Last Ninja 2: The Isometric Peak

There is a moment in Central Park, early in Last Ninja 2, where you have to cross water on stepping stones. You can see the stones. You can see your ninja. You press the joystick diagonally, hold fire, and he leaps in an arc that looks completely convincing right up until he lands in the river and the screen goes about its business without you. You try again. You land in the river. You try again with your face two feet closer to the television, as though depth perception is something you can improve by leaning.

That stone crossing has been costing people afternoons since 1988. It is also, I think, the most honest thing in the game — a moment where Last Ninja 2 stops pretending its perspective is a window and admits it is a grid.

Manhattan, which is the joke

The first Last Ninja, in 1987, was a straightforward pitch: feudal Japan, a sword, revenge on the shogun Kunitoki, presented in a diagonal projection that made a C64 look like a machine from a class above. It sold enormously. System 3 had a sequel to make, and the obvious move was more Japan.

They went to New York instead. Kunitoki drags Armakuni across time to present-day Manhattan, and the game runs through Central Park, the streets, the sewers, a mansion, an office block and a basement. A man in a gi hopping between skips in an alley behind an office building is a genuinely strange image, and 1988 played it entirely straight. The strangeness works. Feudal Japan is a setting that has been storyboarded to death; a ninja in an American office block in the late eighties, drawn by Hugh Riley in the C64’s chunky multicolour, has nothing to be compared against. The environments get to be specific — you recognise a park bench, a manhole, a filing cabinet — and specificity is what the isometric view is best at. It is a perspective built for furniture.

John Twiddy wrote the code, and the engine is doing more than the first game’s: larger scenes, more screen-to-screen variety, and a fight system with several weapons that switches cleanly enough to survive contact with a joystick. But the reason people still talk about Last Ninja 2 has more than one part, and only one of them is programming.

Matt Gray

The Central Park theme is one of the best pieces of music ever written for the SID chip, and it is unusual in the C64 canon because of what it declines to do. The house style of the machine’s most celebrated composers is maximal: Rob Hubbard filling every gap, arpeggios doing the work of chords, the three voices worked so hard they sound like six. Matt Gray’s Last Ninja 2 score is patient. It has space in it. The bassline sits back, there is actual room around the lead, and the whole thing sounds like it was written for a real instrument and then translated, rather than composed at the limits of the hardware.

That restraint is the point. Delta, which I wrote about as the C64 shooter as music video, uses Hubbard’s soundtrack as an engine — the music drives the game forward, and the game is arranged around it. Last Ninja 2 uses Gray’s music as an environment. It is atmospheric in the literal sense: it makes the air in Central Park feel like a specific time of day. Games had used music to signal danger and to reward progress for years. Using it to make a place feel like it exists when you are standing still was newer than it now sounds.

Monty on the Run, which I looked at in the music that made it, is the counter-example — a game whose soundtrack is so good it papers over the design underneath. Last Ninja 2’s music does not paper over anything. It sits on top of a game that has real ideas and one enormous, deliberate, unresolved problem.

The problem is the projection

An isometric view is a lie told with mathematics. The C64 draws a two-dimensional image, and your brain reconstructs three dimensions from it using rules that work perfectly for static objects and catastrophically for moving ones. A ninja standing on a stone is unambiguous. A ninja in mid-air above the same stone occupies a screen position that could correspond to a dozen different points in the fictional space, because height and depth share the same axis on screen. Move up-and-back and move up-and-up produce identical pixels.

Every isometric game of the era hit this wall. Knight Lore hit it in 1984 and Ultimate’s answer was to make the jumps short and the platforms forgiving. Head Over Heels hit it in 1987 and Jon Ritman’s answer was a shadow — a small marker on the ground beneath the character telling you exactly where you are in plan view, which resolves the ambiguity completely and costs almost nothing to draw.

Last Ninja 2 hit it and did nothing. There is no shadow. There is no landing indicator. There is a jump with a fixed arc and a tile grid you have to read by eye, and if you misjudge by one tile you are in the water. System 3 knew — the first game had the same complaint filed against it in every review — and shipped the sequel with the jump intact.

I have gone back and forth on whether this is a failure. The case for the defence is that Last Ninja 2 is a game about precision under pressure, and the jump makes movement itself a skill rather than a formality. Once you internalise the arc — and you can, it is entirely learnable — the stones stop being a lottery and start being a rhythm. The game is asking you to learn its projection the way a fighting game asks you to learn frame data. That is a real design position and the game holds it consistently.

The case for the prosecution is Ritman’s shadow. It existed. It was in a game on the shelf a year earlier. The information cost nothing and would have removed the arbitrariness while leaving the skill entirely intact — you would still have to time and aim, you would simply be able to see where you were. Refusing free clarity is a choice, and it is the choice that costs Last Ninja 2 its claim on being the best isometric game of the eighties rather than the most impressive.

What it got right that nobody copied

The thing worth stealing from Last Ninja 2 is its inventory. You pick up objects, you carry them, and the game almost never tells you what they are for. There is no highlighted interaction prompt and no character helpfully observing that this looks important. A screen contains a thing; several screens later, a situation contains a hole shaped like that thing.

This is adventure-game logic wearing an action game’s clothes, and it survives because the environments are drawn with enough specificity that objects read as objects rather than as puzzle tokens. Modern games have almost entirely abandoned it in favour of legibility, which is usually the right call and occasionally a loss. The satisfaction of realising what the thing in your pocket is for, three rooms after you found it, with no assistance, has a different texture to the satisfaction of executing a plan the game explained to you.

The sequel problem, solved sideways

One more thing Last Ninja 2 does that deserves credit: it understood what a sequel is for. The 1987 original had proved the technology and exhausted the fiction — there is only so much feudal revenge a diagonal projection can carry. System 3’s response was to keep the engine, keep the protagonist, and replace the entire world.

That is a harder trick than it looks, because a setting change breaks the visual shorthand a game has taught you. Everything Armakuni learned about reading a bamboo grove is useless in a sewer. The engine survives the move because the isometric view is fundamentally an inventory of solid objects, and Manhattan has more of those than a temple garden does. The projection got busier and more legible at the same time, which is why the sequel looks better than its predecessor while running on the same fundamentals.

The games industry spent the following thirty-five years mostly doing the opposite — iterating settings while keeping the fiction and expanding the mechanics. Last Ninja 2 kept its verbs and threw away its century. Both of the sequels that followed, in 1991 and beyond, walked that back towards Japan, and neither is remembered the way this one is.

Where to play it

The C64 original is the one that matters — the Amiga and Spectrum conversions exist and neither has Gray’s SID score doing what it does. It is available legitimately through System 3’s various re-releases and compilations, and it runs fine in VICE. Give the stepping stones twenty minutes before you decide the game is broken. Somewhere in the third or fourth attempt the grid resolves, the arc becomes a thing you can feel rather than guess, and Last Ninja 2 turns into what it always was underneath: a beautifully specific place, scored better than it had any right to be, that requires you to learn how to look at it.