I know this is probably one of the most-asked questions on any 3D modeling forum, but I can’t seem to find a definitive answer to it anywhere.
How high-poly is too high-poly, for 3D animation?
I know current-gen game characters average around 10K faces, and Star Citizen is pushing its next-gen characters up to around 100K, but that’s for real-time rendering on powerful higher-end computers. I’ve never seen a rule-of-thumb poly-count for assets and scenes in pre-rendered animation, though. I know it can be considerably higher than real-time, but how much higher before you start waiting days frames to render, I don’t know.
So, basically, I’m looking for ballpark poly-counts for different categories of objects in an animated scene: primary assets (main characters, major props and vehicles, etc) and secondary assets (minor characters and props; sets, etc.)
I’m asking, because I’m constantly scrapping assets I start because they’re either too high-poly for their hierarchy or importance (modular set pieces being 50K polys or more, often) or too low-poly (~1K, max) to look acceptable even after texture and material application.
If someone knows some good numbers to shoot for, I’m all ears (eyes?)
It all depends on the scene view and details you want to provide. I would go with mix of High, low, Medium poly dependent on the basis of the visibility of the object.
Then there are lots ways to reduce the poly count like Instancing (render time) and grouping (3d view) of objects.
For starter, lets say, you have a view of street, full of shops and buildings. if your scene is not going to be shown from behind, you don’t need to fully create the buildings. Just the front is ok. You can also use Pictures/Images too for far behind objects.
I suggest watching some animated scenes to know more about the pre-render animations.
There is a tutorial “Create a Missile Turret Scene” , it will give you lots of insights on animations as well.
Besides the things mentioned by dukejib it also depends a lot on the time and budget you have.
For example you will often get CAD data from a client. CAD data tends to be extremely highpoly and pretty messy. It´s often difficult and time intesive to reduce or retopolgize and can take long to render. If you´re on a tight dead line you will often have to use the CAD model to reduce production time and throw money at a render farm in order to reduce render time.
If you have the time to create clean models it makes sense to do so of course. This rarely is the case in my exprience.
Well, the project I’m working on now is a personal one, with no budget and no real deadlines, so I don’t have any restrictions, other than the limits of my patience and (gradually improving) 3D skills. I’m using the development stages I’m in now to try to learn to do things the “smart” way, with clean model topology and good-looking textures and materials; so that maybe by the time I get into full production mode, I’ll sort-of know what I’m doing.
The idea behind the project is for me to create the first sequence of the pilot episode for the sci-fi web series I’m writing. I’d then be using that sequence as a demo and promotional piece to try to secure a team (and possibly funding) to take the episode to completion, and hopefully create additional episodes.
And in case anyone’s wondering, what I’m working on now is the Roxy Future series, as linked in my signature.
Maybe you can post a couple of your models you think have a too high or low resolution and you can probably get more accurate advice.
There´s really no such thing as a rule of thumb for things like that. It allways depends on a gazillion factors.
I’d post up a few of the too-high-poly models, but, like an idiot, I deleted them in my frustration in favor of starting from scratch. I’m really trying not to end up with a million-plus polys in a small set similar to this, basically.
No this isn’t the one. I tried to find it but failed. Anyway, there are lots of tutorials , non-blender too, which gives the insights about pre-render animation.
I tend to create scenes like this iteratively. In order to do this you will need some sort of film script.
Create the scene very roughly with boxes or simple geometric objects
Create the camera and animate it acording to your script
You now see which elements are in view, which elements are close and require a lot of detail and which elements are far away and require less detail
Don´t forget that some objects offscreen might be needed for reflections, shadows and GI.
Detail your scene a bit more. Lowpoyl, only several thousand objects, so you get the outlines of the objects right and some large detail
Now I usually start lighting the scene. It´s fleshed out enough to get a good feeling for what the lighting will be like in the final image and still sufficiently low poly to get decent rendertimes
It can make sense to add materials now as long as they don´t require intensive unwrapping. But often a a simple box mapping will do for getting a feel as to how the final image will look like
add more detail to the objects that require it
show to boss/client. He will like it.
tweak lighting
add final detail
show to boss/client. He will still like it.
tweak lighting and materials/ add effects/particles/athmosperics/similarstuff
add or remove detail until it´s just right
Show to boss/client. He will now demand to change the camera animation completely and you will have to detail completely different objects and throw away 70% of your work.
Your iterative approach made me realize a few flaws in my own initial approach, so far, so thanks for that.
I’m used to doing real-world filming, where all your assets physically exist in finite detail, whether or not they’re in-frame, and you just pick your shots according to the storyboard. Doing that with 3D animation takes up much more time and system resources than is necessary to produce a final image within a reasonable amount effort and time, obviously, so it’s best to only do as much work as you need to to get the result you want.
I’ve got the first draft of the sequence’s script done, and I’m looking at doing some minor revisions before calling it final. The next step is to storyboard it, which is going to be a bit trickier, since I suck at hand-drawing and I only have the demo version of Frameforge. I’ll figure it out somehow, though.
Your approach made me realize that I really shouldn’t start building the set in earnest until I have a storyboard ready to work from, so I know what parts of the set I should concentrate my modeling efforts on, and not waste time and energy on parts that, more than likely, aren’t going to be seen during the sequence.
