3D models for the Dark Mod

[Frank Cotton]

Is there documentation about creating 3D model for TDM especially for animated characters?

thanks

 

[thebigh]

Yes, the Dark Mod wiki has a lot of useful information. You can find an index to some relevant pages here.

[Frank Cotton]

thanks for indulging a lazy cunt like me

[Frank Cotton]

It says the poly limit is 5000 polys does that mean 5000 quads are ok or does it actually mean 5000 tris is the limit?

And how hard is the 5000 poly limit?

[Dragofer]

I don't think there's a hard limit anymore, especially if you provide lower-poly versions of the mesh to use at a distance and a shadowmesh so that the engine doesn't need to calculate shadows on the whole mesh. For a character the highest detail mesh probably ought to aim for 7-8k triangles, possibly even more. Polys most likely refers to triangles.

[HMart]

6 hours ago, Frank Cotton said:

It says the poly limit is 5000 polys does that mean 5000 quads are ok or does it actually mean 5000 tris is the limit?

And how hard is the 5000 poly limit?

This engine doesn't support quads at all, all models need to be triangulated at export time.

Don't take my word for it but IMO that limit seems to be a old recommendation for Doom 3, today is very possible to have a character with more polys (triangles), those Dragfer suggested sound reasonable but I would be surprised, if you couldn't have one or two 10 000 or more characters moving around, all depends on the amount of characters you want on view and the minimum PC you want to support.

But with all respect, please make those polys mean something, don't make a high poly model, where many triangles are wasted in flat peace's, I've seen this in some games.

Also it seems this engine, cares more about material complexity than raw polygon count, of course within reason, don't go crazy on polys just because, when many lights cast unto a single model, all at the same time, it is re-rendered per frame, for each light that strikes it, that can potentially mean, millions of triangles being rendered per frame (including all the rest of the scene), please have that in mind.

Btw afaik for animated MD5 models the higher poly it is and the more complex the skeleton is (many bones), slower it will become, I assume because bones will have to transform (move around) more vertexes, so imo is always good to test and if necessary optimize animated md5 models and use only the strictly necessary bones.

And to finish , like Dragfer said, TDM has a level of detail system, so please make LOD models.

make a low poly shadow mesh (use the last LOD level for that?).

Not sure if required for animated md5 models (I think those use a ragdool? "articulated figure") for that, but at lest for static level geometry, making a low poly collision mesh, is necessary (can be the shadow mesh), if not the engine will use the real mesh for collision, very slow.

Not important for animated characters or static models but for dynamic rigid body models, (those moved by the physics engine) unfortunately there's a limit of 16 polys/triangles for the collision model the visual model can be higher poly.

Hope this helps. 

 

[Frank Cotton]

thanks for the info guys, really appreciate it.

[Springheel]

Some of our characters have 5-7k high poly versions, as I recall, though they use LOD.  We have some prop models that are 10K plus, although they also have shadow meshes and LOD.  Today I'm sure we could go higher, although mixing high and low poly assets is always bad for cohesion. 

 

 

[kano]

Actually I have a fitting question for this topic. As the game ages, how hard would it be for the TDM team to spit out versions of the models with more subdivisions to make them look better? I mean, since it's all open source, can't one just take the original model files, ratchet up their subdivision, and then reimplement them into the game as hardware gets more powerful?

 

The other day I spawned a whole bunch of guards on my 5950x machine just to see what would happen to the frame rate, because I remember doing that same thing making a Pentium 4 beg for mercy, and it didn't even flinch.

Edited April 21, 2022 by kano

[HMart]

IMO just Tessellating a low poly model, may help round silhouette's a little but will not magically make them look better, perhaps may even make it look worse (and unnatural) and imo is very wasteful, why, because like I said above, when doing that, many polies will very probably end on flat peace's and still look flat, have no visual impact but still make GPU's work harder.

[Frank Cotton]

12 hours ago, kano said:

original model files

probably not

 

first problem no one is that autistic in their workflow and file management, if you work completely non destructive and scalable you have to do more work than simply making a new model, there probably is no clean file to work from.

second problem some of the older models are designed around a relatively low polycount, simply adding more tris won't make much of a difference.

 

I would be happy do make some higher fidelity models based on the existing ones.

Name a character and i will give it a shot.

 

[HMart]

Hi Frank Cotton, please do what you want, but if you don't mind, I have some suggestions for you to consider, If perchance along modeling, you know how to animate, personally and I mean personally, I would love if someone would make more animations for the werewolf character, specially a better run animation and a cool roar one for when he detects the player.  

Also again IMO, if instead of rework existing ones (that imo are fine for the time being), a totally new character would be cool, for example, like we have the small spider like guard bot, to me would be cool, to have a relative big, biped steampunk robot, with animations, at lest walk and search animations, why, because based on Doom 3, imo is sad to me that the engine has AI AAS support (ability to navigate in a level) for big monsters but TDM has none to take advantage of that.

I can imagine some missions where the player goes to some inventor guild place and has to go around a massive robot to reach some high value thing. 

Something like 

 

where the head is a search spot light that goes around and he makes steam boat sounds when he detects the player.  

 

I know this second one is very probably way harder to make than the one above and TDM already has a automaton character but imo is to much a obvious reskin of the builder guard.  A better automaton would be cool, one that is more or less the representation of a human guard but still very different, something not like but akin to the following, world be cool

Where his eyes are red and cast a spot light

mood example

 

Of course all of this new characters without animations, will be almost useless only used has static props, so again if you can't animate, reworking a existent character is obviously better.

Just my two cents. 

[Frank Cotton]

9 hours ago, HMart said:

the werewolf character

where can I find it?

[HMart]

1 hour ago, Frank Cotton said:

where can I find it?

see in tdm_ai_humanoid_beasts01.pk4   (pk4 is just a renamed zip file, you can open it with 7zip) 

[vozka]

On 4/21/2022 at 8:47 PM, kano said:

Actually I have a fitting question for this topic. As the game ages, how hard would it be for the TDM team to spit out versions of the models with more subdivisions to make them look better? I mean, since it's all open source, can't one just take the original model files, ratchet up their subdivision, and then reimplement them into the game as hardware gets more powerful?

 

The other day I spawned a whole bunch of guards on my 5950x machine just to see what would happen to the frame rate, because I remember doing that same thing making a Pentium 4 beg for mercy, and it didn't even flinch.

There's no simple way to add polygons and make the model look better without a lot of work. 

What might make more sense would be replacing low-res textures. Those imo are more pain to look at and more often used right next to high resolution textures on the same surface or in the same room. After playing a few games (Thief 1 & 2 among others) with neural network upscaled textures, I believe this technique could be used with success. 

It takes a while to tune the processing pipeline so that the textures aren't oversharpened, noisy etc., but the results can be good. It takes a while to select the right textures, process them (often needs hand retouching for seamless texturess for example, and alpha channels are not always supported) and test them in-game, but it's doable.

[Geep]

HMart has large AI on a wishlist. I'd kinda like small AI with associated animations. For humanoids like hobbits, elves, kids, goblins, dwarfs, leprechauns. Animals like cats, dogs, snakes.

Less ambitiously, with existing animations, a few more female heads and/or bodies wouldn't hurt. Female inventor or chubby engineer, for instance.

[Frank Cotton]

11 hours ago, Geep said:

small AI

how small could it be? how does it work with pathfinding and collision etc.

could it run under the tables if the collisions are set up that way?

 

I will put this on my list and will make some little humanoid.

 

[demagogue]

There are actually two pathfinding scripts, one for human-sized AI and one for rats. So rat sized AI are okay.

[Geep]

1 hour ago, demagogue said:

There are actually two pathfinding scripts, one for human-sized AI and one for rats. So rat sized AI are okay.

Hmmm, rat pathfinding might work for small animals like cats, lizards. But for humanoids, I was thinking of 3-4 foot tall characters, so probably would need a separate pathfinding script.

[datiswous]

23 hours ago, demagogue said:

two pathfinding scripts,

What, Only 2?? Is this automatically used based on the aas or something? What would the fire elemental use?

Some in-water creatures like fish would me nice, although maybe md3 would be enough for that.

Edited April 28, 2022 by datiswous

[demagogue]

Right. I was talking off the cuff and should probably have said something like "at least" 2 scripts (edit: I almost edited my last post to add "at least" too; but then I guess I forgot to) or 2 relevant to the discussion at hand. I think there probably are other algorithms--I haven't looked at it in such a very long time--but they didn't occur to me.

I remember the rat one of course because if you spend any time with DR, when you run a map after changing the geometry without dmapping it, you'll I think always get an error that the rat pathfinding is broken. So it's easy to recall