Meaningful Character Customization

[pxib]

The human eye have incredible depth of field. It can ascertain mountains on the horizon as capably as it can distinguish pimples on a forehead. Our hard-wired facial recognition systems allow us to identify familiar acquaintances at a glance and at range. This has led designers to care about character faces and to offer a great many sliders related to head shape. Frequently there are as many variables above the neck as there are below it. This is a needless waste of effort, polys, and character load time. Nobody can tell past a distance of a few meters. Monitors just don't have that many pixels yet.

Not that players don't want to be obvious and identifiable. Offered a choice of hairstyles, players frequently choose the ones that look the most outlandish... or at least different from the rest. Given a slider, they slam it all the way to one end or the other. Primary colors are more important than an exact shade of green. As fun as it is to screw around with subtle changes, once we start playing most of us want to be recognizable. There must be a better way to do that than floating names.

Cartoons are a good place to start. They suffer from the same dept of field problem that computer games do. Any cast of characters will, by necesssity, have wildly different hairstyles, head and body shapes, and color palettes. Even as a angular smudge in the distance, a character's identity is often immediately obvious: Broad shoulders and short legs, tall and narrow and bald, rectangular build with tiny head, big hair and wearing  red and brown.  Equally important is the characters' posture and animation: How do they walk? How do they stand? How do they sit? How do they wave?

So how about fewer options within a trait, but more traits. Wildly different body frames, but only four of them with two options apiece for shoulders, waists, hips, legs, arms, hands, neck, whatever. Four options for height. Four immediately identifiable posture sets with two options apiece for walk, stand, run, jump, sneak, duck, laugh, whatever. Have all clothing be crafter-customizable along similar lines. Each type of clothing is wildly different from any other type, but within that type exist obvious and straightforward options that allow players to refine their "looks". Fewer shades, more colors.

If, like armor, differing body types or postures have different stats and abilities, they become as important as "races". It becomes possible to guess players' classes based upon their appearance. It also becomes possible to guess wrong. That lumbering brute with a goatee, big enough to gain extra hitpoints and the "tackle" ability, may actually be a cleric. Cross-sectional area makes it unlikely he'll dodge anything, unlike his lithe, long-haired companion with her big shiny battleaxe.

Blahblahblah.

[Xilren's Twin]

Not that players don't want to be obvious and identifiable. Offered a choice of hairstyles, players frequently choose the ones that look the most outlandish... or at least different from the rest. Given a slider, they slam it all the way to one end or the other. Primary colors are more important than an exact shade of green. As fun as it is to screw around with subtle changes, once we start playing most of us want to be recognizable. There must be a better way to do that than floating names.

Two thoughts.

Firstly, CoH's character designer is by FAR the best for this sort of thing, BUT, they can get away with that much customization up front b/c there's no loot to speak of in that game.  One thing most players seem to dig is having their characters appearance reflect their success in game (the armor and weapons they weild are both charcter effectiveness multiples, but also epeen status symbols).

Secondly, i like the idea of traits chose at character development having a visual cue in game.  Take a game like Shadowbane and their character creation.  If you chose a trait like "half giant", it should change your physical appearance accordingly.  Let's see, 8 ft tall with rough, pebbly looking skin and poor eyesight.  Half mountain giant, get the boar spears!  In a skill based system, it would nice to have each skill have a phsyical counterpart.  Think of what you could do with a Fallout/Gurps style skill system...

Xilren

[edlavallee]

Firstly, CoH's character designer is by FAR the best for this sort of thing, BUT, they can get away with that much customization up front b/c there's no loot to speak of in that game.  One thing most players seem to dig is having their characters appearance reflect their success in game (the armor and weapons they weild are both charcter effectiveness multiples, but also epeen status symbols).

Maybe I am not like most players, but I don't really care that much for my appearance to reflect my success in game. I often would wear a set of gear for a specific look, rather than sport some better stat gear with lesser visual appeal. This is only in the clothing (and makes me think I am obsessed with some dort of paper doll dress up routine, but lets leave that to another thread), where in weapons, I agree that I want some visual cue that I have beaten BadAzz McMob and taken his cleaver of bloody gore.

So, let me gain my stat bonuses apart from clothing and dress how I want, and let me display my sucess in my selection of weaponry.

[pxib]

Well by all means continue to allow the epeen. There can certainly be objects and enchantments (my weapons glows! this ball orbits my head!) that are only accessible via particular accomplishments. Heck, reward players with special dance emotes that nobody else can do. The real barrier to character traits is the amount of bandwidth required to inform other players' clients about how every other character nearby happens to look. Why not have that information be spent on visual differences that matter rather than ones which few people can see? Limit the number of choices rather than the types... a lot of 2 or 4 bit chunks rather than a few 8 bit ones. Fewer textures, more models.

CoH is fantastic, but it wastes a lot of information (polys and bandwidth) on bits people only see from a few feet away. That's all well and good for chats and costume contests, but 90% of the game is running around and fighting in swarms of partical effects. One "huge" superhero doesn't look all that different from any other in sillhouette. Posture and pose would be far more obvious.

Options about how your character chooses to wear a hat are more important than any variety of hats.

[Sairon]

I don't think equipment have any impact on network performance. The models are already stored on your computer, the only thing that needs to be transmitted is "use weapon model x on player y" for example.

[pxib]

Indeed, but compare load time walking into Ironforge to flying through Atlas Park. Polygons matter to your computer, large groups of  customized characters matter to your bandwidth.

[Yoru]

Indeed, but compare load time walking into Ironforge to flying through Atlas Park. Polygons matter to your computer, large groups of  customized characters matter to your bandwidth.

Um, what? For each left-right morphing slider you use a floating-point number to represent the degree of blending between two meshes and you end up with, at most, 4 bytes of data per slider. A hundred characters with a hundred sliders is going to be about 40kb of data on top of position and orientation, which isn't all that much, especially since you'll load that set of data once (when the character comes into range) and then cache it off.

Anyway, one solution that was ventured by Seed was to have a rendered full-facial portrait appear instead of the little static portrait that you see when clicking on someone in WoW or the EQ-alikes. Doesn't completely solve recognition-at-a-distance, but it's something.

[Krakrok]

large groups of  customized characters matter to your bandwidth.

Ted Stevens is that you?


Kidding aside, a combination of EVE empires, Planetside, and the CoH character generator (game combination of the month) might be interesting. You take the CoH character generator and when you first make up a character you only have the options available for what you're currently wearing. So for example if you have on light armor then you get all of the 'light armor' options in the CoH character generator. Farther into the game if/when you change into heavy armor or whatever then the heavy armor 'look' options become available in the customizer. Players could change their look whenever and however many times they want. As an example, when you get tired of the spiky armor you just change to the plate armor without the cockblocks.

This allows you to mostly keep the character customization from CoH available to the players in a Planetside style environment. Additionally it would allow 'empires' or RPers like the Shadow Orc clan or whatever to all choose a player defined uniform (or color, whatever) if they wanted. Lastly it gives a somewhat structured player customization without the lootwhoringcampingbullshit from Diku.

Keep in mind this would be built on a Planetside system which is effectively itemless.

Call it Mutant Wars. Okay that was taken. Chameleon Wars.

[pxib]

A hundred characters with a hundred sliders is going to be about 40kb of data on top of position and orientation, which isn't all that much, especially since you'll load that set of data once (when the character comes into range) and then cache it off.

Then add fifty characters plus their location, direction, speed, and facing plus whatever they're doing or saying plus the normal client-server chatter and watch your latency. It's not wildly bandwidth expensive, certainly. We're not talking about running a bullying bit-torrent client in the background while you play... but it matters. Wander around crowded places in Second Life for fun and profit. In games where reaction time matters (not just twitch reaction time but World of Warcraft reaction time) an extra 40k every time somebody wanders into your field of view could be a matter of life and death.

Anyway, one solution that was ventured by Seed was to have a rendered full-facial portrait appear instead of the little static portrait that you see when clicking on someone in WoW or the EQ-alikes. Doesn't completely solve recognition-at-a-distance, but it's something.

Sure, but it doesn't help you pick anybody out in a crowd without individually targeting every one of them. I want to see a crowd scene and say "Oh hey, is that Gorillajoy?" and, upon clicking on her or wandering closer discover that I was right. I want to see my character within a crowd scene and feel like a beautiful and unique snowflake even if we're all members of the same guard and are wearing identical uniforms.

[Lantyssa]

I have not been following it too closely, but I believe Spellborn Chronicles is going to have a customizable appearance system.  Your gear may help determine stats, but you choose how you look.

[Yoru]

A hundred characters with a hundred sliders is going to be about 40kb of data on top of position and orientation, which isn't all that much, especially since you'll load that set of data once (when the character comes into range) and then cache it off.

Then add fifty characters plus their location, direction, speed, and facing plus whatever they're doing or saying plus the normal client-server chatter and watch your latency. It's not wildly bandwidth expensive, certainly. We're not talking about running a bullying bit-torrent client in the background while you play... but it matters. Wander around crowded places in Second Life for fun and profit. In games where reaction time matters (not just twitch reaction time but World of Warcraft reaction time) an extra 40k every time somebody wanders into your field of view could be a matter of life and death.

You missed the point. A hundred characters with a hundred sliders is 40kb of data. One character with a hundred sliders is 400 bytes; add in position, facing and velocity, each of which are 3-value vectors (12 bytes each). Second Life is entirely different since you have to stream in entire models and textures when they come into view, as it relies on user-generated content and isn't particularly meant to be Counter-Strike. In a WOW-esque game, where your reaction time matters, all the data-heavy crap (models, high-res textures) is already on your HDD.

Really, what I was trying to say was that your assertion that crowds are more bandwidth-constrained than local-PC constrained is probably not correct - I'd be more worried about doing 100 mesh transformations per character, then blending their textures with custom colorations, when they first load up, and then having to cache all that off into memory; you're more likely to enter a state where you're swapping lots pages off the pagefile on disk when in a large crowd.

Sure, but it doesn't help you pick anybody out in a crowd without individually targeting every one of them. I want to see a crowd scene and say "Oh hey, is that Gorillajoy?" and, upon clicking on her or wandering closer discover that I was right. I want to see my character within a crowd scene and feel like a beautiful and unique snowflake even if we're all members of the same guard and are wearing identical uniforms.

Anyway, I don't disagree about the aim here. You might want to take a look at the way Hero's Journey is doing it; you can generally pick out your outfit and keep it constant while playing - you more or less change out 'enhancements' (COH style) on your gear to improve it, but the look stays mostly the same, as I understand it. Given the variety of models they showed at E3, you may be able to put together an ensemble that's somewhat unique.

Also, when you're looking at a character rendered to about 100 pixels high on your screen, you can only get so much variation, even with different models and sliders. Color's your best friend there. Larger, higher-resolution monitors will help since you'll be able to depict more detail at a distance.

[Krakrok]

Second Life does literally download the objects to you because they are player created like a webpage. Hence it can feel slow because objects pop in to existance. CoH on the other hand has all the graphics client side. And in Yoru's example 40kbs was for 100 people with 100 customization settings each.

There are all kinds of tricks you can play so people don't notice models loading dynamically. For example, loading everything w/ a default look and the morphing them to their customized look once the data comes in.

[pxib]

Obviously I was confused. Consider me educated.

[bhodi]

You eliminate modems when you do that. Your bandwidth ceiling is generally 2.5k/sec upstream/downstream unless you want to exclude them.

[Margalis]

I think a simple point to take away from this is that dev time is often mis-spent. EQ2 is a good example. You can move your eyebrows up and down but your starting items looks like crap and most of the races look the same, just bigger and smaller.

Instead of giving you 20 meaningless options they could have spent their time making the models look more distinct in general.

[Strazos]

You eliminate modems when you do that. Your bandwidth ceiling is generally 2.5k/sec upstream/downstream unless you want to exclude them.

Eh...I think I was able to pull at least 10k when I had dial-up, but I see your point.

But uh...I don't know any gamers who still use dial-up.

[Yoru]

You eliminate modems when you do that. Your bandwidth ceiling is generally 2.5k/sec upstream/downstream unless you want to exclude them.

Are you still using stone knives and bearskins too?

[bhodi]

Are you still using stone knives and bearskins too?

Well, no one has a fixed amount, but CNN is reporting about a 60% broadband peneration in the US. Some reports are a bit higher. I can't find any numbers for just the online gaming segment of the population, but I do know all current (popular) MMOGs work on dialup. It's still in at least 25% of internet-connected households.

Another thing to consider is operating bandwidth costs. Double or tripple (or 10x in your case) bandwidth requirements for each user makes your infrastructure costs skyrocket.

[Margalis]

Character data only has to be transferred occasionally. Position and movement must be transferred continuously, but what hat you are wearing only needs to be transferred once until you switch hats. (Slight simplification but the general idea is sound)

[bhodi]

Character data only has to be transferred occasionally. Position and movement must be transferred continuously, but what hat you are wearing only needs to be transferred once until you switch hats. (Slight simplification but the general idea is sound)

True. I'm sure the "next generation" of MMOGs will have increased custimization, since it's apparent that there's an entire subclass of people that *cares* about it. It's why WoW implemented the "dressing room", and it's why UO and others had dyes.

[Yegolev]

There are all kinds of tricks you can play so people don't notice models loading dynamically. For example, loading everything w/ a default look and the morphing them to their customized look once the data comes in.

This totally sucked in Horizons, but I suppose it would not take much to "do it right".

[Yoru]

Are you still using stone knives and bearskins too?

Well, no one has a fixed amount, but CNN is reporting about a 60% broadband peneration in the US. Some reports are a bit higher. I can't find any numbers for just the online gaming segment of the population, but I do know all current (popular) MMOGs work on dialup. It's still in at least 25% of internet-connected households.

Another thing to consider is operating bandwidth costs. Double or tripple (or 10x in your case) bandwidth requirements for each user makes your infrastructure costs skyrocket.

Indeed. The case I presented was extremely unoptimized as well. You probably don't need a full 64 bits of precision for character customization mesh morphing; there's probably less than 256 visibly distinct interpolations, so you transmit all that stuff as 1-byte values and now you're down to 100 bytes. If you have a large, discrete package of data (like character data!), then you could also use a compression mechanism before it's transmitted, e.g. gzip.

I mostly just wanted to make a snarky remark. :P

[Krakrok]

Bandwidth is dirt cheap. You can get a gigabit sustained for $60k/m as the more you buy the less it costs. All the people thinking that YouTube is spending millions a month on bandwidth are full of crap.

[Trippy]

Bandwidth is dirt cheap. You can get a gigabit sustained for $60k/m as the more you buy the less it costs. All the people thinking that YouTube is spending millions a month on bandwidth are full of crap.

Except that a gigabit connection can only sustain 2731 simultaneous 384 kilobit connections (the approximate bit rate YouTube encodes at). Let's say they average 200K simultaneous connections per second. That would be: 60K * (200K / 2731) = $4,393,994 a month. Of course, as you said bandwidth gets cheaper the more you buy but "millions" is not an unreasonable estimate.

[Jayce]

A bad example of this was SWG.  The idea of morphable faces is a cool gee-whiz factor at first, but it does kill your performance in crowds.

And I really don't think all this figuring of bytes transferred makes a big difference.  It might add to latency, but I think the bigger bottleneck is your graphics card being asked to calculate 100 individual and slightly different meshes on the fly, and additional ones as people show up.  They can't be pre-calculated because there are so many possible combinations - though they can be cached, but that adds to the memory footprint of the game.  You also have to figure out when to flush a cache, for example of the guy's mesh who drove by 10 minutes ago and you never even looked at him.

The fact that it spends SO much relative time doing this for a feature of which you can't even realistically see the results is pretty silly.

[Krakrok]

Their July 16th number was 100 million a day or 138k/s @ 384kbps (52 gigabits a second) based on an average 2 minute video. Their June numbers were 83 million a day or 115k/s (44gbps) . Their February numbers were 15 million a day or only 20.8k/s (8gbps). If they are paying a million a month now it's only been since June or so. I could see them paying $20-$25 per meg sustained @ 10gbps+. I had read $300k/month which would put them at ~$35 a meg at 8gbps.

People were speculating a million a month in December and January though.

---

Back on topic, yes 5 billion face combinations is a little silly to say the least. There is only so much LOD can do to weed out the faces that aren't in your field of view. And while your artists are building new kinds of eyebrows they could be building more 'useful' objects/content.