It Takes Two

Mechanics and settings were prototyped separately, then paired up for level designers. Art explored places a shrunken couple could visit, with the tree an early pick, while design prototyped many mechanics at once, including the sap and match guns. Each mechanic was matched to a setting, and each designer or pair got a level to fill.

Oliver Granlund and Robert Johansson built the Tree and the snow globe level.

The guns had to do more than explode before they could carry puzzles. The first sap and match gun only sapped and exploded, which was great for combat but useless for puzzles. The team pushed weight and spinner mechanics, so sap gained weight and could pull things down.

The team chose early to let sap land anywhere rather than on one white spot in a black room.

Remove the spots where a mechanic feels bad, keeping the puzzle intact. Because sap could be shot anywhere, Cody could sap the outside of the first wasp nest cage where May couldn't reach, and in version two he sapped the roof so May had to look straight up. Much of the work was eliminating those cases, with one or two passes on nearly everything.

"A lot of the work we did with these mechanics was eliminating places where you can do things that just doesn't feel nice."

Fold controls into the core weapon instead of adding UI. The boat first moved by recoil, which failed because you couldn't see where you were going. A rudder was fun but broke the level's rules of kinetic play and minimal UI, so it ended as simply pressing fire to go.

"Instead of having a lever you press Y on, you would blow something up." The boat only reached its final form through the team's make, playtest, talk and change loop.

Stop each mechanic at good enough, because unused depth adds nothing. Variety was the core pillar alongside finding good enough. The boat could have been a 30 minute segment or a whole game, but it lasts five minutes, and Oliver Granlund says a longer boat would have made a worse game.

The team put its best finds into the level rather than dragging ideas out, and applied Nintendo's four step level design rule only in places. "Depth doesn't mean it's good if you don't have time to use it."

Budget polish time: a one week prototype takes far longer to make fun. Hazelight built things in about a week and dropped them in, but taking an idea to fun gameplay took much longer than expected. Half the work was making mechanics clear and half making puzzles smooth.

The vault door first exploded when sprayed but had to be tiny to keep the camera tight, so it became a turning lid. Ideas stayed similar while execution ended up wildly different; Hazelight is not triple A and could not spend unlimited time on one thing.

Give artists context before they build, and let them surprise you. Handing art abstract gameplay and asking what it could be is no fun for anyone. When the big spinning wheels had no context, artists turned them into an elaborate lab that shaped parts of the story.

Costly features like the swarm technology, the hammer and the seamless fight-into-slide had to earn their cost. "Game development is, you know, it's a team sport."

Mix parallel, step-by-step and simultaneous co-op, and pace between them. Parallel co-op is two people playing their own solo game side by side, step-by-step is turn-taking where one action enables the other, and simultaneous needs both acting at once. Hazelight wanted all three and calls pacing them the big challenge.

The side-scroller camera came in fairly early and fixed camera problems in the spinning lab section.

Give each player a distinct role, and tune the game to keep both busy. In the wasp fight Cody is the instigator who saps and May the detonator who ignites. Since May has little to do until a wasp is sapped, wasps are tuned to attack her while she waits.

A separate tutorial was added because without it both players shot every shooting sign and the same yellow highlights and got confused.

Give each level its own visible rules, learnable in minutes. With a systemic sap gun, anything weighed down that makes something go up must be visibly connected to it. There are no Mario style floating platforms; each level sets its own grounded rule set.

"It's basically like teaching a player a new game in an hour."

Fill playtest gaps with hidden side interactions that replace collectibles. After each milestone, playtests showed too little to do, so the team went on a spree of minigames and side interactions. Collectibles had no good purpose, so the hidden minigames became the collectible.

They were hidden because they could distract. One began as "wouldn't it be fun if you could trap your friend" and went from ridicule to torture.

Competitive minigames break up constant cooperation and echo the bickering couple. In a game built on cooperation the pace is hard to vary, so rivalry adds a change of moment, and when the characters bicker the players do too. Minigames like tug of war and a literal chessboard suit different players.

"Not everything would have to be technically skilled, because then the same person would win over and over."

Obvious co-op health systems failed; a respawn timer worked. A shared health bar failed because the weaker player took damage until both died, and a Gears of War style revive failed because players are often apart or platforming. The final design is a respawn timer that runs while both live, so each player effectively has two lives.

The timer turns into cheering: "don't die, I'm almost back." "It's fun to die because then you feel like I'm challenged and you get that stress, but it's not fun to redo it." Replaying parts goes against the motto of variation.

Lenient aim makes combat about dodging, so mixed-skill pairs can play. Auto aim is very lenient so the challenge is dodging wasps and spraying the right target, not landing skill shots. The Tree is the only level with aimed combat and the hardest for new players, yet it ended up more usable than most shooters of the time.

It is still hard for people who can't use both sticks. The beetle, meant as a mini enemy, became a boss by accident; an earlier version made players shoot it on the butt, needing aim and communication, and adding grates reduced the aiming.

Prototypes piled up and stretched a one hour level to nearly two. The Tree was storyboarded as one hour but grew to nearly two, diluting the story, so cutscenes such as the robot squirrel scene were added late to carry it.

A segment's story weight only shows once the game is nearly whole: too long without story bores players, and too long a bare cutscene loses them.

Constant variety still needs an intensity curve that rises and falls. Most games pace repeated platforming and combat, but here every stretch is new. Oliver Granlund says intensity must still go up and down so players aren't exhausted or bored, and admits the team should have paid more attention to it.

Nothing limited how strange the level got; meeting feedback was to make it crazier, "like Avatar on LSD", and players ask whether they are still in the tree.

Pair each mechanic with a theme so gameplay carries the story. Magnets are about attraction and clockwork is about time; the team calls these emotional levels.

Hazelight's earlier game, A Way Out, was story focused; here gameplay won, but the motto was to keep gameplay and story together.

Players follow environment cues, so make cues point somewhere safe. Ants added by art as decoration led players down a hill where they jumped off and died. Henrik Sandin then used ants on purpose to show the way players should go.

His calm section came very late, between storm sections, yet looks like the first thing you play; it introduced swinging, which then had to be added across the rest of the tree.

It Takes Two lets players skip button-mashing quick time events. Mark Brown counts button-mashing QTEs as a barrier and names It Takes Two, Guardians of the Galaxy and House of Ashes as games that let players bypass them.

Menu narration lets blind players start the game and reach settings. It Takes Two ships a menu screen reader, alongside Forza Horizon 5, Age of Empires IV, Far Cry 6 and others.

Some of these games turn narration on by default so players can reach the settings unaided.

Beat analysis paralysis with an arbitrary deadline and minimum scope. Oliver Granlund, who worked on It Takes Two's magnet snow globe level, advised Mark Brown to take a game jam mindset: about one month, only the core mechanics, and five levels.

"I would set an arbitrary time limit (that you're probably gonna overshoot)." Mark committed to 30 days for five levels of Mind Over Magnet.