VN Daz Studio Workflow/Compositing assistance

[Fekkin]

I'm new to this and I've recently started working on creating a VN via Ren'Py using Daz Studio and Daz3D assets. I'm mostly hoping to use 3D assets in my scenes as opposed to relying on integrating characters with HDRI backgrounds. I've opted to predominantly work with Genesis 8 and 8.1 characters and associated assets/props.

My GPU is an RTX 3080 (unfortunately the 10GB version) so I know I'm somewhat limited in VRAM. I've bought tools such as scene optimiser and camera view optimiser alongside other methods for reducing VRAM usage, but I'd still like my workflow to be modular so I don't have to bake full renders for every scene I create.

I'd like to have the same background/main scene for multiple shots, with the characters & some props being more dynamic (various poses, movements, etc). So I was hoping I could render the main/background scene seperately, then independently render characters (while having them affected by same lighting) and composite them in Photoshop afterwards so the main scene plate can be reused.

Unfortunately I've been really struggling to find an optimal way to do this. I've tried using canvases, but unfortunately there seems to a bug with the Iray Alpha canvas that completely ignores the assigned node list, so I can't seem to mask my characters from the main scene. I tried adding an alpha channel to a second beauty canvas with a node list but it missed half of the assigned nodes. I also tried applying the matte property to emissive buildings as found in a suggestion but that didn't work either. I've tried other methods, but they either remove the effects of emissive lighting on the characters, or are unable to effectively seperate the characters from the main scene for compositing, forcing me to rely on baking my characters into a full render for every scene. Reducing opacity & increasing luminance, adding shadow catchers... No matter what I try it just doesn't seem to work the way I hope.

Maybe I just started with a very awkward scene (outdoor night city scene with emissive buildings, street lights, etc), but at this rate it seems like it'd take forever to create a decent VN with fully baked renders and no modular approach. I'd be extremely grateful to hear any advice, ways around the issues i've been having, and general workflows/best practices/methods that current VN creators have when working on their renders and how they approach dynamic scenes. I want to be able to show different expressions & movements on characters without having to render a full scene every time.

This has been driving me nuts - please help! Many thanks!

 

[SirLecherous]

That's alot of words my friend -_- , almost exceeded my reading quota for the day. Some more qualified will answer later but if not why not do batch rendering, for the lighting you can check out 'Daz ghost lighting setup" on YT . Also check out renpy layering (idk how efficient this is but some swear by it)

 

[Egglock]

I'd like to have the same background/main scene for multiple shots, with the characters & some props being more dynamic (various poses, movements, etc). So I was hoping I could render the main/background scene seperately, then independently render characters (while having them affected by same lighting) and composite them in Photoshop afterwards so the main scene plate can be reused.

I don't render in Daz3D, So I'm not sure if this is the optimal way of doing things. Something you could do is, hide your objects using the outliner. To the left there is an Eye icon you can click on to hide/unhide objects. Do your BG render with characters hidden, and then characters with BG hidden.

I don't believe Daz3D has any built in features to exclude objects during the render phase. As in, X,Y,Z should not be included during render. So you might have to do what I mention above if you want to render objects separately.

 

[caLTD]

Daz3d has render visibility option for each object and it can be on off in each timeline frame. Check parameters.

Save your scene as scene subset.

Use lights as another scene subsed. delete scene keep lights. then rencer character.


My Sugestion, if financially possible buy secondhand rtx 3090. Which I do and my daz workflow much better now.

Of course, I still hate daz3d my bones. But it always taming the beast...

 

[Fekkin]

Appreciate the advice, thank you all!

 

[AllNatural939]

Frankly, I don't understand what your problem is with rendering full scenes... Look, this image was rendered with a 12GB RTX 3060. All the characters are G8... I don't remember the exact render time, but it was under 10 minutes.



The view from the viewport so you don't think I'm lying...



And when rendering in the viewport, it uses more or less the same amount of VRAM as the final render, so...


With some basic optimization, you can achieve much more than that...

This images has 20 G8 characters on screen and was rendered all at once. Some of the characters and their clothes, further from the camera, were pre-optimized.





"Physical" obstacles don’t count for much when you make up your mind to overlook them.

 

[SirLecherous]

AllNatural939 TF ...so many questions
how much ram do you
have also did you change the resolution
or sub division?

 

[AllNatural939]

AllNatural939 TF ...so many questions
how much ram do you
have also did you change the resolution
or sub division?

32 gigs of RAM. But the images of the 20 characters were made when I had 16. No, I haven’t touched the subdivisions, I only changed the resolution, but not for all the characters. Around half of them, the ones farther from the camera, are set to Base resolution and the others are still in High. Hairstyles for everyone are in Base. Half of the clothes don’t have textures, the other half have the textures reduced to half... Skin and hair for all of them with textures cut in half. With ridiculous stuff like eyelashes, textures reduced to 512, why the hell would I want them in 4K xD
This is just basic pre-optimization in the 20-character scene. In the indoor scene with 9, there’s almost no optimization, just some textures like hair reduced to half. Maybe also hair resolution reduced, I don’t even remember, but generally, hair in high resolution at mid-range shots doesn’t show any visible difference compared to base resolution, so if you’re tight in a scene, keeping high resolution is useless.

 

[CtrlAltNEET]

To composite you need to ensure:
1) you have maintained the major shadow-casting and light-emitting objects in each render
2) you must ensure that overlapping objects are rendered together within one render (e.g., if two characters are holding hands, there must be at least one render where both their hands are maintained in the render, and probably their whole bodies because they will be shadow casting onto eachother)

The two points above seem quite obvious and you would have arrived at these independently already I assume while working through the problem.

In practice, this might look like rendering one pass with just the background. Then a second pass with the character and background visible but with all insignificant (non-shadow casting non-emitting) props hidden. Then compositing them together in Photoshop etc.

Pro tip: spot rendering out to new window will make rendering those parts you spot render way faster than if you rendered out the whole image.

To people proclaiming you can render out 20 x G8 characters in one pass. Be that as it may, it's going to be difficult or next to impossible to do so if they are all high subdivision characters, or each have different textures, and each have detailed clothing and detailed SBH with millions of vertices each, and the background actually has props and detail in it. spot rendering out character details like clothing, hair, and hiding background parts will obviously help, but generally you are going to need to do multiple passes and compositing for most situations where there are crowds. Or sometimes even just two high subd characters in a scene.

subdividing a character just once quadruples the polycount. A strand-based hair with millions of verts might be 10 x the amount of polys as hair cards. If 20 characters use the same textures, that's just one texture call to VRAM. But if they use different textures, then they might each have 20 texture maps to load. So "number of characters in a scene" is not a perfect proxy of VRAM use, and even a low number of characters could require splitting up the render.