Old Dog Needs New Tricks - Game Asset Advice please?

Replies

• 
  Options

  Offline / Send Message

  Kanni3d ngon master

  Jun 2021

  Hey man! Here's my two cents on what I can answer to help clear things up.
  

  
  

  1. Polygons
  

  Do we still refer to Triangles as Polygons rather than Quads? This was starting to get a bit blurry last time I was in a studio. More recently, I've seen meshes referred to more by their vertex counts. Is this the new standard, or does it vary?

  
  

  In terms of stuff that goes into engine, almost always they are referred to as triangles. Polygons is a bit misleading, since they can be approximately twice the actual geometry (a crate that is 10k "polygons/quads" can be ~20k triangles in the end).

  
  

  2. Poly Counts

  What would you suggest is a common poly/texture budget for a hero prop like the crate design below?  I understand that a hero prop featured more prominently will have more polys and probably a bigger texture size than something that's background dressing, and I've probably erred well on the side of caution with the poly count here. A lot of the detail I pushed into the normal map that could be modelled into the actual crate itself?

  
  

  A crate like that, just by giving it a glance and seeing tubing/wires, large chamfers/beveled shapes, and a lot of rounded elements- I would expect its LOD0 to be anywhere from 6-12k triangles.
  
  Just use enough geometry as you need to get your shapes and silhouette across, without sacrificing fidelity, and without being careless as well. Have your mind focused on how good the shapes read, and not focused on hitting a budgeted ceiling.
  
  Geometry cost is at its lightest as it has ever been, and as you know, is not as such as a tight bottleneck as it was 10 to 20 years ago. If you notice any faceting/chunkiness in your model, especially a hero prop, you do not need to think twice whether or not you need to add some more loops or subdivisions - absolutely go for it, and make it look slick. But it is still nice to be resourceful and terminate your loops and remove any erroneous and useless loops/triangles.
  

  
  

  3. Topology

    Is it still the case that a mesh has to be airtight, with no geo embedded or resting on the surface of another piece? There was a time when it all had to be merged into one continuous mesh with no T-junctions, or a vertex sitting on an edge and not connecting to another vertex. Is this no longer the case, or is it a situation based decision? Doing it the embedded way saves on polys but can cause issues with wasted UV space and polys don't appear to be the issue they once were but texture sizes still are?

  
  

  I think this would be more of a situation based decision. Meshes can absolutely have some detached/unwelded elements to them if the reasoning is sound. Especially if things are interactive/move/rigged etc.

  
  The embedded/floating geometry method saves on tris, and is much easier to LOD as well. Depending on the asset as well (which is why it's situation based) it just may appear unsightly if its a hero prop or a baked asset and has these floating/unwelded details.The slight disadvantage/downfall is very much that it would waste UV space, as the space under the embedded/floating geometry is unseen etc.
  
  

  It's nice for many reasons for things to be watertight, but it wouldn't be bad for the reasons that you'd be concerned with from games back then.

  
  

  4 Textures

  Like most artists trying to get up to speed, I've been looking around at what's considered cutting edge as a good benchmark to try and emulate. Consensus seems to be Star Citizen for the kind of stuff I enjoy designing and building. I realize its PC only, and you probably need a mortgage to get a rig that does it any justice, but are these texturing techniques filtering through into current console and indie games? I know it was used in Alien Isolation a few years back, but how many studios use it as standard today?

  
  

  To recap and make sure we're on the same page about Star Cit/Alien, the way that they handle textures is by having almost damn near everything procedural/tileable textures followed by some real robust decal shaders to embed tons of details/breakup on these procedural materials. They handle edge highlights/beveling with surplus geometry, negating the need for unique normals or any baking at all. This is mostly for environment art and filler/background props, and doesn't apply too much to main/hero props (by this I mean that hero props at least are almost always uniquely unwrapped, baked, and textured). It was pretty groundbreaking 6-7 years ago, some would say revolutionary, as no one has really seen such quality/fidelity until then, but I wouldn't say this level of quality is a standard. It's definitely put into practice on a bunch of games and is widely recognized as a quality bar though. Most modern games handle texturing with atlases/trims/procedurals, and only now slowly using geometry to their advantage to handle edge beveling etc (Couple of Far Cry titles I've worked on rely on this for enviro art especially).
  

  
  

  Something Star Cit is also known for is Face Weighted Normals/midpoly technique which is a "newer" workflow that definitely was not in production in the olden days. Alien Isolation was a spearhead for this too - this method is worth looking up and handy to know.

  5. POM and Bump Offset.

  Trying my hand at the Star Citizen workflow (be gentle, I'm a noob) I can definitely see its worth in preserving that pin sharp clarity of the decals but also notice you have the drawback of the tiling textures repeating. Any distinct wear and tear on the texture noticeably repeats. 
  

  That's natural, and a small disadvantage to using tileable textures - you lose any unique wear and tear, and also risk having repeating details. It's worth looking into making sure your texel density is set up correct, and is uniform (and also respects your texture map sizes). I haven't mentioned this yet but texel density is huge, and seen as king in making environment art look consistent, sharp, and uniform across using many different texture sets.
  
  

  Am I correct in thinking a lot of UV sets per asset are needed to blend a range of materials and weathering  over the top to break up this tiling? If so, how many usually?

  If you mean UV sets by different material IDs (like a uv set for the lid, and another uv set for the body etc), it can help, but it won't solve or mitigate the inherent repetitiveness of the tiling textures. Assets do receive many tileable materials, which in the end are just seen as drawcalls. How many really comes down to the asset - drawcalls are the "budget" here. Seeing a surplus of drawcalls for an insignificant/tiny prop is concerning for example. I've dealt with some large and important assets that have anywhere from 5-10 drawcalls on them.
  

  
  

  Example:  Asset_A has 5 drawcalls as it references:
  

  • two instanced metal procedural materials
  • an atlas for screws/bolts
  • an edge wear trim material for metal scratches that are floating on the asset
  • another atlas decal for branding/text.

  I could be a bit foggy on this though, as there's much more to drawcalls than simply what materials are being referenced by the model, but at least it would be a big indicator of how expensive an asset is, and can easily be assessed if it's warranted or not.
  

  
  

  You could blend materials in a few ways to break up the repetitiveness/uninteresting nature of just having a plain tileable. Masking with utilzing secondary uv channels is pretty nice since you can target/style precisely how you want another material set to blend on your model. Vertex painting is a traditional but valid option.
  

  
  

  
  

  I hope this helps and was clear/concise. I hope others can chime in on any additional info to help you out further.

  
  

  Also I'd like to also say that it's very admirable, and really cool to see really long-term vets from the "dark days" of game dev still kicking around and making a leap to "modernize" their workflows. Good on you, and can't wait to see some of your work in the future
  

  11 · Share on FacebookShare on Twitter
• 
  Options

  Offline / Send Message

  LeeRay polycounter lvl 2

  Jun 2021

  Many, many thanks to Kanni3d for that fantastic reply above. Working in isolation is a nightmare when you need solid info to learn from, and the above has been really useful in confirming what I thought I knew and informing me in areas I was ignorant.

  With this new guidance, I've gone back to the Cryo-Crate  mesh that was way under the  triangle budget and brought it closer in line with the 12K suggestion.
  

  Primarily, I've transferred the major details I had put into the normal map and found I could add as geo to the mesh.

  Also bearing in mind that the POM workflow is not as dominantly used as I suspected I've unwrapped this 12K version and treated it like a standard prop with mostly unique texture space but wanting to preserve a 512 x 512 pixel = 1 metre squared texel density on a 2K texture there has to be  a few areas where UV space has been shared which shouldn't be too noticeable if I've done it correctly.
  

  
  

  I've still a few polys left that I will use for decals and some smaller geo details in certain areas once I've got further into the materials and texture baking.
  

  
  

  My questions in this post relate to UV unwrapping, such as pipes or tubes.

  I've seen a few tutes where artists have forced an unwrap to be a regular grid of uniform squares even though if you relaxed it, it would take a more organic irregular shape, as they have done in my UV layout. I can see the advantage would be more efficient use of the UV space, but doesn't this also generate uneven texel density with compressing and stretching in bends on inner and outer edges if the quads have equal UV coverage?

  
  

  What am I missing here?
  

  
  

  I'm also a bit unsure of the best practise when it comes to the rounded, angled insets on the panelling. If I detach them and release the shells "tension" by cutting its continuous looped flow and unfold them, the shell edges will straighten on the parts they were connected to (the actual side panels of the carte for instance) but the layout will be less efficiently packed as a result.

  
  

  Does it matter in this case as there's nothing noticeably distorting such as labels or text I'm baking into the base texture, and are small distortions like this an acceptable trade off for efficient UV packing?
  

  
  

  
  

  0 · Share on FacebookShare on Twitter
• 
  Options

  Offline / Send Message

  LeeRay polycounter lvl 2

  Jul 2021

  UPDATE AND FINAL VERSION.

  
  

  Based on what I've read regarding texture budgets, I've taken this as far as I can.

  I used Substance painter and Photoshop to produce PBR textures - 2K for the crate itself, using 512 X 512 PX = I metre squared as my level density. A 1K Colour Decal sheet also with an emissive channel for the lit displays and lock screens and a 1K trim sheet of sorts for the finer normal details such as the screws, panel lines and more greebly parts on the crate ends.

  
  

  Turntable rendered in Marmoset Toolbag.
  

  
  

  Am I missing anything?
  

  
  

  Feedback always welcome
  

  2 · Share on FacebookShare on Twitter