Sakurai on Creating Games: Motion (18 Topics)

Masahiro Sakurai, the creator of the Kirby and Super Smash Bros. series, shares a wealth of game-development know-how on his YouTube channel “Masahiro Sakurai on Creating Games”. This article summarizes and restructures the content of that channel by topic.

That said, a summary is only an entry point. So much of the value — Sakurai’s own words, his real-world examples, his pacing, and the footage itself — can only be gotten from the videos. So please don’t stop at reading the article; I strongly encourage you to also watch the original videos embedded under each topic.

This category ranges from the principle “overdo it for just right” to techniques like interpolation, blending, and rotation. Here we bring together the key points of both videos in the “Motion” category (18 individual topics), structured in 2 parts following the 2 source videos. Each part opens with its explanatory video, so please watch along.

Part 1: Overdo for just right, the 4 parts of an attack, flat scaling (#01–#09)

Motion (character movement) is an especially important element in Smash and the like. From how to specify movement to the principles that create good feel, the production know-how is discussed.

https://www.youtube.com/watch?v=W1Qfqa2G1Cc

1. How to specify motion

When planning a Smash fighter, Sakurai explains how he conveys the nuance of moves and motion to motion designers. First, indispensable is the motion list. He writes out all motions as text, noting the image of the movement, the motion’s length, and the time until an attack comes out. At this stage, feeling the mind games, he makes it reasonably appropriate.

Next, he specifies key poses — idle, movement, attack — with photos of figures. He takes out a figure, photographs it, removes the shoulders, adjusts contrast, and works it into the spec. This makes the move’s image clear. He often prepares around 70–80 photos per fighter.

But even gathering all this, it’s just the “beginning” in the production process. A figure is a figure, and its range of motion is on another level from the game model, so there’s a mountain of things to have the motion designer think about. He supervises repeatedly, and sometimes gives up if it doesn’t work, but it’s produced with care so the character has identity. Note that he has all motion designers hear his directions together, to avoid repeating the same ruts.

2. Overdoing it is just right

Stop fussing “this is broken, it’s a bug” when you see the waist twist of Kazuya’s “Spinning Demon” — so says Sakurai. A motion swung so extremely that the model’s weights break is a good motion. He OKs this because it was so in the original, but really it should be pushed even further, he says.

When making attack motions, there’s almost never “too much”; “nowhere near enough” is an everyday occurrence. Since it’s a small thing on a small screen anyway, think of overdoing it as just right. The first 3D fighter Virtua Fighter even has movements where the upper and lower body face nearly opposite directions. A range of motion being technically correct isn’t a good thing. Note that for linear movement, “going too far then pulling back a little” is also effective — he’s done this since the Famicom days (Kirby’s Needle ability).

3. The parts of an attack motion

The attack motion found in many games. Sakurai divides it into four: “idle, wind-up, attack, follow-through.”

Idle is the origin of all; since most actions are built from here, not wavering it is essential. Wind-up is the anticipation/charge (the pose starting the attack). Attack is the key pose, and it’s vital that the pose at the instant the hitbox comes out is crisply struck (if that’s no good, it’s all no good). Follow-through is the recovery, the time to return to idle. Set cancel frames in the follow-through so you can move or jump before fully returning to idle. Up to Smash Melee there was no such mechanism, so controls felt a bit heavy and cramped.

4. The wind-up: instant and extreme

The wind-up is the pose from pressing the button until the attack actually comes out. Smash’s wind-up has traits. First, the pose changes quite a lot from the instant you press the button. Carefully interpolating from idle increases a sticky feel, but Smash prioritizes response, returning a reaction as soon as input is accepted. Precisely because online and TV lag can’t be prevented, instant response to input is ideal.

On the other hand, there are attacks with almost no wind-up. Zero Suit Samus’s weak attack pops out in 1 frame, and Final Fight’s jab can be mashed instantly and feels good. A game holds up even without a wind-up. But without a solid “charge,” it lacks the impression of a forceful hit, so it’s not just about making it fast. From the viewpoint of firmly conveying input acceptance, it’s good to consider the wind-up.

5. The attack pose

Whether the attacking action feels good greatly changes a game’s feel. In Smash, hit-stop kicks in when an attack lands, and the pose at that instant stays in the head most, so the figure photos form that instant too. Sakurai’s points:

• Make the action exaggerated overall, clear even from afar
• Make it an attack that crosses the center line (drift to the back or front and it stops hitting)
• Avoid symmetry (left-right mirror) (use it only when there’s comical intent)
• Avoid making the torso perfectly sideways; twist a little for sharpness
• For a sword swing, hit-stop at the swung-through point as far as possible
• Endure some perspective (the difference in look between left and right of the screen)
• And above all, make the silhouette look clean

Since Smash is a game seen from the side, the silhouette from the side matters. Striking that is absolute. This sense, he says, is cultivated by jiggling 3D objects around.

6. The follow-through decides the impression

The follow-through is the action that becomes the opening after an attack. You might think “what happens after attacking isn’t important,” but this is the part that catches the eye most.

In Smash, to make the attack response fast, going from wind-up to attack is overall quick. A standard attack mostly finishes in under a second, and the attack comes out at about the first 1/3–1/4 of the total time, 1/5 if early. In other words, all the remaining time is follow-through. Short until the attack, long after it.

Since the attack itself ends in an instant of seconds, the follow-through must be a motion that sufficiently conveys “what it was doing before that.” Rather than just interpolating back to idle, hang on to the very limit of what’s tolerable, expressing the attack a touch long. Then set cancel frames in some of the interval returning to idle, and actually accept input on poses that look operable. The follow-through after an attack, too, should be staged firmly without slacking to the end.

7. Act big because information decreases

An episode that stayed with Sakurai is 2002’s Clock Tower 3. This was the final work of the great film director Kinji Fukasaku. The acting in the visual scenes is so exaggerated it’s almost funny. About that, the director reportedly said: “Characters lose the various information humans carry, so you must act bigger to make up for it.”

It should have been his first game production, yet it hits the essence — Sakurai was surprised. Because game characters don’t actually exist, acting big is just right. Even captured via motion capture, it’s usually unusable as-is, and moving it quite big is about right. The character’s size relative to the screen matters too; being small or far makes movements feel cramped. Even in action games, the basis is to move big and boldly. For a sword-swinging action, just swinging isn’t enough.

8. A recommendation to pose

Playing “vroom” with an articulated figure is called “bundodo” in Japan. Per Sakurai, this is a fairly decent pastime for motion designers. It may not suit pursuing realistic movement, but striking a pose is very important.

Japan has mountains of good action figures. How do you make a pose work? What’s a good angle? Actually picking up a 3D object teaches you all sorts of things. Using the figma Pit as an example, Sakurai lists pose knacks: avoid symmetry, avoid facing dead-on, tuck the chin, don’t make the neck perfectly vertical, don’t line up the upper body, waist, and legs in a row, raise the shoulders a little, show the left and right arms from different angles, bend the wrists lightly toward you, break the center of gravity a bit, splay the feet slightly… Even a motionless standing pose can be thought through in many ways. Ultimately it’s not logic but “what you can think looks good.” To cultivate that sense, nothing beats handling 3D objects.

9. Flat scaling

Scaling a model up and down is called scaling. CG movement is in principle composed of three elements: scaling, rotation, and translation (movement). Movement and rotation are everyday, but scaling isn’t so much. Yet you mustn’t take it lightly.

A bouncing ball’s movement, too, with scaling combined in rather than movement alone, gives a completely different texture and feel. This is called squash & stretch. The point is to roughly preserve mass. Squash vertically and stretch horizontally, and so on. Kirby’s crouch and landing have been flat since the Game Boy days, but simply squashing lacks a sense of volume, so they devise giving volume to the edges. Mario’s fist enlarging on a punch is more of an exaggeration and a bit different. If you’re stuck on flavor, try working in scaling.

Part 2: Damage expression, flip, motion blend, 2-axis rotation (#10–#18)

Where the first half was about principles of movement like “overdo for just right,” the second half centers on more technical know-how — interpolation, blending, rotation. Concrete knacks for moving characters in 3DCG are discussed.

https://www.youtube.com/watch?v=XraQdYJCxiQ

10. Damage motion

About the damage motion when hit by an attack. Smash assumes that the instant you take damage, it immediately switches to a painful-looking pose. The first frame of the damage motion looks long because hit-stop is applied, so making it look painful here matters. Some works have the hit pose close to the idle pose, but Smash makes it take an extreme pose totally different from idle, so you can clearly tell it took damage even from afar.

However, from the pose at the instant of being hit to the damage motion, it’s joined with a slight interpolation. Since it always becomes the damage pose no matter the situation, if the opponent’s hurtbox is long, you end up taking the damage motion at a fairly distant position. There’s a hand of keeping the hurtbox extended while applying hit-stop, but he does it this way to emphasize the sense of being hit.

Smash’s damage reactions split into upper, middle, lower, taking the reaction near the hit position. Further, on the ground there are small, medium, large by strength, making nine combinations with upper/middle/lower. Launch and the variations increase further. And the trouble is that almost everyone in Smash is a non-humanoid with a too-unique body shape. If everyone were a standard humanoid you could reuse motions, but they aren’t, so adding even one fighter is a big deal. For every damage motion, that the silhouette is firmly visible is emphasized, twisting the body so the limbs’ silhouette shows.

11. The trap of motion interpolation

Motion-creation tools are very convenient: set a few poses (keyframes) and they interpolate the in-between for you. Fixing breakage like clipping is easy too. But this convenience becomes a pitfall.

When fitting a 2D game into 3D, the 2D snappy impression is often lost even with the same poses. With the same time and same poses, evenly interpolating the in-between movement tends to look “mushy.” Sakurai cites Ryu’s heavy punch (uppercut) as an example. The original is a very well-made movement expressed by cycling 3 frames, but just interpolating that as-is into CG kills the momentum.

So in Smash, they weighted the time to the wind-up and attack, making the approach to the key pose (keyframe) rather extreme. Draw the forearm back further to take a longer charge, and send the swing-through in 1 frame all at once, crisply. The follow-through doesn’t trace the same path going back, hanging on to the very limit of becoming the same pose as idle. Interpolation fills and supplements the in-between, but you must be careful not to make the impression sluggish with it. Whether you’ve managed exaggeration should always be considered.

12. Facial (expressions)

Facial means adding expressions. Even in Smash Ultimate, where characters are small and hard to see, expressions are made fairly seriously. Since it’s a work that doesn’t need lip-sync or subtle expression changes, you want to move them energetically and big overall.

In development, before making the model, the direction is decided by retouching art. Submitted art often has small expression changes, but Sakurai puts his hand in and directs moving them bigger. Using Wolf’s damage expression as an example, “judging even from the jaw’s structure, this much intensity is appropriate” — making it bigger from where you thought you’d worked hard is good.

There are snaggy points too — show the upper and lower whites when opening the eyes wide, don’t mesh the upper and lower front teeth — but these change entirely by the game’s aim. Precisely because CG tends to lose information, exaggerating is just right. What you do differs for realistic and comical types, but you’d want to put expressions on big. Incidentally, robot-type expressions, or Steve’s blink-only facial, are fine facials too.

13. Flip motion

When two Links face each other, one shows the belly, one the back. When two Marios face each other, both show the belly — this latter is flip motion. A technique that mirrors poses and movements left-right; with seemingly no official name, it’s called this within Sakurai’s development.

Street Fighter II has characters as pixel art (sprites), so left-right flipped art is used as-is. Street Fighter IV is polygons, but following II, it properly flips left-right. This takes quite a bit of effort: a right straight becomes a left straight, and you need to swiftly swap an item’s holding hand. It’s a technique hard to apply to left-right-asymmetric fighters, like a sword-wielder.

Why do it? Of course, because it looks good. Made normally, the person on the right shows their back, the expression invisible, low in information. If they’re going to, facing front is better. Up to Smash Brawl there was no flip motion, so you could only see things like running from the side — and the side doesn’t show the face, which isn’t fun. Half the movement ends up hiding the face.

As an aside, the reason Street Fighter II’s “Dee Jay” wears shorts reading “MAXIMUM” is that it’s a letter combination that’s fine when flipped left-right (mirrored). Kirby’s “Maxim Tomato” can also be made instantly by left-right flip, which is why it’s an M design.

14. Motion blend

Motion blend is a technique that combines two or more motions. The cancel frame touched on in the attack-variation topic is an example. From a certain opening of an attack onward are cancel frames, and operating here makes it not suddenly snap into action but rather a pose where the operated movement and the movement from the cancel frame onward are combined.

The instant you enter walking from idle, and the instant you enter walking from an attack. Though the walk motion is one, the starting pose looks quite different. This is the effect of motion blend. You can specify the degree of application and its time, but more in-between motion reduces sharpness and is prone to breakage, so it’s done swiftly without much time. A shield-stance tilt shift is also a cross of the motion circling the outer edge and the motion fixed at the center.

In snappy Smash the impact isn’t that big, but motion blend rather comes alive in 3D games. Separate control of upper and lower body, footwork matched to terrain, reflecting orientation and incline — it gets so complex you can’t tell how much is blended. It’s not enough for movement to merely be alive, nor to technically fill it without breakage; an endless research probably continues somewhere.

15. Finger expression

When holding something, finger expression is a snaggy point. When gripping something, gripping perpendicular to the hand is natural, but when matching forearm movement, it goes diagonal to the hand. Making the grip natural using finger angles matters.

But fingers are very complex for their small size, and considering processing load, they’re also fated to be cut first. The model doesn’t vanish; they may look like they’re moving but be frozen, or have a few finger patterns assigned. He’s sometimes felt the number of movable fingers tells the story of that hardware’s performance.

Either way, making a game always entails anticipating cost-effectiveness. If fingers had many hierarchical structures constantly computed, it’d be quite a loss. It depends on hardware and aim, but cutting where you can cut and routing it to processing the important parts is always considered.

16. Always be conscious of hitboxes

Since Smash is seen from the side, the side silhouette is emphasized. As a premise, the hitbox against the opponent isn’t attached directly to the shape of a sword or fist, but placed to cover a wider range. Smash makes intense, fast coordinate moves, so hitboxes are on the larger side. Otherwise attacks wouldn’t land, making a stressful game.

So a common occurrence at the motion-creation site is the problem that an attack doesn’t pass through the center line and is hard to hit with. Of course they have it fixed. The planner side can place the hitbox to make it hit, but then look and hitbox diverge further. Some divergence of look and hitbox size can be accepted, prioritizing fun, but if the motion’s position also differs, the breakage grows large.

Further, the time the hitbox is out, and the pose at a likely clean hit, also matter. This instant has hit-stop applied and stays in the eye, and on a finishing blow the screen crisply freezes. A pose with weak attack feel there is no good. When swinging through a sword, the hitbox is sometimes called at a slightly swung-through position. Even if the planner does the strict game tuning, motion greatly affects feel, so the motion side should work conscious of the game itself.

17. The limits of the skeleton

When moving a CG character, it’s convenient to set up a skeleton (bone structure). For ordinary moving objects, the base moving moves the tip — shoulder to elbow, elbow to wrist, deciding the wrist position; this is FK (forward kinematics). Conversely, pinching and moving the hand position decides the shoulder and elbow angles; this is IK (inverse kinematics). Convenient for editing and easy to express ground contact — a basic technique with a long history.

About that skeleton, Sakurai has some thoughts. A joint’s range of motion is naturally 180°, but “can’t it bend further?” He wants to bend past straight to make animation livelier. A skeleton that simply bends the reverse way would increase everyday unwieldiness, but precisely because it’s CG, sometimes you want to exaggerate beyond logic.

Making it multi-jointed or increasing structures expands what you can do, but there’s processing load and it also takes a toon-anime-like sense, so Smash doesn’t have it set up that far — one of the issues proposed early in development but not realized. An existing convenient system is convenient when it fits, but what spills out of it is hard to make, and it draws a line of limits in your head. Creations are free, so he wants to exceed limits with various ideas.

18. Rotate on two axes

Try rotating the neck sideways. Doesn’t it look mechanical? — Sakurai asks. Rotation is expressed by the X, Y, and Z axes, but rotating on the Y axis alone just because it’s rotation around the vertical axis is a no. When making human movement, you’d want to be conscious of 2-axis rotation.

For example, to turn the neck sideways: look down and turn (X axis + Y axis), or tilt sideways and turn (Z axis + Y axis). The wrist, too, looks unsatisfying like a drill with simple 1-axis rotation, but just tilting a little before turning makes it plausible. Originally you move the linked parts too; a twisting action links the shoulder and collarbone, and an action like opening a motorbike’s throttle also uses wrist movement.

Rotation speed looks plausible even left to a 3D tool’s function curve, but shifting the timing of the two axes a little makes it even better. Even with the same end point of rotation, you can give it an entirely different character via the process. Even the same “turn the head” looks different from below, from above, or dead sideways. Note that when moving the neck, also be conscious of the gaze’s movement. With the gaze firmly added, sometimes it holds up without changing orientation that much.

Summary

We’ve now gone through both videos and 18 topics of the “Motion” category. Finally, let’s recap the key points of each part (= each source summary video) by individual topic.

Part 1: Overdo for just right, the 4 parts of an attack, flat scaling (#01–#09)

Part 2: Damage expression, flip, motion blend, 2-axis rotation (#10–#18)

There’s a lot here, but I hope you’ll revisit it starting from whichever part caught your interest. Please also check out the related categories.