how to get same detail as TFD CPU render in Octane?

[sdanaher]

Think large pyroclastic clouds with lots of surface detail. Tried decreasing voxel and step size (0.5 and 0.05 respectively) and increasing density but I don;t get the same intricate details as the TFD render which also renders much much faster. Is it possible for octane to be faster with more detail or am I expecting too much? Seems Octane is fast at more transparent smoke, while TFD is better at dense clouds.

[aoktar]

As i know Octane evaluate voxel with full contribution to all pathtracing. So when you compare the speed you should activate GI, raytrace shadows, etc.. for C4D renderer.
About other question, have you tried to use VOLUME GRADIENTs? It's similar way as TFD's gradients.
TFD has some upsizing option to add extra voxels on render time. But we don't have this. I think you can use voxel size = TFDvoxelsize/2 for same details. But decreasing size 2x time will cause 2x2x2=8 times more calculation and VRAM usage.

[sdanaher]

As i know Octane evaluate voxel with full contribution to all pathtracing. So when you compare the speed you should activate GI, raytrace shadows, etc.. for C4D renderer.

I understand Octane has different priorities, but if that makes it slower than TFD because its calculating stuff I don't need then it IS slower than TFD.

About other question, have you tried to use VOLUME GRADIENTs? It's similar way as TFD's gradients.
TFD has some upsizing option to add extra voxels on render time. But we don't have this. I think you can use voxel size = TFDvoxelsize/2 for same details. But decreasing size 2x time will cause 2x2x2=8 times more calculation and VRAM usage.

OK, TFD must be adding some kind of displacement detail not present in the sim because Octane can't recreate them no matter what resolution I use.

Just tried volume gradients, they don't affect the surface detail though. Will there be a way to add noise shaders to the volume to get a similar extra detail effect as TFD?

[aoktar]

Sorry but It may not make any sense on you but this is a pathtracer. I don't think that TFD consider any pathtracing in their voxel calculations.
All of you asked questions are about renderer. Plugins cannot change the fundamentals of renderer. I think Thomas can answer better. But as i know that when OSL comes will be possible to assign textures to voxels.

[sdanaher]

Sorry but It may not make any sense on you but this is a pathtracer. I don't think that TFD consider any pathtracing in their voxel calculations.
All of you asked questions are about renderer. Plugins cannot change the fundamentals of renderer. I think Thomas can answer better. But as i know that when OSL comes will be possible to assign textures to voxels.

I have no idea about any of that. My questions are from a purely aesthetic, time/cost perspective. That's that point of my post, not to make an abstract comparison between technologies but about tools in the real world used to produce actual work for real clients and deadlines. When I make these evaluations its costs me time, hence the questions, because I don't want to be wasting it chasing something that's not possible. its not a criticism, I just need information so I can make a decision on whether to spend my R&D time on something more productive.

If it makes you feel any better I just ditched Arnold in favour of Octane on a recent job, because Octane was quicker and had a better look.

[aoktar]

I'm just trying to make best bridge between C4D and Octane. Also i'm not just developer but 3d generalist for years. So i'm not frenchman for what you tell. But there is a truth and math behind all these tools.

[aoktar]

I have implemented VOLUME object since a few version back. I think not so much people is aware of the power of that. I suppose this is the way to go rather than pyrocluster things.

[haze]

Think large pyroclastic clouds with lots of surface detail. Tried decreasing voxel and step size (0.5 and 0.05 respectively) and increasing density but I don;t get the same intricate details as the TFD render which also renders much much faster. Is it possible for octane to be faster with more detail or am I expecting too much? Seems Octane is fast at more transparent smoke, while TFD is better at dense clouds.

I'd like to know more about your use case of volumes -

1. how complex is the scene besides the volume?
2. what is the volume resolution?
3. can you post one or PM me the vdb?

As a result of being physically based, Octane inherently takes into account self-shadowing and multiple scattering. Experiment with the following to reduce the effects of these (it will speed up render significantly)
1. reduce diffuse/specular depth to 2 or 3
2. turn off alpha shadows

[sdanaher]

Hi Haze

It's rendering high-res print imagery for advertising, competing with photography so it mustn't look 'cgi'. At all. Type 'powder explosion' to see the kind of look and level of detail we're talking about. There are multiple components in this kind of volume structure, from particulate debris to smokey puffs. The middle structure is the 'pyroclastic' like stuff, not very transparent quite dense and with a lot of fine surface detail. We need dark shadow details in some places with rich structure. This kind of image would be built up in post with multiple cgi assets using different techniques for the different parts, even combining different renderers. I've tried Arnold, Krakatoa, xParticles, TFD and Octane. But I've not found a way to get the look I'm after in a reasonable time.

I've tried reducing both diffuse and specular depth but not alpha shadows. TBH it seems TFD's own renderer shows the most promise, but times still become exponential when you start ramping up the frame size and sim quality.

(Volume resolution is key to getting fine detail, but the higher you go the more you have to subdivide the overall explosion into passes for it to be manageable, but then you're just loading the back end of production to take pressure off the front end. You can choose where to put your resources (higher sim detail with Octane doing the rendering, or higher post-sim detail with TFD) but its always a compromise and which will allow you to make changes more easily when feedback comes in? It quickly becomes unmanageable for the level of feedback/changes my clients expect)

[MoGrafik]

Hi Haze

It's rendering high-res print imagery for advertising)

Hey, just thinking sideways a bit here, but if it's a print / still thing, I'd imagine you'd be much better off forgetting about rendering the smoke / atmos / volumetrics, and do yourself a favour and look to source some high-res smoke stock and implement it in post production. That ought to save you tons of time and technical workarounds.
If it really has to be rendered, why not render out TFD as a separate pass and composite it on top of an Octane render? I've experimented with this and found it can work well. For me the big difference is that TFD native renderer allows for sub-grid detail / veolcity displacement, which can be nice for achieving a sharp, wispy look.