The Art of Not Reinventing the Wheel in 'Wild Rift' Asset Pipeline

[Music] hi everyone welcome to gdc and today our topic is the art of not reinventing the will in wildlife asset pipeline hi my name is sean low technical art director at wright games so happy to be back again to gdc this year last year i gave a talk about ux of r2 development in general so this year i thought maybe we can talk more about pipeline and specifically on what happened in baltimore and before that quick intro about myself i started working on games when i first joined disney interactive and worked on disney infinity 2 and 3 as a technical hard disk in the tools department

infinity is a massive project where we got to work with studios like disney marvel and lucasfilm in everything relates to the ip and i learned a lot from these projects as you can see there's still a lot of toys that i collected during infinity times those are sweet memories and it's so great to have them working alongside with me throughout these years all right and there were also many amazing prototypes and ideas that i worked on but those belongs to the unpublished titles here i then joined unity technologies serving several several internal demo teams before i got pulled into the adam project it was a very ambitious project because we want to prove that we can use game engine to produce high graphicality film and after the success of adam our team

got to work with disney tv this time we want to prove that we can use game engine for long form animated content like baymax dreams my last project at unity was overhaul the entire unity editors interface which is official starting 2019. another super ambitious task where i got to work closely with many engineering and designer teams it's a long process but i'm glad that we put it off successfully it is now officious and default in the unity you download today so hope you all enjoyed it now i'm with riot games working on ballroom which i wasn't fully aware when i first joined so if this is also your first time hearing about well reef please allow me to briefly explain it here

wall reef is league of legends but on mobile it is a 5v5 moba or multiplayer online better arena genre here you log in the game and form a team of five so does your opponent team everybody got to pick a champion to play then the 10 will get sent to a battlefield or we call it sumner reef battle you work as a team and fight for the objective that's to take down enemies base or just as fun fighting each other very straightforward but because each champions that you choose is so unique with their abilities there's a lot of time you can sort off your 200 iq gameplay and skillful control mechanism and when you outplay your opponent successfully it becomes the highlight but when you don't it goes to the funny moment compilation then

and all these memorable moments is the reason why players are engaged within these games enjoy playing it as well as enjoy watching pro players they meet in the e-sport events i myself love this trailer very much because i think it captures the light tone the fun long time we hope players can enjoy it here in huawei all right so to reiterate worry it is on mobile and we rebuild it entirely the art the code on you this means we got the opportunity to revisit many parts of the games like the design and gameplay to make it refreshing and fun but this also means we got tons of work to do because not only we have to do the new works we also need to ramp up our content to match with what we have in league of legends pc whenever makes sense

and for myself joining this theme while it is scaling or exploded from a handy prototype to global live games presented one huge problems that i need to solve which is how do we keep thousands of art assets each with huge dependency in a trackable state so for today i'd like to share my experience in solving these problems we'll go through the context scopes actual implementations and finally the learnings all right first i'd like to explain about wildly champion assets uh because it's a pretty unique structure to serve this mobile genre for a champion asset it mainly consists of the icon splashart

in-game model and turntable model the icon splash turntable other one player will see before an actual games or battle but when a battle start is all about the in-game asset here that's why it is the most complicated asset we have and also the one that we will spend most time talking about for today's topic and when inside the game each champion has four abilities to learn and level up you can think of this ability as the champion's power for example fire cannon causing aoe damage or heal and for each ability we will have a dedicated set of animations effects and sound to pair with it take leona our support tank for example her first ability is shield of daybreak

so you'll see her animations and the effects i said that we need to play when it got triggered second ability eclipse and the third ability will be jennifer blade and the fourth or her ultimate solar flare you'll see over here when triggered her animations of her pointing the sword up to the skies followed by the declines that we need to project onto the round and that's for a champion as a base to further expand the fantasy of a champion we will have skins that visualize the champions with different thematic or sound for example here we have leona the radiant dawn as a base champion well we have barbecue leona as a skin which we re-imagine her as a chef

now her sword becomes a huge fork and her shield becomes a grill and to do the skin technically we will have to rebind the new bbq leona mesh onto the base leona skeleton so the animation files can be shed there will be also time that we need to add bones for unique animations or effects designed specifically only for a particular scheme so basically when we look at a champion headset hierarchy it will be something like this we have animations and effects as set for each ability for a base champion and for skin we might reuse everything on the base but do a special ability for the ultimate like what we see here for skin 1 and for skin 2

we might reuse more than one base asset or we might even create a whole new set for everything like what happened here for skin free also bear in mind this is only a simplified version we showed here in the actual full asset hierarchy we will have more skins as well as more animations and effects set for different actions that we need in the game therefore if we want to track an asset of a skin in our game it could be really confusing and tricky yes just for a champion currently as we speak we have over hundreds of skins in our game when we zoom out it will be more like this well again this is also a very simple version of the reality uh in the productions

all right moving on expanded scope wall roof started as a small r d project in the beginning our main goal is just to figure out what uh figure how should we what are the things that we need to do uh if we want to put league of legends on mobile uh how do we make it right from there one features after another one milestone after another suddenly we were released and moved to close alpha and open beta at several regions in the world and start to serve millions of players now so in order to live up to our player expectations we have to pump up more content meaning more champions and more skin for example during alpha we only have 40

champions with no skin now we have 68 champions and 189 skins in our recent release and the number is still counting so as we need to produce more content we also need a whole lot more people to help us in doing this and now our teams also got scaled up rapidly but in a relative small ind skills team members were able to manually track and update the current state of a particular asset and keep communications consistent through regular income however with current state of war with size to coordinate the production works with multiple internal and external teams the previous method immediately becomes unsustainable to illustrate the situations clearly this is a very true conversation i have

with team members happen so many times especially especially with those outside of the core death thing so say someone will come and ask me hey uh can i get the remodel i would say yeah of course which one the latest then normally i will ask i mean do you want an in-game or turntable model ah the turntable got it just a base skin right oh actually i need the kda skin guess what okay but which one kda the og or kda or the new one yeah we have two versions of kds inline the original one in 2018 for pc and the new one that we released in 2020 both for pc and wow so confusing so hopefully now you can understand the

pin and why we absolutely need to solve these problems and how things could be worse well endemic bye bye office and hello working from home imagine having to deal with this crazy amount of information while the working environment have changed from everyone working in a closed office environment like a part of 10 people to everybody working from home no longer you can lean over to your teammate beside you less than six feet apart for a quick answer now everything needs to needs to need to go through slap messages video calls like google me or zoom meetings this makes the process of getting a simple yet derivable answer suddenly become so much harder and time consuming so to summarize our problem space

wall reef has a complicated asset hierarchy to deal with our teams got scaled up rapidly more human more questions needing more answers and working from home just make complications harder than how we used to in the end simple yet crucial tasks like tracking a state of our asset is a huge pain and now to the goal what we need is a tool that can quickly help us to get all art asset information above broad and deep brought as in things like how many champions uh we have now or say how many support tanks we have for the current build deep as in what are the list of animations that we use say for bbq leonard ultimate

skills for example then for the restrictions the big caveat of this task is that we can afford to revamp the current workflow because production is full stream ahead we have deliveries to meet we can't afford to ask parties or pm to help us fill up those data through their daily routine or even train them to use some new tools to help us gather information that means reinventing the will or introducing new process at this point is a big no no so to start what do we need to know now and where can we find those answers this is our discovery phase my first approach is to survey existing tech that we have in the project

this is an very important step i need to know what we already have or even what we might have in the near future and then not to make any assumption here so through reading a lot of documents uh and checking in with team leaders from art engineering and even design i was able to capture a broad knowledge of where and how our art assets are being used in the project so one important thing that i learned is that we have an internal game logic system that is used to set up every champion in our game for example which character animation clips and effects are supposed to play during a champion ability or which model is supposed to load when players select a champion skin

the systems were separated into few different places for different usage like we mentioned just now but that's okay i just need to know where to look for them and guess what the best part is these data are serialized in separate json files which means i can pass them directly without any additional truth therefore in this case a very specific and custom made tools in our project helps me in verifying the first and major parts of my hypothesis and i'd like to emphasize that this kind of very niche deeply integrated systems is typically very hard to understand at first documentation might not be available and even if they exist they might be outdated because how fast things have changed to meet the

production's needs so whenever a document is not available i have to check in with relevant teams or even individuals to find the answer and when both document and person is not available what should i do yes we have the code the answer is right inside the code what i will do is deep dive into the code you know set a breakpoint followed each step to prove my guesses and get the answer by myself it is tough but it's definitely rewarding too if you can make it through the steep learning curve of this kind of in-house tools the benefit will definitely avoid the cost at least for me all right next is what i got for free the established asset system within unity

uh for example in this case what helped me are the prefix systems and the meta files and since this is universal i mean same rules apply for all unity projects out there therefore i can safely rely on these generic systems and use it to build my tools around it in a unity prefab file it actually stores a game object complete with all its component property values and even the child games objects in it is like a template and we use it to configure many assets in value including our champion and as for the meta files it is created alongside with each asset file which is used to store the uni a unique id or gui id that unity assigned to it from the beginning these two unity files helps me in developing the solutions

which we will explain it in more details later and good things about this kind of established system is that they are is that their level knowledge is publicly available either from the official documentation or community forum therefore the learning curve of this is much lower than the in-house tools that we discussed before and the last part of discovery is the naming conventions you probably heard this million times already but it is still a very true thing and since and since what i'm trying to solve here is related to asset the naming convention is even more important here and the way i build it is from two categories one is system facing like the folder and file names

the other one is more for asset like how the rig animations and effects object is being named inside a file for example for rig we might have prefix like l r or c to indicate which side of a root control or skeletons belongs to animations might have like in out trends while effects might have things like hit above to explain their usage with a strong and clear naming convention a lot of questions can be answered right away however we must also admit that this is extremely weak because if anyone in the team is not committed to these conventions you know like wrong folder inconsistent names or even like typo it will immediately break and make things worse and confusing

i'm just super grateful that our team did a very good job in this so whenever i have to relying on these naming conventions did help me to solve the other parts of the problems when the previous two methods were not able to all right so now that we already did our homework survey and gather enough information let's get to work for a solution and the solution is an asset analysis tool which is able to scrap the entire game project for relevant champion asset data and in this process we will first ingest all json files that we use to set up our champions through the logic system that we discussed before and these stage 1 files contain a general champion info like their name

their attack range or lane we can also know their skin and skill names along with their 2d and 3d asset and more importantly their path in our project because with this asset path stored in the json file here we can now use it to load the asset for more details information maybe something like the mesh and animation data in a fbx file then based on this raw data the two will then create containers for each asset type with proper structures that reflect our asset hierarchy as we discussed before so we have the top levels which is the entire project right then we break it down into the champion levels then we have uh the skin level with their respective asset

for example again you know like their animations and effects list for each actions that we have again idle run recall things like that pretty standard oop here but with these structures we can treat each container like an actual asset object we should make the next steps so much easier and user friendly yes the next step is an easy api for the user to interface with we really want this tool to serve as an easy data point for user to access so just import the module you know call it aet is our asset analysis tool and point it to the project report and basically you are good to go so for example once we set up we can ask for a very specific champion like ari here after that we can

get uh continue to gather details data like her specific skin say get me the first one or even all her skills in a list form the nice thing about this method is that now data could be obtained in an organized format and at the same time it is easy to customize for many different purposes the possibility is quite endless here and we will look at some examples later on too and the reason uh we can achieve this kind of result is through many integrations you might already notice that we write these two using python so first what we did was integrate packages like unity puzzle and yemo to help passing the unity side of system files so in this case unity puzzles is for us to pass the

prefab so now i can know which components are contained in that prevent for example like the mesh or material with that we can continue to trace the actual asset that is being used uh within that prevail this is where we will use the yaml package to help us in verifying the gui id of the asset that we are looking for so once the asset is verified we can now safely ingest or load the asset for example we have a lot of fbx files that we use for models and animations so what we did was having this tool to integrate with the fbx sdk this enables us to get detailed data inside fbx like the mesh data animation curve and even other custom properties that we store

in there this is so great because now we can query the state of an art asset entirely this even give gives us the ability to qa the art asset if we want to for example we can ask the tool to scan for any model with a word count that sits certain limits so you might ask why python then your project is in unity why don't you stick with unity's c sharp yes that would be the straightforward choice but launching a unity project at this skills it normally takes about 5 to 10 minutes before we can even start navigating inside the project for an answer so what i really wish for is something that is light and fast basically i just hate waiting when i

need an answer also i like to keep these two as standalone so what that means is it doesn't need any tech dependency from our actual projects we don't even need to install unity in order to run these tools for example what i just need is a python ide you know and outside of the project and start coding not to mention iterations times is faster as well since we don't have to wait for unity to compile the script every single time having these extra layers of flexibilities makes our two development a lot more efficient and this is our final result now tech artists have the abilities to track asset dependency easily provide quick and accurate answers to anyone who need it directly and

it can be customized to fit any specific need and to find any specific answer for example it can allow us to do a project-wide asset analysis like what's shown here we can get all champions available in the current project look through the list and ask for maybe just a base model in this case we just have to throw in the id of zero and that will give us access to every base model fbx files immediately from here we can do many analysis like hey just calculate the max or mean uh word count in the mesh or get me the keyframes data in animations uh clip so one real scenarios i mean share here uh is in a zoom meeting that we had someone asked the group hey which champions has the highest poly count in our project now

so what i did was quickly launch visual code and write a few lines similar to what we saw just now get our champion right go through all skins and for each skin grab the in-game model word count append it in a dictionary sort it and tell me the max and all this was done while the group is still talking and before the topic changed i was able to inform the group and say seraphine has the highest polycount i guess thanks to her long hair and floating stage imagine if i were to do this directly in our unity project first i'll have to launch unity five minutes in say even if i have some tools ready to scan or champion fbx but i'll still have to modify the c sharp script to get me the max word count model and wait for it to compound

now compiling even and a simple editor script like this in a project of this skill is going to take maybe like a few minutes therefore having this standalone tool is definitely easy to modify and faster to get a answer so with the data that we gather from these two we can now output it in whichever format we want like a spreadsheet a csv or excel for other teams to use since not everyone in our team is able to write a python stream but that's okay they can still benefit from these two with the data provided in the format that they are familiar with to continue their work not only that we can even directly connect this data or these two to any framework or applications that we need like perforce maya or other standalone tools that we have

for example we actually use this tool to populate our shotgun asset database when we started using it early this year it was a breeze to see how easy and accurate the asset data can be migrated over the new venue all in all these two help us in getting the single source of truth for internal use as well as help us in team collaborations uh with because now uh we can share info and tech easily with other teams like console and pc or basically any team in riot or even even with our partners looking back what are the lessons that i've learned so in order not to reinvent the will the first thing i learned is to understand your playground and here's the pictures of a playground

that i took during pandemi just in case some of us already forget what the playground looks like okay now the playground i'm referring here is the existing data and tools there will be tons of data linked around in a massive project like this some data is ready to be used like in my case the json files are already there to be ingested there are also data that are not existed so we'll need to find a way to formulate that and fill in the blank so in my case naming conventions and those tribal knowledge helped me to bridge the missing gap also understanding what tools are at your expense is equally important some tools are custom made like the game logics that we discussed before and some tools are generic just like the unity system file

once we have gathered this kind of information it will not only provide a broader view of your playground but more importantly it will reduce the urge of build something from scratch yes we prefer to build something on existing foundations rather than start from zero not an inventions but accent integrations the pro of doing this is definitely faster development since the amount of work and code that we need to do is normally broad but thin we are linking many working parts together and drive from there the con of this approach is that you might have to upfront a lot of communication cost to fully understand the overall pictures like what we discussed before this really depends on how easy you can obtain the information from assisting documentation or code base

but even in the worst case scenario i would still believe that the cost is absolutely worth it when we look at the benefit we gain in long term second thing that i learned is choosing the right tool and construct a plan that is future proving we want to make sure the tag that we choose can be easily supported in long term rather than something that might not be even relevant in the next five years we also want to make sure something that is flexible uh to make necessary adjustments if we need to it would be nice uh if it is easy for us to pivot to accommodate any future production scope along these lines we are looking for something that has very little dependency so we can move things around without too much restrictions travel life is what came in mind when i'm thinking about this again i think i really need to get out from

this house because i got stuck in here during pandemi for more than a year now okay so for our case here this is so true uh that when we choose to use python to write this analysis first reason is because they are already many python users within riot which mean is is a relatively low barrier in terms of the learning curve but this also means that it is easier for us to get help uh in working on these tools whenever we need to just have to call someone from other things and we can learn them so we have sufficient support in terms of both tech and people and talking about support python also has a lot of publicly available resources and an active community this makes the development a lot more easier when we have this level of

confidence and at last even before working on the actual code base i know that it is best for us to make these two with less dependency in terms of tech meaning we want it to be more of a plug and play kind of easy mechanism also with this loosely attached interface we are able to change or expand so these two can connect to other applications or engine so we need to pivot in the future last but also the most important lessons at least for me is to start small so we could start fast and possibly also feel fast keeping your goal too ambitious sometimes make things really sluggish to even kickstart by sculpting things accordingly you can at least focus on a well-defined area to start with prototype something really fast so your

result to team members and stakeholders make sure everything is aligned and informed then go back read the rate and keep building the next block with this approach i was able to keep expanding the features while fixing the box or even remove some features that we thought was a must in the beginning another thing that i benefit from this approach is it keeps reminding me to make sure the code plays that way right is always modularized smaller lego blocks that are put together to form bigger blocks then fit into other moving parts we can reuse and even it is easier for us to take out the blocks if we need to uh in the future and talking about uh pivoting we are currently in the phase of evolving this analysis tool into something bigger and even more integrated

remember we talked about using this tool to build the foundations of our shotgun database so now we want to use it to extend our shotgun workflow and this will let us build a more streamlined workflow by layering this analysis tool on top of the shotgun database give it a tour of knowledge and of our asset conventions and rules integrate many processors and automations via the shotgun api i'm confident that while rif art production is going to have even more potentials after this so to summarize the art of not reinventing the will for me it is all about integration you might heard me saying this repeatedly already throughout today's presentations yes the way i view this integration is like a partnership and partnership is more important than

ownership i like to think of our code base like having many people working and collaborating together each person has a dedicated task to complete but it also needs to work well with other people and this is so true for our wildlife team we work with many partners around the world and only with this kind of strong relationships we are able to achieve such a wild project and bring the best mobile mobile experience to our players and as for 10 hour people this partnership genetics lives deeply in our blood we are the bridge and our job is to bring the tech and the art together in order to make things better and also because working together is more valuable and more fun than working in our own silo especially when we are all working from home for more than a year now

i just hope to take this opportunity to encourage everybody to actively reach out to each other even in the virtual format just like this one we are all here in gdc virtually these temporarily physical separations won't be the reason we stop working together to solve difficult problems i want to believe this and we are not found when the pandemic is over we will be able to hang out together again and come to gdc together in person this time lastly and truly one plus one is more than two we are better when together so hope this presentation helps you all either in your works pipeline game or even working relationship with your team members and partners people around you let's make more awesome more fun and more well-gained

thank you all stay safe and see you around [Music] you