Below is the full Post Mortem Interview Game Developer Magazine conducted with SWERY and the Access Games Staff back in August 2010! It’s a great read for any Deadly Premonition fan!

Access Games’ Deadly Premonition had a long and difficult development period that was beset by technical problems. The game was dragged from the precipice of cancellation several times and ultimately persevered to become an underground hit. In this candid postmortem, the staff of Access Games describes how a passionate commitment to their game saw them through the production’s highs and many lows.

UPDATE 02/01/2014: Following the magazine’s closure in July 2013, GDC Vault compiled an archive and made all past issue freely available for all to enjoy. You can download the entire August 2010 issue there!

Vol 17 No 08 August 2010

POST MORTEM- ACCESS GAMES’ DEADLY PREMONITION

This project began in the early fall of 2004. Looking back, it was more than five and a half years ago. After we released Spy Fiction in late 2003, Access Games wasn’t lucky enough to have the opportunity to work on another original title, and we wasted nearly a year deciding what our next game would be. Of course, it’s possible we only received the chance to make such an uncommon game because we had so long to harge our batteries. The creativity that built up over so many months was about to explode, looking for an outlet to express itself. Our energy found an explosive release in the form of Marvelous Entertainment, and crystallized into what became Deadly Premonition. The process was logical, and you might say that we set out to follow a standard development course. But that course turned out to be more precipitous than we could have imagined, and it was, quite literally, a bloody few years. Numerous obstacles stood in our way: unclear nextgen console specs, the decision to go multi-platform, and the threat of project cancellation. It was all we could do to stand up to the hardships and remain steadfast in our implicit faith that the game would eventually be completed. In the end, I learned that true effort is always rewarded. After pushing this project through, we found a gratifying reception waiting for us in a distant, foreign land.

>> What Went Right

1] CHARACTER BUILDING AND BACKSTORY (ESPECIALLY YORK AND ZACH)

– SWERY (director) & Hitoshi Okamoto (art director)

We believe our greatest success in Deadly Premonition was the establishment of our main character. Many different types of protagonists have been created – and loved – in the history of videogames, so our primary mission was to devise a type of hero that had never existed before. DEADLY PREMONITION is essentially a mystery, and although quality peripheral characters are crucial in good mysteries, the main haracter is especially important. I believe players pick up on the fact that while York may appear very handsome, he’s a true eccentric inside. York speaks his mind with no regard for the feelings of those around him while constantly muttering to himself … In other circumstances, it would be no surprise if players grew to dislike him, wondering “Who the hell is Zach?” [York has a second personality, named Zach, living within his mind. There is a strong argument to be made for Zach being “the player.” – ed.]

Yet everyone who plays the game seems to love him. Why? Perhaps it’s because players recognize that he’s extremely charming and reliable – a friend worthy of admiration. Of course, we don’t think we gained that recognition for free. There’s an important device at work: the invention of the “Zach as Player” relationship. Agent York pursues his murder investigation in Greenvale, the player munches popcorn in his living room, and Zach is the bridge between them.

You might say that the character we were aiming for is established the moment York and the player become true friends, joined through the conduit of Zach. Agent York is a character who only springs into existence with the assistance of the person playing the game … and that turned out to be exactly the new type of hero we were after.

We knew going in that a character created with the exclusion of the player just wouldn’t work. Please show York the same love you’ve given countless other video game characters before him. That’s all we – and he- can ask.

2] STORY AND WORLDBUILDING (BOTH INITIALLY TOO EXTREME)

– SWERY

We spent as much time polishing our storyline and worldview as we did our protagonist. During early development, our setting was much more urban and cynical than the final product. The presentation was also more violent, and decidedly more extreme; I think the one-year gap between original projects may have turned our thoughts toward violence. Our final world came together only after several iterations, and the overcoming of multiple near-cancellations. For the story, I kept as much sense of reality as possible while embellishing it with a somewhat dreamlike milieu. As a result, I believe the sections depicted as eality have a dreamlike feeling, and likewise, the actual dreams are given a sense of realism- we were able to smear the boundary line. There are very few video games so particular about depicting such things, and I think DEADLY PREMONITION may have become a unique example among them.

A realistic setting was also completely necessary for bringing this story to life. To build it, I posited three “reals:” “real time,” “real scale,” and “real life.” Implementing these with our limited budget and resources proved to be extremely difficult, but despite a great deal of resistance (even from within the team!), I felt these three concepts were essential.

First, to reproduce our five-square-mile town, we visited America for data collection, using measures to determine the width of roads, billboards, and railroad crossings. Second, to depict 24-hour time spans, we calculated weather patterns and the angle of the sun by giving our fictional town a real-world latitude. Finally, we wrote extremely detailed profiles for the townspeople (blood types, birthdays, favorite foods, favorite music, people they disliked, what age they were when they had their first kiss, etc.). While not necessary for the game proper, these were used to create individual 24-hour activity patterns for each character.

For example, when the punctual character Thomas rises in the morning, he goes to the toilet, washes his face, and brushes his teeth. If you have a moment while playing, try spying on his apartment- you’ll actually see Thomas performing these actions. Despite not appearing anywhere in the game, we even set the hourly wage for the A&G Diner (about $3.50 per hour, and $25-$30 per day, depending on tips). And, of course, every street in Greenvale has its own name. For a project as rash as this-replicating an entire rural American town from inside Japan- details like these were very important.

These factors layer and combine to form the town of Greenvale, and imbue DEADLY PREMONITION with a sense of reality. This approach is particular to our team, and I wouldn’t necessarily recommend it to other developers … but should you get the chance, why not give it a try?

3] DISTINCTIVE MUSIC

– Keiji Teranishi (planner)

Music was another focus in DEADLY PREMONITION, and I feel our success in this area was worth the effort. The soundtrack served to further bolster our unique story and world; it may even strike a mood never before heard in games. Our unusual score owes a great deal to our success in helping the composers appreciate the unique aesthetic behind the game. We showed them countless design documents, played music we thought would fit, and sometimes even hummed song ideas – efforts that resulted in the musicians’ understanding of the work.

Doing our utmost to convey the game’s aesthetic caused the town of Greenvale to take shape in the minds of our sound team. By the time its citizens began living their daily lives inside our composers’ heads, the melodies of Greenvale were created almost automatically … or so it seemed to me! As proof, I submit “Life Is Beautiful,” a memorable whistling tune that conjures up the feel of a peaceful country stroll, and “York And Zach,” a song that plays during monologues representing the mysterious image of our hero. Both initial demos were given the okay almost immediately, and both were easily worked into the game. I can think of no other songs that match the worldview of DEADLY PREMONITION so perfectly.

Several other tracks, however, were created only through much trial and error. Particularly difficult were “Red Tree,” a theme representing madness heard in the Red Room at the outset of the game, and “Miss Stiletto Heels,” Carol’s song, which might be considered DEADLY PREMONITION’s second theme. We were extremely specific about our orders for these tracks – not only for the arrangements, sound levels, and effects, but also aspects as minute as the timing of hi-hats within single bars! The lunatic improvisational section in “Red Tree” also went through numerous takes with different instruments and arrangements before being judged complete.

In the end, I believe DEADLY PREMONITION’s music succeeded in adding depth to our unique world and characters. If I were a greedy man, I might say that having a few more tracks would have made it even better!

4] CASTING AND VOICEOVER

– Keiji Teranishi

I feel that the voice work in DEADLY PREMONITION was even more successful than the music. We spent two weeks recording at WebTone Studios, which was in San Jose at the time (the studio has since relocated to Los Angeles). Over 6,000 total lines needed to be recorded, including pre-loaded and streaming samples. That meant we had to get approximately 500 lines in the can every day. Two weeks may seem like a long time, but Access Games is located in Japan, so it was absolutely imperative we lock down everything required in the time allotted. The pressure was extraordinary.

We spent those two weeks completely sequestered in the studio … but like Agent York himself, we were also becoming familiar with a new town as we shuttled back and forth from our hotel. These experiences proved invaluable after returning to Japan, so one might say the San Jose trip allowed us to kill two birds with one stone.

Before recording began, we showed our actors images of the characters to help create their roles. Although we recorded only after explaining in detail other aspects like the general tone, personalities, and family relationships (Greenvale is full of eccentrics with more than a few quirks), directing the performances turned out to be quite a task. When he had difficulty getting his point across, SWERY occasionally gave direct instructions by getting in the booth and performing lines himself.

Among all our fine voices, one actor performed his role exactly as we imagined … and perhaps even better: Mr. Jeff Kramer as York. Jeff brought even greater depth and style to the lead role than Access Games had envisioned, for which we are deeply grateful. Despite having only five days to record some 3,000 lines (he was the lead, after all!), Jeff presented us with a truly brilliant performance.

Of course, each of the actors portraying Greenvale’s vibrant characters – George, Emily, Thomas and the rest – turned in wonderful performances. If the fans who play DEADLY PREMONITION get a sense that our characters might be real people living their lives somewhere, then our casting was indeed a success.

5] OUR LOVE AND PASSION FOR DEADLY PREMONITION

– SWERY

In trying to compile “Five Rights,” one might expect some talk about fabulous play control or revolutionary graphics. I can’t say either of those were particular successes in DEADLY PREMONITION, so I don’t think such topics would be appropriate for our fifth “Right.”

After much internal debate, I came to the conclusion that Deadly Premonition’s greatest “Right” is the fact that we poured our love into the game. That really says it all, so I’m going to use this space to touch on a few aspects we were particularly obsessive about. I hope this text gives you some idea of how much we cared about our game, and indeed how wasteful some of our efforts may have seemed.

· Side Missions. I was insistent that the content of our side missions always draw upon aspects of the NPCs’ personalities, thus adding further depth to the world of the game. We built the missions so that players who pursued them might learn more about the personalities and lives of our characters – aspects not revealed in the main story.

· Observation And York’s Monologues. Examining certain items in DEADLY PREMONITION (products in stores, posters on walls, objects inside NPCs’ houses) will reveal messages unrelated to the main story, written from the perspective of York speaking to Zach. We created these messages with an emphasis on the question, “If York were actually investigating this object, what sort of impressions would his peculiar sensibility elicit?” For players who want to get the most out of the game, these insights serve to engender further sympathy with York.

· Coffee Fortunes. Coffee is a major factor in the main story, but we also included a feature that allows players to view York’s “coffee fortunes” whenever they wished. Like a line in the main story says (“my coffee warned me about it”), we added these interactive “warnings” to bring an extra bit of depth to the concept.

· Turn Signals And In-Car Navigation. Although they have no bearing on the game itself, turn signals can be activated on every car. And have you noticed that the car navigation systems are actually functional, and synced to the HUD’s automap? Features like these may seem extraneous, but I truly believe they accentuate the game’s sense of reality.

· Conversation While Driving. DEADLY PREMONITION’s game area is five miles square, so a lot of time is spent traveling in cars. What happens on long drives with someone in the passenger seat? Conversation-though not many games have done it. Incorporating the chat feature brought extra realism to these sections of the game. Zach’s presence is also given a chance to shine, as he ceases to become a simple internal monologue and actually enables Agent York’s conversation with the player.

· The Greenvale Map. If you look very closely at the full map of Greenvale displayed in the menu screen, you might notice that the town is shaped like Kaysen’s dog, Willie. Willie adds a small accent to the story by appearing with Kaysen, but he is in fact an extremely important character with an intense, unrevealed backstory. We didn’t have a chance to expand on Willie in the main plot, so we secretly concealed a hint in the form of the town map for those who are truly perceptive.

>> What Went Wrong

1] MEMORY ALLOCATION AND PROCESSING SPEED

– Wataru Nishide (lead level artist) & J’s Kataoka (lead programmer)

DEADLY PREMONITION was our first stab at next-generation development, so we began the project astounded by the vast amount of RAM available- much more than any previous consumer hardware had offered.

Astonishment gave rise to overconfidence, and eventually to the worst-case scenario: our data management became sloppy.

Attempting to work on memory allocation in such a state was very dangerous, but we went into production unaware of the risk. The result was a constant struggle with remaining RAM. This was glaringly evident with the motion data in particular-it ended up occupying a truly massive chunk of memory, which led to system restructuring further down the line.

In addition, we set far too many objects in our outdoor scenes, and could not process them all effectively. Dealing with trees and shrubs was especially trying. With that said, cutting too many objects would have reduced the object density of our expansive outdoor map to unacceptable levels, so we had to keep both the optimization of code and the appearance of our world constantly in mind.

Similarly, we had entirely too many points of collision enabled, which became a primary factor in processor bottlenecking. This required scrupulous massaging of collision data right up to the end of the project. Lighting and shadows (which we will go into more detail about below) had a huge impact on our cycles as well. Our model forced us to do constant optimization throughout all our code, which led to a huge loss of working time.

Many other similar issues can be cited: our water ripple effect, refraction effects, and reflections in mirrors. We do think memory management on this project was an exceptionally useful learning experience, but from the perspective of development, it must be considered a failure. When considering the impact memory allocation has on a game, it goes without saying that this aspect was one of the least successful on the entire project.

2] LIGHTING AND SHADOW PROCESSING

– Wataru Nishide & J’s Kataoka

Since this project allowed the use of pixel shaders, we experimented with various types of lighting- specifically, real-time processes like flat lighting, point lights, and spot lights. Unfortunately, we were a bit overzealous with our lighting calculations which resulted in severe processing bottlenecks.

Even worse, we based our data formats on the assumption that we would be implementing these different lighting techniques as is, so there were absolutely no lighting effects done with textures; this had a huge impact on later code revisions.

Point and spot lights take a heavy toll on graphics rendering, yet brightly lit interiors like the diner had been built with dozens of them. It wasn’t just a question of adjusting the positioningor number of lights; the situation escalated to the point that we had to modify our original resources.

Dealing with shadows brought similar hardships. The process of implementing shadows for flat lighting outdoors, point lights indoors, and spot lights used at other specific locations soon became a mighty battle with CPU cycles. A spectacular amount of objects in the vast outdoor areas needed to cast shadows, and because there was little in the way of surface obstructions, the shadows had to appear exceedingly sharp and tight. However, generating such sharp shadows required an enormous amount of VRAM, which simultaneously caused our performance to drop significantly.

A great deal of labor was then expended on repetitive testing and adjustment of various shadowing techniques: perspective shadow maps, light space perspective shadow maps, and cascaded light space perspective shadow maps. In the end, we were not only forced to compromise on the sharp shadows we wanted, but the s ituation degenerated to the point that we had to modify almost all our resources. Indoor shadows, too, would be processed redundantly several times if an object were lit by multiple light sources, which also led to serious speed drops.

In addition, we faced similar consequences with the self-shadowing on our characters. Selfshadowing can be highly noticeable onscreen, so we strove to render it attractively. Of course, the more high quality our self-shadows became, the more cycles and VRAM were consumed, so we had to work within a delicate balance using jitter textures and the like.

These aspects of game development are commonplace today, but at the time, they were tremendously difficult for us. We simply did not have the proper technical know-how yet.

3] USING THE PHYSICS ENGINE

– Yutaka Ohkawa (programming support)

At the beginning of development, I was convinced that implementing a physics engine (NVIDIA’s PhysX) would allow us to build an exciting, neverbefore-seen world. It was indeed a great boon to the project, driving the behavior of vehicles, the interaction of objects, and even the fluttering of hair. PhysX did allow us a glimpse into a new world. However, the use of a physics engine was not all positive. After incorporating too many realistic physics behaviors into the world of our game, we often found ourselves at their mercy. Game worlds require not-infrequent “cheating,” but physics behavior is brutally honest. In situations where we had to implement convenient cheats-bending the behavior of a character’s prop in a demo scene, or forcing vehicles to navigate properly, for example- physics behavior would conflict.

During development, physics behaviors would often be at odds with what we were expecting, so adjusting the parameters to make things look right was rough indeed. This was especially true for characters, whose physics behavior would change any time their motion data was altered. As a result, even parts of the game that had already been approved would immediately break after motion tweaking. The problems manifested in gameplay, of course, but also in demo scenes; the process of correcting them ended up depleting a tremendous amount of time.

The processing load for physics operations also proved to be much higher than expected, which was another large miscalculation on our part. Because we originally had everything that could possibly be displayed with a physics engine running through PhysX- hair, clothing, even fishing rods- we failed to produce an enjoyable gameplay environment. In our zeal to pursue individually moving objects, I might say we lost sight of the overall processing picture.

To maintain frame rate, it became necessary to optimize; we reduced PhysX actors and adjusted computational loads for each scene individually. However, this made an adversary out of the game’s key eature “freedom” as we had to consider an enormous number of variables when making the adjustments. This process was not quick, and when combined with the aforementioned physics behavior tweaks, it consumed even more of our time and energy.

Physics behavior can certainly bring dynamic expression to game worlds, but we learned that it isn’t necessary for everything. Determining exactly when to deploy it is of the utmost importance.

4] SOUND EFFECTS AND SURROUND SOUND

– Keiji Teranishi

As I stated above, the music and voice acting in DEADLY PREMONITION went very well, but I cannot say the same for our sound effects or 5.1 surround sound. With nextgeneration hardware, we knew sound specs would improve alongside the graphics. The increase in breadth of expression was attractive, but we were incapable of bringing out the full potential of the hardware at the time.

What was the source of this failure? Although there were quality issues with our original sound resources, the biggest problem lies with the fact that Access Games has no sound department. Because we entrust all audio aspects of our games to an outside partner, there is little technical knowledge of sound within the company, which made it tough for us to provide satisfactory quality control. Having turned a blind eye to the problem, we missed the opportunity to try and compensate for the lack of manpower assigned to these issues.

Of course, it goes without saying that many development teams lacking sound departments release games with high quality audio. We should have studied and discussed such games, asking ourselves how we might incorporate their quality sound into our product. Indeed, we should have been closely considering measures to amend our faults. Instead, we put everything into the voice and music, which effectively reduced the priority of the sound effects. This is something I deeply regret.

And why is that? All audio produced by a game is important-music,sound effects, and voice. Only when the three come together as one can we truly speak of a game’s “sound.”

Looking back, I think something could have been done about the schedule/manpower issue and other problems before they became too unwieldy. Avoiding the creation of problems is certainly an important factor in game development. But this project reminded me that when problems already exist- or loom on the horizon- the ability to work as a team and find the best solutions is even more essential.

5] SCHEDULE MANAGEMENT AND MOTIVATION

– SWERY

I would say that the final and greatest “Wrong” on DEADLY PREMONITION was schedule management. The ideal director is supposed to be able to handle both game quality and project progress, but that was a very difficult proposition for me at the time.

I regret forcing my development team to work longer and harder than was necessary. It may be true that the lack of technical expertise covered above contributed to schedule delays, but as the man surveying the entire scene, the greatest responsibility lies with me.

I have no intention of criticizing a game I obsessed about deeply, and poured my soul into. Those were good things, in fact. However, managing team motivation so that such games can actually be realized is equally important.

On this project, I worked myself to the bone, single-mindedly absorbed in production to the point of blindness, unable to see the people around me- which may be why DEADLY PREMONITION has such a strong sense of “authorship.” In the future, however, I want to forge a game production path in which both schedule and motivation management are properly implemented.

I ALMOST FORGOT THE MOST BASIC OF FACTS!

» Somehow we overcame a prolonged production schedule, near-cancellation, and countless other hardships to get DEADLY PREMONITION into gamers’ hands. Through the process of producing this game, I learned: never give up. It sounds so obvious, but the meaning is simple: even though you may believe in what you must do, nothing will come to fruition until you actually do it. DEADLY PREMONITION may not have been a huge financial success, but what may be more important is that it was a work that allowed our staff to grow, strengthen their bonds, and better our relationships with all the companies involved. This is not the goal, but the starting line; now, we embark on our next stage in game development.

York: “Say, Zach. What do you suppose is waiting for us on the road ahead? No, don’t answer that … Hardships are part and parcel of life. They’re what let’s appreciate the good times.

Isn’t that right, Zach?”

And that’s it. I love you all!

GDM – This article was translated by Nick Des Barres.