Gojo

Gojo is a Jujutsu Kaisen cursed technique visualizer built with MediaPipe Hands and Three.js. This version focuses on smoother animation, lower render cost, and significantly more stable hand tracking.

Techniques

• Secret Technique: Hollow Purple
  • Trigger: Pinch (thumb + index)
• Domain Expansion: Malevolent Shrine
  • Trigger: All 4 fingers up (flat/open hand)
• Domain Expansion: Infinite Void
  • Trigger: Index + middle up only
• Cursed Technique Reversal: Red
  • Trigger: Index only

Optimization Update

Performance

• Particle count reduced from 20,000 -> 12,000 for smoother FPS while preserving visual density.
• Particle generators now return flat arrays instead of object allocations, reducing per-frame GC pressure.
• Optional low-end mode: hand detection can be throttled to every other frame to cut MediaPipe compute cost.
• Only the first detected hand is processed for more stable and less jittery gesture reads.
• Detection confidence and tracking confidence increased to 0.75 / 0.75 to reject noisy frames.

Smoother Animation

• Screen shake now uses lerp easing instead of instant random jumps.
• Rotation transitions ease between techniques instead of snapping.
• Particle lerp factor tuned to 0.085 (from 0.1) for silkier transitions.
• Technique label color and glow now transition smoothly with CSS transitions.

Hand Tracking Reliability Fixes

Root issues that caused unstable tracking:

• Single-frame detection directly changed technique state (flicker).
• Overlapping gesture rules caused random switching between techniques.
• Pinch threshold was too tight.

What was added:

• Vote buffer over the last 8 frames.
• Activation threshold: gesture must win at least 62% of votes.
• Hysteresis: leaving an active state requires a higher threshold (72%) to avoid wobble-induced drops.
• Mutually exclusive priority order in gesture checks:
  • Pinch -> Purple
  • All 4 fingers up -> Shrine
  • Index + Middle only -> Void
  • Index only -> Red
• Looser pinch threshold (0.055) for more reliable pinch detection.

Code Cleanup

• Technique visuals and labels consolidated into a central technique config.
• State switching is centralized through target updates instead of redundant resets.
• Gesture logic is now explicit and non-overlapping.

Getting Started

Prerequisites

• Modern browser (Chrome, Edge, or Firefox)
• Webcam access

Run Locally

1. Clone the repo:

   git clone https://github.com/abijithraja/Gojo.git
   cd Gojo

2. Open with Live Server in VS Code, or open index.html directly in a browser.

Credits

This project is inspired by and gives credit to the original SAT0RU repository:

• Original repository: https://github.com/reinesana/SAT0RU
• Original README reference: https://github.com/reinesana/SAT0RU/blob/main/README.md?plain=1

Built with MediaPipe Hands and Three.js.