik) do this with GI in a single pass, which does make it a great timesaver feature, but it should be called light layers, or multipass rather than relighting.

At the risk of straying away from Maxwell:

Mdme, given the layering + 32bit features, is TIFF a better format than Open EXR for use with compositers like AfterEffects or Combustion? And if so, can an app like Max actually write layered 32bit TIFF's?

Thanks to all for a really interesting thread!

[rocarpen]

Hmm... did some searching and found the following from OpenEXR.org:

http://www.openexr.com/about.html#history

Conversely, 32-bit floating-point TIFF is often overkill for visual effects work. 32-bit FP TIFF provides more than sufficient precision and dynamic range for VFX images, but it comes at the cost of storage, both on disk and in memory. When creating background plates for VFX work, film is usually scanned in at 2k pixels wide or more, and 24p HD images are 1920x1080, so background plates are already quite large relative to "typical" digital images.
(http://www.openexr.com/about.html#history)

[Per-Anders]

yeah that's more or less the trouble with 32bit tiff. exr is smaller, but the app you're compositing in needs to support it properly. the benefit of 32bit is fidelity, 16bit unclamped is like 8bit in that 0% to 100% is still 0 to 255, it just goes higher than 100% intensity, whilst 32bit is like 16bit clamped, single precision floating point values, i.e. 0% to 100% is 0.0 to 1.0 with 7 decimal places of signifigance (i.e. a lot more accurate).

I'm not sure if Max and it's various render engines support layered 32bit TIFF or not.

[tweeeker]

Of course OpenEXR also supports 32bit float. From the same page:

"In addition to the half data type, OpenEXR supports 16-bit unsigned integer and 32-bit floating-point data types. OpenEXR images can have an arbitrary number of channels, each with a different data type"

So even if you _think_ you need those extra bits, OpenEXR is still often the best option.

T

[stew]

However M~R is the only one to currently (afaik) do this with GI in a single pass, which does make it a great timesaver feature, but it should be called light layers, or multipass rather than relighting.
As I wrote earlier in this thread, C4D R7XL renders all these things in one pass, so I'm pretty sure AR/95 would still do the same. Just because it's called multipass doesn't mean it'll render the whole thing multiple time (in C4D, "HyperNURBS" aren't NURBS and "Radiosity" isn't radiosity either...).

[rocarpen]

As I wrote earlier in this thread, C4D R7XL renders all these things in one pass, so I'm pretty sure AR/95 would still do the same. Just because it's called multipass doesn't mean it'll render the whole thing multiple time (in C4D, "HyperNURBS" aren't NURBS and "Radiosity" isn't radiosity either...).

What's funny is who you're explaining that to. Mdme_Sadie knows a thing or two about C4D, I suspect.

:thumbsup:

[rocarpen]

Of course OpenEXR also supports 32bit float. From the same page:

"In addition to the half data type, OpenEXR supports 16-bit unsigned integer and 32-bit floating-point data types. OpenEXR images can have an arbitrary number of channels, each with a different data type"

So even if you _think_ you need those extra bits, OpenEXR is still often the best option.

T


Thanks, Tweeker. I wondered about that. The site says it has 32-bit support but Sadie had indicated 16-bit. I'm looking to add some serious compositing capabilities to my company's services roster (currently we're an archvis shop, interested in moving into commercials and whatnot) and am trying to research as much of this stuff as possible.

[Per-Anders]

As I wrote earlier in this thread, C4D R7XL renders all these things in one pass, so I'm pretty sure AR/95 would still do the same. Just because it's called multipass doesn't mean it'll render the whole thing multiple time (in C4D, "HyperNURBS" aren't NURBS and "Radiosity" isn't radiosity either...).

Ah no I think you misunderstood me. C4D renders out the seperate lights etc, but it doesn't render out seperate gi passes for each light (without requiring additional passes), and that's more or less what Maxwell is doing here. I don't think any engine apart form MR is capable of generating seperate light layers from it's GI pass (without additional rendertime) currently.

[CupOWonton]

I think they ARE capable of doing it, they just never implemented the idea.

If they can seaperate GI from Diffused Lighting, Reflections, and Refractions, they can sertainly do combined layers attached to individual lightsourced.

[Coliba]

They are called Deep Rasters.

In Houdini you can, and I'm sure you can in PRMan. Any renderer that allows you to output arbitrary variables from shaders (Light, shadow, surface, whatever) will let you shove any information at all into the channels of an image. You aren't limited to RGBA and Z either. You can make an image that's RGBA in addition to containing point cache info, surface IDs, sidedness, normals in one cahnnel or 3 channels, UV information... You could make an image that shoved integers into every pixel that repeatedly spelled out your name when decoded if you wanted to. Any information you can generate can be put into an image channel.

I don't think you answered my question. I wasn't asking if you can put additional info into an image, but what renderers allow you to output in an HDR format, which is layered. Like for example C4D does, but with 16bit psd's. So you have ONE image, but reflections, shadows, GI, caustics are in layers, for EACH of the lights in the scene.

Of course OpenEXR also supports 32bit float. From the same page:

Yes, but again, is it a layered format, or would you have to render a separate EXR image for each light? You would have to render one EXR image, for each of the lights in the scene....to have a HDR format with which you can do what Maxwell is doing.

So for each of the lights in a scene you would have to go through all the rendering process, incl GI precalculations, final gathering etc......which means for 5-6 lights you would have about 5-6 times longer render times....

[tweeeker]

Yes, but again, is it a layered format, or would you have to render a separate EXR image for each light? You would have to render one EXR image, for each of the lights in the scene....to have a HDR format with which you can do what Maxwell is doing.


Not quite sure what you mean by a 'layered' format, but OpenEXR can store an arbitary number of channels, all of which can be different bit depths. So in short you could have a single exr file that, combined with something like shake, allows you do to exactly what maxwell's doing in the demo.

How easily that multi channel exr can be created is another matter, and largely boils down to how flexible your renderer is.

T

[stew]

Ah no I think you misunderstood me. C4D renders out the seperate lights etc, but it doesn't render out seperate gi passes for each light (without requiring additional passes), and that's more or less what Maxwell is doing here. I don't think any engine apart form MR is capable of generating seperate light layers from it's GI pass (without additional rendertime) currently.
I did misunderstand you then. Hm, come to think of it, I'm not even sure if it'd be that easy for C4D to spit out separate GI passes for each light. If their irradiance cache is built like most others, it doesn't keep track of where the light came from initially.

[Mauritius]

Not quite sure what you mean by a 'layered' format, but OpenEXR can store an arbitary number of channels, all of which can be different bit depths. So in short you could have a single exr file that, combined with something like shake, allows you do to exactly what maxwell's doing in the demo.

Another issue is whether you want to put everything into one image. As if one channel is wrong or needs tweaking in the renderer, you have to re-render all channels or split the image into its channels and then re-assemble them after re-rendering the wrong channel.

A lot more pain than rendering the passes into separate images to start with.

Most places I worked at would stay away from putting too many passes into a single image (even though that is easily possible in e.g. PRMan with the RiDisplayChannel stuff), for exactly that reason.


Cheers,

Moritz

[Coliba]

Not quite sure what you mean by a 'layered' format, but OpenEXR can store an arbitary number of channels, all of which can be different bit depths. So in short you could have a single exr file that, combined with something like shake, allows you do to exactly what maxwell's doing in the demo.

How easily that multi channel exr can be created is another matter, and largely boils down to how flexible your renderer is.

T

I'm not talking about channels, that hold info such as position, or UV's. I'm talking about layers that actually show the pixels affected by each of the light source. For example, you have a scene with 3 lights, red green and blue. Your renderer should allow you to render that EXR image in one go, but separate the influence (including GI influence, not only color and shadow) of each of those lights in layers. So effectively you get 3 images in one, and by adding those three layers (as discussed previously) you get an image exactly the same as if you had rendered the scene with GI and all lights on. Or you could adjust those layers individually in Photoshop or whatever to change the intensity of each light.

If this is not possible, then you would have to render the same scene for as many times as you have lights in your scene, and for each render you would have to go through the GI calculations and then the final render......not very efficient.

[stew]

Whether or not you have to render the whole thing multiple times or not has nothing to do with the file format used. When I let C4D spit out separate jpg files for each light, it still renders only once.

[Jozvex]

Another issue is whether you want to put everything into one image. As if one channel is wrong or needs tweaking in the renderer, you have to re-render all channels or split the image into its channels and then re-assemble them after re-rendering the wrong channel.

A lot more pain than rendering the passes into separate images to start with.

I know what you mean and under normal circumstances I agree, but this Maxwell feature isn't meant to be a render passes replacement. It's just letting you do a normal old render and then turn up a light that ended up too dark, etc.

[tweeeker]

Coliba - I think your just getting confused with the terminology. As far as exr is concerned, a 'channel' can contain whatever you want. So two 32bit channels might contain u,v data, or, they might contain the red and green components of a light. It makes no difference. So, a typical r,g,b,a image has four channels. But if you wanted you could create an exr file with 10 channels:

r,g,b,a,light1r,light1g,light1b,light2r,light2g,light2b

that would contain rgb channels for 2 lights. 3 lights would need 13 channels etc etc.

Moritz - your right, this wouldn't be a good idea at all - I'm just trying to point out that it's possible.

T

[CGTalk Moderation]

This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.