What is Agency?
Just in case you have never heard of Counter Strike: Global Offensive, it's a hugely popular online FPS, successor to Counter Strike: Source and the original Counter Strike. The original came out in 1999 and the core gameplay has remained almost unchanged. Players are split into two teams and challenge each other in various game modes such as Bomb Defusal (one team has to plant and detonate the bomb while the other tries to stop them) and Hostage Rescue (one team must rescue the hostages whilst the other attempts to prevent that). The Bomb Defusal mode is by far the most popular, with maps designed with such detail that players can predict down to the second when another player is due to arrive in a certain area of the level. It's also the only mode played in competitive events and for huge prize money.
This leaves the poor Hostage Rescue mode sitting on the sidelines twiddling it's thumbs and feeling a little rejected. In part this is because the Hostage Rescue mode is far more of a roleplaying experience, often with very poor odds of success for the team tasked with doing the rescuing. Often the levels are designed in such a way that the defending team has a large positional advantage, where simply staying-put will give them a good chance of winning.
That's where we can start talking about Agency. Agency is a Hostage Rescue level, created as a collaboration between level designer Patrick Murphy, and myself doing the art. The basic idea being that Hostage Rescue could be just as precise and exciting as Bomb Defusal. It's been included in three official releases from the games creator, Valve, as part of their community level packs: Operation Bravo, Operation Phoenix and Operation Bloodhound. Phoenix being a community-voted choice, which was especially great to see that players enjoyed the style of gameplay and visuals that Agency brought with it.
In this article I will go over the process of creating the art, from props to set dressing, texture creation and lighting, while maintaining a visually pleasing aesthetic and serving to enhance the gameplay. This isn't a postmortem but rather a walk-through of the various stages, hopefully to give some ideas to others, with lessons learned both positive and negative.
Iteration from Whitebox to Final
Starting out you should always have an idea of what you're going to create, even if it is quite vague, as it'll point you in the right direction for both creating architectural spaces and letting your imagination fill in the blanks as you build the basic shapes of the level. We knew we were going to build an office space, but style was leaning towards an older government building with red bricks and musty wood. As I started to put in some basic textures we decided it felt too bland, and similar to other levels in the game. In order to stand out and create something really interesting and intriguing that would entice players to want to explore the level we decided to modernize the space and use white as the primary colour - this would help players see each other more easily and provide a striking visual setting it apart from other levels.
"Modern Office" is not exactly a style that has a single look, if you search for images you'll get back a lot of contrasting designs and ideas, trying to put every single one of those into a level would create a visual mess with no consistency. It's important to choose the right references for what you are building, something that looks cool in a single image or from a specific location might not fit into the theme of the level, and in a worst-case-scenario it might actually start to detract from the level as a whole. Trying to cram in as much content as possible simply makes your level feel less unified and jarring.
Unfortunately when you are presented with so many fantastic designs and ideas it can be hard to pick out what is important. After settling on the location: a modern advertising agency's office, I broke down the needs of the level into a few different categories:
- Area Specific
- General Use
- Overall Theme
The Area Specific content is "hero assets" for each location in the level. These are the things that help the player tell different areas apart from each other, a reception desk, a kitchen, a bathroom, etc. Assets that won't be used anywhere else except in their specific location.
The General Use content is the backbone of the building, it's wall sockets, ventilation tubes, sprinklers, desks and chairs. The things that could be used anywhere and would blend in to the background and not stand out unless you were specifically looking for them.
The Overall Theme content is what sells the theme of the level to players, advertising boards, company logos, large art installations and so on. These can be used everywhere but sparingly and should only be used as a subtle reminder to the player of where they are thematically. They shouldn't detract from the Area Specific content but should stand out more than the General Use content. This came in the form of abstract paintings, corporate logos, rotating advertisement panels and so on - things that would subtly tie the level together.
Once these categories were laid out, searching through reference images became much simpler as you know what you need and only have to find an interesting design or detail that enhances a specific category.
This isn't to say that everything was completely planned out or that development was flawless. Sticking to a plan only works until you open the editor, and if you try to force something you'll end up frustrated when it consistently fails to work. As an example we originally had the level set on the ground floor of a tall skyscraper. I spent a few weeks working on content for the ground but never really getting it to feel right within the theme of the level: the contrast between a dirty exterior street section and a spotless interior didn't feel right for the level, and felt a little too similar to another Counter Strike level. Patrick played around with some ideas and tried something I was afraid of: simply deleting everything I had done on the outside and adding an epic city vista. Instantly it felt right. The important thing to take away from this is that just because you have worked on something doesn't mean it's the right thing to be working on, and that getting input from other people with different ideas can vastly improve what you are working on.
Another incredibly important thing I realised is making use of modular assets. If you are going to duplicate something in your particular modelling software you should ask yourself: is this efficient? Chances are you're just making things harder to change later and locking yourself into a particular shape; eg: a walkway has a railing around it, you model the entire railing as a single object. Now if you need to change that walkway a month later you're going to have to go back and change your railing model. It's better to create a smaller tiling mesh that can be used multiple times, as often you'll find you can use that model in other areas and in different ways than you had initially intended. You're simply applying the concept of tiling textures to models, and in the process saving yourself a lot of time.
A Believable Clean Art Style
Creating a clean environment can often be more difficult and time consuming than a very dirty and cluttered one, simply because any mistakes are magnified by the lack of other objects to disguise them. A room with a single chair in the middle is going to end up with the focus being on that chair, if you fill that room with a hundred chairs you're going to be less concerned with the details of the chair and more worried about why someone would fill a room with a hundred chairs.
In the modern office setting of Agency it would have made little sense to fill it with props and clutter, but a large empty space would just feel unfinished. A delicate balance of larger architectural shapes and smaller objects was needed. I like to think of this as functional art: it serves a purpose in the lore of the game world. Window and door frames, electrical sockets, thermostats and card swipes along with the maintenance apparatus of ventilation systems. These are the general use objects mentioned earlier, they fill out space and prevent an empty wall or ceiling from actually looking empty and at the same time they contribute to the believability of the level. It's important to think of the infrastructure of the building when placing these assets - if a wall has an air vent on it then the wall needs to be thick enough to support the ventilation pipes that feed it, Card swiping mechanisms need to be placed near doors at the correct height, electrical sockets should be placed logically in areas where they would be of use to the fictional inhabitants of the level and so on.
One of the most important things to do right when creating clean environments is to get the most out of the materials. It's not possible to cover every surface in dirt or decals, so the surfaces themselves become your way of showing detail.
For Agency this was achieved by making liberal use of the phong shading techniques in the Source engine for models, and cubemaps for world textures. Almost all models in the level have some amount of phong shading, and although it doesn't produce a completely physically accurate result it can be used to create materials and surfaces that look relatively accurate. Simply by increasing or decreasing the intensity of the phong amount allowed for a vast majority of the levels surfaces to be rendered accurately. As I didn't need to have a lot of noisy detail in the materials due to the clean style I simply used a small phong texture as a mask for 75% of the models and let the lighting and general shapes of the models do the rest of the work.
As most of the surfaces had a single layer of material, ie paint or coloured metal, the phong shading could be completely even without breaking the illusion; however some of the dirtier surfaces such ventilation tubes and water pipes had several layers: a painted metal surface with area peeled away to reveal with metal underneath or a layer of dust. These had specific masks that would enhance the different materials, and showing wear and tear in the background assets added an extra layer of depth without compromising the clean style.
Most of these textures were created with dDo, an excellent tool for quickly creating textures. I generally started with quite a dirty texture preset and toned down the details and noise until they were barely perceptible surface imperfections.
Agency features probably close to 95% custom art, and that's a lot of work for a single person. Using dDo allowed me to make a lot of content relatively quickly, and kept it all visually consistent.
The process of creating the assets with dDo was quite simple: first I modeled the basic ingame asset, then did a very quick and dirty placement of edge loops that allowed me to smooth the mesh and get a workable high poly. A very rough normal map was baked (along with a more solid ambient occlusion map), this rough normal map would never make it into the game, it was used purely for texturing with dDo. This rough-and-dirty technique was mostly used on the more general purpose assets that nobody would spend a lot of time looking at. For the objects that were in high traffic areas or that required finer detail a more robust normal map was created.
Tiling textures used throughout the world were photo-sourced and tiled in Photoshop. A few examples worth pointing out are the plaster wall textures and the marble floors:
The image above shows the ingame result, the diffuse texture, and the normal map of the standard plaster that is used throughout the level. The normal map was authored at 1024x1024 compared to the diffuse texture which was 512x512. I created several colour variations of the diffuse texture and for a very plain surface using a 1024x1024 diffuse didn't make much sense. The final touch was to add a subtle cubemap effect to bring out the normal map and add interesting coloured reflections in various areas.
Another example is a marble floor used throughout the level. The normal map is unrealistic in that it portrays an uneven bumpy surface when in fact it is more likely to be uniformly flat. However to break up the reflections and add some visual interest to such a large and empty area I added a subtle bumpy normal map which warps the reflections, but is subtle enough that it doesn't get picked up by the lighting and actually appear like a lumpy mess.
Good shading only gets you part of the way there, however. A poorly scaled model can break immersion instantly, especially when you are trying to create a believable real-world environment. There are tried-and-true metrics for Counter Strike so having a base to work from helped immensely, but these only give you a good starting point or a bounding box for your object. It's important to study real world reference and make sure your object is proportional to the world around it and also to itself. A unit in Hammer is an inch, so having wood that's 2 units thick, or a doorway that is 1.5m wide quickly makes things look wrong.
Working with Designer Blockouts, and not Destroying Gameplay
Agency was a collaboration, with Patrick doing the design work and me doing the visuals, this meant there was a lot of potential for overlap and working on the same areas, the potential for breaking things was huge.
Often when you create things as an individual you don't have to worry about version control or stepping on someone else's toes, however when you work with other people either for pleasure or business you, as an artist, need to change your mindset. You are not creating a portfolio piece but rather something functional that has to withstand hundreds of hours of real people playing it.
Your first role is to support the designer, and this benefits you as well. By creating the basic structures of the level: doorways, window frames, stairs, railings, cover objects etc, you are allowing them to work with the final assets and tweak gameplay according to those assets. Nothing needs to be finalized instantly, it's better to provide a rough mockup of the intended asset so the designer can play around with it and give feedback on the shape, size and silhouette. Once you are both confident it's going to work they can populate the level with these assets which saves you time in the long run, and once you finalize the model and textures they are going to be updated across the entire level without having to manually replace assets.
It can be difficult to determine exactly when you should start an art pass, especially when a level is constantly evolving. Rather than sitting idly by whilst Patrick was ironing out the design of the level I started on the creation of a few visual test levels to explore materials, lighting and modular assets. Once the first iterations of Agency were created, with rough shapes for important cover and controlling lines-of-sight. I went in and created an art pass and altered many of these original gameplay ideas, simply experimenting with different shapes and designs for the rooms. We had a constant dialogue and never considered something finalized just because it was finished. Playtests would determine whether an idea was valid or not in a way that speculation can only hope for. The most important lesson learned during this process of constant iteration was that work is very rarely wasted, and it is far more important to stay true to a gameplay ideal than to have an area that looks interesting in a screenshot but utterly fails when players get their hands on it. A box is a box is a box, it is down to you as an artist to imagine how that box can be interpreted within the context of the environment.
An important part of any environment is the lighting. Too contrasted and moody and it becomes hard to identify players, too bright and monotone and it becomes boring and a strain on the eyes. For Agency I used a series of instanced lighting setups: a model to visualise the light source, a spot light to direct the light, and a sprite or light cone to add a visual effect around the light. Each light setup was unique to the type of model used for the actual light source, ie: all spotlights were identical, all fluorescent lights were identical etc. This meant I could change a single light and have the others update automatically, and always get an accurate result.
Then it was just a case of placing these different types of lights where they logically made sense in the environment, and if an area was too dark an appropriate light source was added, and if an area was too bright lights could be moved around or removed entirely. This made it quite easy to light as everything was guided by reality, which has plenty of reference material, and had the side effect of helping to make the environment more believable. By using various colours on the floor and walls I could direct lights towards them and take advantage of the Source engine's excellent radiosity and spread interesting colours to nearby surfaces.
In many areas the ceiling was opened up to reveal the sky and to let natural sunlight into the interior spaces, this was done to provide contrast to the electrical lights and to get extra radiosity bounces into the environment. Some areas had lights removed or toned down to allow other more important gameplay areas to stand out, for example the image below shows how the corridor here was darkened both by using darker textures and by using restrained lighting to make the room in the distance appear brighter as this is an area that enemy players will appear from.
This could have been taken even further by possibly using emergency exit signs to add hints of colour to important gameplay areas and chokepoints. A consistent lighting language would have helped guide players during the first few times playing the level. There are some large open spaces that would have benefited from some coloured screens or lighting panels, or possibly making some of the larger glass surfaces tinted, to add a little extra colour and prevent such a monotone look whilst not being over-bearing or detracting from the realistic style of lighting I was aiming for.
During the course of developing Agency I had a chance to learn a few things and come out the other end a, hopefully, better artist.
So, what went well?
The iteration process never had any hiccups, by using modular content and being prepared to discard ideas and art styles that weren't working we ended up with a better level. If we had tried to force the original idea of a ground-level government office we would have ended up with a completely different level, complete with underground parking lots and elevator shafts. Exciting stuff!
The power of iteration cannot be understated, and understanding that a mockup or a blockout of a level is simply a temporary phase that doesn't represent the end result. Areas changed drastically between versions, sometimes due to design requirements, and sometimes of shifts in art style; but each version was better than the last, more refined and polished.
What went less well?
In direct contrast to the statement above, sometimes the iteration interfered with more important tasks. I got stuck on areas trying to get them to work instead of letting them sit for a while and returning to them later. I tried to force an idea for the exterior part of the level and it never felt right and consumed way too much time, when all it took was getting some outside perspective. Luckily during the process I learnt to trust designers when it comes to art, just because they might not build high poly meshes doesn't mean they aren't artistic.
Another problem was building too much content completely unique for an area which meant when we inevitably changed things it became time consuming to shift assets around, and makes it less easy for others to re-use that content without creating an almost replica of the area it was designed for. These unique assets helped sell the realism of the level but made them harder to work with.
Hopefully this has been interesting and insightful!