How 3D model fidelity is usually measured?

Cake_Seller

Hi. I'm learning 3D art and I wonder how 3D artists and their clients usually negotiate the fidelity of the assets.

Let's say I'm searching for a 3D artist who will create some game assets for me. For example, I need a game-ready asset of a gun. I expect it to have some minimum level of fidelity. How would I describe this level?

I would probably try to describe it with some of these statements:

• I want an asset to have like 8,000-12,000 triangles.
• I want it to have one 2K texture.
• It should look good when the gun is rendered on a 1000×1000 pixel image (no visible shading artifacts, the silhouette doesn't look too low poly etc.).
• Show some references and say "It should be this good".

WDYT does it makes sense? Please share your thoughts. Or maybe you can share an example of a good specification that is provided to the artist?

Kanni3d

Geometry wise - as long as silhouette is established and there are no visually faceted looking areas should be your rule of thumb. Within reason, the 8k-12k budget is arbitrary, as all assets are different. That 12k limit may work for one asset, but what if another asset is much more organic with tubes, pipes and cables? Once you put a budget/ceiling on something, there could be some areas of the model that will suffer and look chunky/low res because the artist was forced to be under that limit. Whereas if there was no limit, the model could look much more defined with an additional 2-3k triangles.

Texture wise - most of the asset should have a uniform texel density, with the exception of hidden/not very commonly seen areas being shrunk down to increase resolution. There shouldn't be any hint of low-res/blurriness. Again - 2k is arbitrary here, as some assets can look great with a 1024 or 512 (if small enough in size). Most artists will abuse overlapping/symmetrical UVs to maximize their UV space for crisp textures.

Cake_Seller

@Kanni3d is it common to have maximum triangles amount in asset specification? I understand that sometimes an additional 20% of triangles can make the model look much better but still probably there should be some target tris count (maybe +/- 10%) otherwise something that in artists' opinion is low poly might appear mid poly in the client's opinion. So probably it makes sense to specify target tris count as a starting point, and then if artists think that additional tris can make model look much better they can negotiate it will the client. Does it make sense?

Kanni3d

It's more of a "soft limit". Assuming this is for current/next gen hardware and platforms, geometry really doesn't matter too much. This soft limit can simply be like "ok wow this asset is 40,000 triangles.....pretty sure this can look identical at 20,000 triangles given its scale + amt of details".

Nowadays, just make things look good within reason, and without having unnecessary loops not offering anything to the silhouette. But this doesn't invite artists to be reckless and go wild with spending geometry everywhere. Artists should still be optimal and vigilant, but generous with topology in areas that really need it (curvature, organic areas, soft areas etc.).

Cake_Seller

@Kanni3d thank you for your reply 

Alex_J

agreeing with kannid, in general it is easier to get a first draft artwork produced first, then review it and see how it can be improved (both technically and artistically.)

And that is easy. If they give you a 2k map, just reduce its size and if you can't tell a difference, great, go with the smaller texture. Same thing with geometry, just see if you can delete loops. 

A more experienced artist will do that sort of stuff before delivery, a less experienced one might need more pedantic instructions. Some things should just be obvious - like if you need a small prop to work as clutter in a top down game, and they have three materials used for this...they are just a knucklehead and you'll have to say more specific instructions like, "max one material." 

But it's better not to shoehorn from the onset because a million things change and consolidating the art assets is the easier part of game dev - so it makes sense then to let them be oversized and modular until you really know where the bottlenecks in your project are.

poopipe

your reference point of a 1000x1000 pixel image is really useful.  

the thing that most artists seem to struggle with is that polycount is entirely dependent on view distance - I'm constantly asked what the triangle budget is for assets and the answer is simply  - as many as you need for it to look good at the distance you see it from.  

the advice above about not having geometry you don't need is entirely valid  - whether you need it or not is determined by how many pixels it covers on screen