How We Built Sushi Ninja With AI: When a Slice Game Became a Recipe Game

The brief was a Fruit Ninja clone with a sushi theme. The AI turned it into a sushi assembly game with nine recipes, four unlockable blades, and ingredient physics. We almost simplified it back. We're glad we didn't.

March 17, 2026 · Vibe Arcade

Want to play first, then read the build story?

▶ Play Sushi Ninja Now

The original concept for Sushi Ninja was simple: slice flying ingredients, score points, avoid bad items. A Fruit Ninja-style game with a Japanese aesthetic and a sushi theme instead of tropical fruit. Fast, tactile, visually satisfying. Build it in an afternoon.

The AI's first interpretation introduced something we hadn't planned: the sliced ingredients needed somewhere to land. It generated a row of sushi plates at the bottom of the screen, with each plate tracking which ingredients had been added to it. The clear implication was that scoring would be tied not just to slicing, but to what you sliced onto which plate. We were one prompt away from having the AI build a recipe assembly system into what was supposed to be a pure reflex game. We made that prompt.

The Mechanic That Changed the Design

The recipe system emerged from asking the AI to make the plate mechanic "meaningful." The response was to give each plate a target recipe — a specific combination of ingredients that, when correctly assembled, scored significant points. Incomplete plates scored nothing. Incorrect combinations wasted ingredients that had been sliced onto the wrong plate.

This transformed the game's core tension. Fruit Ninja is about fast, careless slicing — you want to hit everything. Sushi Ninja required you to be selective. Not every falling ingredient should be sliced. Slicing the right ingredient onto the wrong plate wastes it. Slicing a hazard ingredient like Spoiled Fish costs you a life. Suddenly the game was asking players to make decisions rather than just react.

The recipes themselves were detailed: specific combinations of rice, fish, vegetables, and optional garnishes like Tobiko fish roe. The AI wrote them with actual Japanese sushi knowledge — the ingredients for a California Roll, a Tuna Roll, a Dragon Roll — not just random combinations with Japanese-sounding names. The Tobiko was designated as an optional ingredient that added a score bonus, which created a meaningful choice: take the extra second to add it and earn more points, or skip it and keep the pace up.

We kept all of it. The recipe system was the best accident in the game's development.

Building Ingredient Physics on a Budget

A slicing game needs satisfying physics. Ingredients need to arc through the air in a way that looks natural but is readable enough for players to track and intercept them. Too chaotic and it's frustrating. Too predictable and it's boring.

The AI's first physics implementation used basic parabolic arcs — ingredients launched from the bottom or sides with randomized initial velocity and angle, following a gravity curve to wherever they landed or fell off screen. It worked, but it was immediately apparent that the randomization range was too wide. Ingredients would occasionally fly off so fast they were impossible to intercept, or arc so slowly they felt like they were floating.

The fix was constraining the velocity distribution: narrow the range of initial speeds so that the slowest and fastest ingredients weren't too far apart. This made the game feel more consistent without making it feel scripted. The AI implemented it as parameterized values — minimum and maximum launch speed, gravity coefficient, angle variance — which made it easy to tune. We spent more time adjusting these parameters than on any other single element of the game.

The slice detection itself was handled with a gesture trail — as the player drags a finger or cursor across the screen, the game checks for intersections between the trail and ingredient hitboxes. The AI's first implementation drew the full gesture trail on screen as a visible line, which looked good on desktop and less good on mobile (the trail sometimes obscured the ingredients themselves). The final version draws only a short trailing segment that fades quickly — enough to provide visual feedback for the slice without cluttering the screen.

Blades: Cosmetic Depth That Isn't Just Cosmetic

We asked for unlockable blades assuming they'd be purely cosmetic — different visual trails, different slice effect animations. The AI designed four blades but made each one mechanically distinct:

Shoshinsha Hōchō (starter) — standard slice, no modifications
Yanagiba — unlocked at 5,000 cumulative points. Wider slice trail, scores additional points per cut. The visual upgrade makes it feel like a meaningful reward.
Deba Bōchō — unlocked after completing 10 Chef's Special orders. Has a 25% chance to triple-slice, hitting up to three ingredients with a single swipe. High skill ceiling.
Katana — unlocked at 50,000 cumulative lifetime points. Can hit multiple ingredients across the screen in one stroke, fundamentally changing how the top end of the game plays.

The actual Japanese names were accurate — the AI used real Japanese kitchen knife terminology rather than invented names. The Yanagiba is a real knife used specifically for slicing sashimi. The Deba is used for breaking down fish. That kind of authentic detail wasn't something we specified; it was something the AI included because it had the domain knowledge to do so.

The Chef's Special unlock condition for the Deba Bōchō was an interesting design call. It tied blade progression to completing premium orders rather than just accumulating points — which means players who master the recipe system get rewarded with a mechanical upgrade. That's good game design. It creates a positive loop where the recipe complexity pays off in a concrete way.

Hazards: What Not to Slice

We introduced Spoiled Fish and Pufferfish as hazard items relatively late in development, after testing sessions revealed the game was too forgiving. When everything that falls is safe to slice, the game becomes a pure speed challenge — slice everything as fast as possible. Hazards force players to make distinctions, which is more interesting than raw speed.

Spoiled Fish was straightforward — a recognizable visual cue (greenish tint, visible degradation) that players learn to avoid. The cost is a life, which is punishing enough to matter without being unfair. The AI initially made the Spoiled Fish look too similar to regular fish during fast sections, so we asked for a more exaggerated visual distinction. The updated version added a small visual indicator above it, which solved the readability problem without making it trivially obvious.

Pufferfish was more interesting as a design problem. It shouldn't cost a life — that would make it feel arbitrary and frustrating. Instead, slicing a Pufferfish causes it to explode and clear your active plate. You lose your recipe progress for that plate but keep your lives. It's a setback rather than a punishment — and it creates a moment of genuine tension when a Pufferfish appears near a plate you're close to completing.

Power-Ups and the Six-Item Rule

The six power-ups in Sushi Ninja (Zen Time, Master's Plates, Double Points, Ingredient Magnet, Tsunami, Auto-Roll) were designed as a set, not added one at a time. We asked for power-ups and got a complete system in one response.

Each one addresses a specific player pain point:

Zen Time slows ingredients when the screen gets overwhelming — a mercy mechanic for difficult sections.
Master's Plates enlarges the target plates, making ingredient placement more forgiving — useful when managing multiple complex orders simultaneously.
Ingredient Magnet pulls sliced pieces toward the nearest plate — a precision assist that lets players focus on slicing rather than aim.
Tsunami clears the screen with a wave animation — a panic button for when plates get hopelessly mixed up.
Auto-Roll temporarily automates the ingredient placement — a "catch up" mechanic that gives players breathing room.

The balance question was whether the power-ups were too accessible. If players can Tsunami their way out of any bad situation, the skill of managing multiple plates simultaneously loses meaning. We asked the AI to make power-ups spawn based on completed recipes rather than on a timer — that tied power-up access to performance, which kept the skill floor meaningful.

What We Learned From Building It

Sushi Ninja is the most mechanically layered game in the Vibe Arcade collection. It started from a concept — sushi slicer — that seemed simple and ended up with interlocking systems: recipe tracking, blade progression, hazard avoidance, power-up management, and a scoring structure that rewards both speed and precision.

Almost none of that was in the original brief. The AI filled in the design space between "slicing game" and "sushi theme" with ideas that were consistent with the concept, mechanically sound, and genuinely interesting to play. Our role was less about originating ideas and more about evaluating which ideas to keep, which to adjust, and which to remove.

That's the vibe coding pattern at its most honest: you bring the concept and the taste, the AI brings the execution and often the ideas. When it works, the result is something neither party would have built alone.

Play Sushi Ninja — try to unlock the Katana, and see if you can complete a Chef's Special with Tobiko on the first attempt.