How to render lots of background characters quickly?
Testing6790
Posts: 1,091
Hi! I'm working on a huge project with tons of people. A lot of them are background figures, so if there is a way to increase render speed while having a large amount of people, I'd love to hear it. Ideally it wouldnt be something like reducing the geometry, because I do occasionally do closeups of the background people too.
Thanks!
Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2025 Daz Productions Inc. All Rights Reserved.
Comments
I was wondering the same thing. I have a render in mind with a lot of statues, and was wondering if I could set up the basic shape for a figure, and then export it as something that I could then load into a scene that wouldn't take up as many resources as a complete Genesis figure for each figure in the scene.
I'd try by setting all distant Genesis figures to base resolution in parameters. Distant/background "extras" characters really don't need to be sub divided. I'd also go into surfaces and make sure SSS is not active, as its not necessary for distant characters either. If those distant characters are using AoA SSS shader I would change the shader to HSS or Daz default as they dont have the waiting period before rendering starts.
I however never really have more then 5 genesis figures in any scene simply cos its to hard on the computer, not only that but the layers of clothing on top of that with all the lighting/shadows and the set/environment its all very taxing. So typically I will render in layers, of say 4 characters and each layer has its own characters which I combine in post work in Photoshop.
Or if possible use one of the Lo Rez figures for the crowd, that's what they were made for.
Powerage's M.O.M Crowd Generator springs to mind, I understand that this uses 2d Billboard type props to create background crowds/groups, there are various add on/expansion packs available at Renderosity, the original pack is here: http://www.renderosity.com/mod/bcs/m-o-m-crowd-generator/71815/
Please note that I do not have this (but have looked at getting it) so can not confirm how effective it is but his products are normally high quality stuff.
The other way is to render several smaller groups & combine in postwork, I have something I did this way a while back but I need to head to work so I'll dig it out later if I get a chance.
My first move is the Lo Rez characters which work great and after that you can always layer things in Photoshop. Realistically on my PC I can 3 full high res characters in a scene, with background and props, before it slows down to the point of a crawl. So the Lo Rez characters are awesome fillers for my scenes.
..I have it and used it and it works well. You may need to do a bit of adjustment to the surfaces channels to get them to look right depending on the light setup you are using.
Question, is there some way to reduce the level of detail. I know you can increase it.
Use a low res proxy figure from Make Human. They have one that's under 1000 polys.
How about instancing. I've never rendered a crowd scene though, only tried it on lights, which it doesn't work on as they have no geometry.
I think it uses less memory as the object is already loaded. Not sure if you can change the colour of the clothing etc. on the instances though.
Yes you can use instancing on figures but you can't change anything except the XYZ coordinates, rotation and scale. If you alter the source figure the instance will change as well.
You don't need to make a long sequence-plan with all the takes in the same scene setup... so when you go for the background close-ups just substitute the few characters that will be focused on in the close-ups with hi-res ones, using low-res crowd people for the background rendering.
Texture files can chew up a lot of memory, there is no need for 4000x4000 pixels textures on background figures.
As I mentioned earlier I have built images from several renders that I assembled into a final image in postwork, using GIMP in my case. Below is one that I have along with the main images (in my next couple of posts) used to create the final image. This was created for the last PA wallpaper contest (Jan 2012) as a banner for the relaunched contest that was supposed to happen after the website relaunch but it never happened. The full final image is 2400x1280 as per the contest rules and I took first place in the contest but as the PA Wallpaper contest never came back my image was never used.
This was done on my current machine, an Intel Q8200 (a 2.33GHz Quad Core) with 4Gb of RAM, far from cutting edge. Loading and working with these files did cause the machine to slow down noticeably but it was still perfectly usable. It should also be noted that the figure groups in the images where often assembled as separate groups of 2 or 3 figures that where saved then later merged into the large groups you see rendered in my next post along with a few more details.
Following on from my previous post:
I started by building the basic structure of the image, the canal, path wall & tower then positioning & locking the main camera and setting up the lighting (an uberenvironment sphere in this case). This was saved and then loaded as required to position figure groups on it, deleting segments that where not wanted in individual renders.
The figure groups where posed as small groups on a single section of path, attached to a null, the path section was then deleted and the scene was saved.
I would then load the main image, delete the unwanted sections (water, wall & path sections etc) leaving an overlapping section of path & wall to help blend into the next image, merge in the required figure groups and reposition them. Then I saved the scene with a new file name before rendering and saving the image as a PNG file, to preserve the transparency.
Repeat as required then load the images as layers in your preferred image editor (I use GIMP) and then I erased parts of the overlaps to get the final image, adding a background, effects & text to produce the final image. Note that the individual images contain elements of the next image to make the blending as easy as possible.
Note, the separate parts below are PNG files (1000 pixels wide rather than the full 2400) so retain their transparency. If anybody wishes to use these to play around with the blending method then please feel free to do so but please do not repost or distribute the files, claim as your own etc.
The exception is if you wish to post your efforts here in this thread to show off your work.
Thanks.
You will need to add a background of some sort, and any other elements that you wish.
As the maximum number of images per post is 5 I attached the final couple of images in the next post
The last couple of images to build the final image as in my earlier post.
Wow thanks for the detailed reply
Texture atlas can be used to easily reduce the the image texture sizes for all the images used by a figure all at once. Of course that does potentially freeze your textures in a far less easy-to-modify state, so it would be most easily used once you were happy with a figure and intended to then concentrate on other parts of the scene after that.
With instancing, there is one tiny cheat that you might be able to use to get an instance that doesn't have EXACTLY the same pose as the original: Resize the object to -100% width. It will be a mirrored pose, which if also turned at a slightly different angle and used in the right place in a scene, might allow you to create an instance that looks like a different non-instanced figure, if the textures should still be nearly identical.