The 18-month fence hop, the six-day chair, and why video games are so hard to make

Early in the supernatural teen-drama game Oxenfree, there’s a scene where the main characters hop a fence. In the game’s roughly four and a half hour playthrough, the scene lasts all of 40 seconds. But according to Sean Krankel, co-founder of Night School Studios, it took the team the game’s entire development — 18 months — to get it right. It was Oxenfree’s most prevalent, annoying bug, he says.

The studio’s latest challenge? Getting characters’ butts to line up in chairs in the upcoming Afterparty, a game about two friends who die, go to Hell, and must outdrink the denizens of the underworld to save their souls.

“[It’s] a crazy nightmare when you think about differing sizes between humans and demons,” Krankel says. “Something as simple as, yeah, literally getting a butt to match up with a stool is incredibly difficult,” he adds, laughing.

Video games are notoriously hard to make. The largest AAA games in 2019 require hundreds, if not thousands, of people at multiple studios around the world just to get out the door. They take years of development. Beyond the stories, gameplay, and the obvious features that make their way into previews and store descriptions, though, there are the minute details that have to be handcrafted and placed, sometimes hidden from the player, to make the game feel believable and immersive — like a quick animation of characters jumping a fence.

Whether or not a player notices, appreciates, or is able to see these details, everything from a pen on a desk to a chair in a room has to be meticulously made, scrutinized, and tested. But at what cost? How does a developer decide how much time to allocate to set dressing a small room versus a game’s main character? How many polygons should an asset in the corner of a players eye get versus something directly in their face?

We recently talked to a handful of developers, making games of varying sizes and scopes, about the decisions behind some of the small details in their games.

It’s never as easy as it looks.

The chair problem

In 2001, a team of 24 at Remedy Entertainment shipped the studio’s first tentpole title, Max Payne. In 2019, Remedy executive producer Tuukka Taipalvesi points out that entire team is smaller than the current studio’s animation department. Each subsequent console generation, he says, brings with it higher expectations from players in fidelity and visuals, which increases the amount of work a developer needs to put into “basically any materials that the game plays with.”

“Everything. Exactly everything” takes more time to develop than people would expect, Taipalvesi says.

When working on Max Payne, making something as simple as a chair might’ve only taken a developer a couple hours of work to get into the game. These days, on Remedy’s upcoming narrative-driven action game Control, even something that minute requires a lot more time — and people — to get up and running.

“Previously you had just a polygonal model of the chair and then a texture map mapped to that and that’s your chair and that’s it,” Taipalvesi says. “These days, it depends [on] where the chair is used. If it’s a gameplay object, it has collision boundaries; it has breakable physics in it; it has collisions; it has three to four texture material maps, plus then details, as well as then the polygonal models.”

“[The time required to develop] depends on the chair — if it’s a cushy chair, if it’s made of leather, if it needs to deform, then four to six days is definitely an option,” he says. “If it’s a metallic chair in a cinematic that the player never interacts with and it doesn’t need physics or collisions or anything else, you’ll be done in a day or less. So it depends on the use case of the prop as well quite a bit.”

When many of the smallest details in a game take days to develop, scheduling is paramount. But it’s not an exact science, developers interviewed for this piece say; it’s mostly ballpark figures that get ironed out throughout a game’s entire production. “I would like to meet the producer who can actually plan all this out during pre-production and give you a mile-long spreadsheet that’s like, ‘This is the asset that we’re going to build, and these are all the details that we’re going to use,’” Taipalvesi says.

“It’s a push-and-pull type of a situation,” Taipalvesi says. “We have white box environments and we then start to see, ‘OK, so the level is this big. We have this [budget for assets]. So we can put three chairs and four lamps into this room. And now we basically need to know how long one of them needs to take.’”

But not every asset is built equally. Sometimes, even in the best-looking games, a developer can get away with tricking the player’s eyes into thinking something looks a lot better than it would otherwise.

Illusions of depth

In the room with the chair, each asset also has a different level of fidelity. The main character in the room, for example, will need a lot of polygons to look good to the player — especially in a narrative-focused game like Control. A developer may be able to get away with making a pen on a desk with a far lower level of fidelity, then, because the player will rarely see it up close. Doing so helps stabilize the game’s performance, allowing memory to be used on more important assets.

“If you build a pen that is down to microscopic-level of detail accurate, and you don’t have enough technology within your title to support streaming in that detail, that detail is going to sync with memory; it’s going to eat it up from something else that might be much more crucial,” Taipalvesi says. “So we have source assets that are in multiple resolutions of the stuff that actually ends [up] in the game because we have in our export pipeline automated steps that will automatically then reduce the fidelity of an object if it’s deemed to be unnecessary. Then of course we also have run-time streaming and levels of details per object that change based on how far the player or the player camera is from that object.”

It’s a lesson Night School learned the hard way on Afterparty. Originally, the game was planned as a fully 3D adventure — until the studio’s first pass on the game looked, as Krankel puts it with a laugh, like a “bad World of Warcraft mod.” So it became 2.5D.

“The further we went, the more we were realizing the approach to building it all in 3D was losing the flavor and character. And it would have required a team of 100 people to ever get it done,” Krankel says.

The studio’s solution was creating the illusion of depth. Objects in the game’s foreground — such as tables with demons passed out on them, per Krankel’s example — receive quite a bit of 3D detail. On the contrary, the game’s background, something the player will never see up close, is comparable to a painting — largely 2D plains with normal mapping on it, a technique used to fake the lighting or texture of an object without using additional polygons. Between the two is a little bit of both.

“We always try to set the types of polygonal count according to the distance of our camera,” Ruel Pascual, Night School’s art director, says. “If you’re only going to see it from this camera view, we don’t really need to invest that much time in the amount of polygons we apply to it.”

“And that comes down a lot to the design needs. It’s not just the poly counts, but it could be, ‘OK, if there [are] 20 characters in this bar on screen’ — originally maybe we had them all have complex behaviors,” Krankel adds, saying originally Night School gave Afterparty NPCs behaviors like walking around the location and having their own drunk meters.

“And then we were like, ‘Now let’s try another version, where most of this is just done by hand and we’re not doing this crazy AI and playing all these extra animations,’” Krankel says. “When you look at both, you’d never be able to tell the difference, yet it was a performance hit to have these more complex behaviors.”

“But it’s really like — it’s like making a dish or something,” he adds. “You’re kind of taste testing it as you go. You don’t start and go, like, ‘This is the exact poly count.’ I mean, [Pascual] and his team would set some parameters for that, but they’re always a best guess, because we don’t really know until it’s all in.”

That illusion, of course, scales all the way down to the game’s smallest assets and details. In a game about drinking, the drinks and glassware are pretty important. But in Afterparty, due to its artstyle, they appear small on screen; a player isn’t going to see them up close and personal.

“Early on, we designed some glassware, but then we were having trouble seeing what the glass looked like because everything is so transparent,” Pascual says. “We needed [to] up the poly count on it to even be able to see the type of material, or the type of rendering or shading we had.”

“Yeah,” Krankel says, “it’s one of those things where we started and you spend all this time having, like, a fluid simulation in a goblet that’s flying around, and you’re like, ‘This looks so badass’ totally out of context. And then you look at it in the game, you’re like, ‘A, I don’t see any of this, B, our performance is taking a giant hit. What’s a better, more effective way to do it?’”

Night School ended up using a shortcut, as Pascual puts it, simplifying drink assets to just different colored stickers they slapped onto simpler glass assets. In something like a pitcher of beer in Afterparty, it’s just some gold liquid that may slosh around slightly. The team doesn’t bother with adding details like foam; it’s too small and the player won’t notice it.

“If you were to zoom in [on] our drink assets that are in people’s hands, they’re very simple shapes that have just a flat image inside of them with particle effects so that it looks and reads like a kind of cartoony, cool drink,” Krankel says. “A player’s eye would never see that, but if they saw it up close it would look pretty lame.”

The solutions and shortcuts used by Night School and Remedy are integral in getting a game out the door, optimizing its performance and keeping developers focused on larger, more important assets and aspects of their titles. But sometimes, even with these solutions, one team — even a team of hundreds, as is the case with Remedy — still isn’t enough.

The outsourcing solution

Remedy didn’t make any of the cars in its 2010 game Alan Wake. It outsourced all of them.

“A lot of the props that we do in here are actually destined for external art outsourcing because it is so easy for realism, or reality-based games, to take a photo of a chair, make up the technical requirements, and give it [to] someone else outside of the studio to work on so that we keep the experts actually working on the interesting problems internally,” Taipalvesi says.

Relying on outsourcing partners has become a more and more common element of the development process over the past 20 years. As games get increasingly complex, developers continue to employ external studios to handle some of the heavy lifting for them — especially when it comes to smaller assets that can eat up the time a developer could spend on bigger issues.

Such is the case for Remedy, which decides what assets get developed internally or externally based on their importance, frequently outsourcing work to companies such as Virtuous or Dhruva. “We try to do a lot of the hero props by ourselves,” Taipalvesi says. “So if there [are] multiple uses for something, we try to do it internally. But if it is one-use or a cinematic prop or something obviously that’s set dressing, definitely we do that outside.”

“Basically, the artists in the team can decide what they want to work on; they can cherry pick assets that they want to do,” he says. “Most of the people that we actually have in here are more larger-scale architects, so they build the levels, the buildings; the big lines basically, the environments themselves. And then a lot of the extra stuff that we use to liven up the scenes comes actually from outsourcing.”

Similarly, Night School’s relied on outside parties to get Afterparty up and running — though not always for smaller, minute assets. As Krankel tells it, the studio’s brought contractors on-site to help with design on the game’s script — which he calls “massive.” Night School’s been able to keep the game’s art all in-house, he says, thanks to the aforementioned shift in visual direction. “I think that’s really maintained a high level of quality control,” Krankel says.

For developers without the resources Remedy and Night School have, employing outsourcing studios or freelance help may not be a viable option — but that doesn’t mean there aren’t outside parties they can use for help.

David Whele, the sole force behind the 2018 indie game The First Tree, an exploration-driven adventure game about a fox finding its lost cub, had to fit in working on his game when he could. At the time, he was working full time at the virtual reality experience The Void. He’d also just had a kid. His time was, suffice it to say, pretty limited. “I basically had to tell myself, ‘I either can play games and watch shows after I get home from work, or I can work on my game.’ And that’s what I had to do,” Whele, who’s now, thanks to the success of The First Tree, a full time independent developer, says. “Every night after our baby would go to sleep, I would work for about two hours [...] and on the weekends maybe two to three hours. And even though it stopped being fun a long time ago, it needed to be that discipline that got me to the finish line.”

Getting to that finish line, Wehle says, was only possible thanks to the use of stores like the Unity Store, which makes pre-made animations, assets, textures, and environments available for public use. Without them, he says, his game would’ve never come out. “I’d use those [publicly available resources] as like a springboard to save time,” Wehle says. “And that included some fox animations, some environments. I tried to add my own creative touch to those and remodify them and remake them so that it wasn’t an asset flip, but even with all that help and the thousands of hours it saved me it still takes so much work to do the smallest things like a fox jump.”

“I don’t know if the world realizes how every little piece of a game has to come from a person. If you’re a solo developer or a small indie team, that can be so overwhelming,” he says.

Getting the fox’s jump animation right proved to be a headache for Wehle, despite it being something a player wouldn’t think twice about if it worked right. Wehle says he was able to use a third-party tool within Unity to rig up his jump animation and refine it; he didn’t have to reinvent the wheel, designing and animating the jump from scratch. But, on the flip side, that’s partially what created his headaches, he says. “I tried to modify the jump and it made it way harder than if I had done it from scratch — if that makes sense. So that’s, like, a downside to it. Right? You’re kind of limited by other people’s creations.”

Outsourcing and third-party asset stores have helped smooth the development process in a lot of ways, giving studios and developers the option to hand off smaller-scale tasks to companies specializing in such things. But video games are continuously getting harder — and more expensive — to make. With the next generation of consoles on the horizon, developers are beginning to look to the future, assessing how they’ll tackle the same problems, but more complex versions of them.

The future issue(s)

Ubisoft may have some of the most expensive trash in the world — $3,438 USD, to be exact — if estimates by Andrew Price are to be believed.

Price is the CEO and founder of Poliigon, a third-party service offering high resolution textures, models, and assets for creators to use in everything from games, to movies, to architecture. Scroll through Poliigon’s website, and users can find everything from pre-made blenders, to power plugs, to carpet patterns.

Much like Wehle’s use of the Unity store, Poliigon offers developers photorealistic assets they can plug into their games without having to make something small or minute from scratch. The company makes it cheap too, offering a $20 monthly subscription for its entire library.

Price’s company is making development quicker for companies, removing their need to allocate time to designing the litter that covers their worlds — and he’s got some ideas on how developers could continue to streamline their process going forward.

Last year, Price gave a talk at the 2018 Blender Conference. In it, he estimated that one screen in Ubisoft’s 2015 game The Division cost around $200,000 to make — with assets like hero characters taking up the bulk of the cost at around $49,000, and small assets, such as trash, costing just under $3,500. His estimates are based on the average salary of game developers combined with how long it typically takes for a AAA developer to create the asset in question. Assets, Price says, are “unreasonably expensive.”

“Basically it’s all time,” he tells Polygon. “It’s all to do with time.”

Price uses an example of developing and designing an apartment building in a typical AAA game. First, a developer will need to create the broad strokes for the building — parameters like its width, how many floors, and so on. From there, a developer will need to make smaller decisions, such as the width of the buildings windows, whether or not it has graffiti on its outer walls, and decorations like plants. There’s also a texturing phase the apartment must go through, where developers pick details such as the width and type of brick the building’s made of. All combined, Price estimates, this work could take days, costing around $4,000.

But that’s one building. For a developer wanting to create variations of the same building, the set parameters and small details strewn throughout can easily lead to a lot of additional work. “Like, let’s say you just want to change the [height] of the floors, well now that’s completely messed up everything because now the window decorations are out of alignment, the staircases don’t match,” Price says. “If you want to duplicate a lot of these apartments, you often have to redo a lot of the work, because the moment you change the modeling, the texturing has to be changed as well, so then you’ve gotta repaint textures. If you do things in this traditional way, it just doesn’t scale.”

“In some ways, I feel like the workflows that were made to build games in the ‘90s or whatnot haven’t changed that much to today,” he says. “A lot of people are still hand-modeling these buildings.”

One of Price’s proposed solutions is procedural workflows. Rather than hand-building each apartment and the small details within, developers can set up a program that generates endless variations of the building with set dimensions and parameters.

“Like, ‘Let’s not just make one asset, let’s build the system that will enable us to generate 1,000 variations of that asset,’” Price says. “And then now once you’ve set up all these different things one by one by one, typically using software called Houdini — once you’ve got that, then you could put it through a sampling system and it will literally spit out infinite variations within the ranges that you’ve set.

“I think the gaming industry’s slowing entering into that,” he continues. “You’ve seen in the last couple years a lot of talks by top game studios — like they’re talking about how they made Far Cry, the latest Spider-Man game — they do actually use these procedural workflows. I think they’re going to become more common instead of the expected.”

As game worlds get bigger, and the expectation to fill them with tiny details to make them feel more real increases, options like Price’s offer a way to streamline the development pipeline. And as the next generation of console hardware gets closer and closer, developers speaking for this story are already considering their options — or at least acknowledging they need to be thinking about it.

“All of our source assets are currently made into a higher resolution than they actually end up in the game,” Taipalvesi says. “So therefore, with the current restrictions that we have from technology, from platforms, from anything else, we tend to make source assets into a higher degree of fidelity already because it is so much easier to scale down the quality than to scale up. Scaling down goes automatically, scaling up means manual work.”

“We’re being very shortsighted right now,” Krankel says, laughing. “We’re only thinking about the games right in front of us. [...] But yeah, I think ultimately if the expectation continues to escalate in terms of fidelity, then we will continue to have to find outsourcing solutions that we partner with in ways that we’re not carrying 50 to 100 people. We don’t want to, as a studio, ever put ourselves in a situation where we’re at a lot of risk either.”

If history is any indication, when the next generation of video game consoles shows up, most of these challenges will only get more difficult. And while that may be a necessary evil, developers of all sizes are not only rising to the challenge, but finding smart ways to work around them. There’s no going back from here. Players are expecting more from games, visually and mechanically.

“We try to be slightly ahead of the curve whenever we can,” Taipalvesi says. “But yeah, it is a race. It’s a technological and artistic game for developers to stay relevant to platforms that they use to release the games.”