#Aim assist

A production-quality aim-assist script with:

• FOV gate (only act on targets near the crosshair)
• Confidence gate (filter out low-confidence detections)
• Exponential smoothing (feels human, not robotic)
• Tunable strength
• All settings exposed as GUI controls

#Full script

//!gui:section label="Targeting" mode="expanded"
//!gui:slider var=strength label="Aim Strength" min=0 max=1 default=0.5 step=0.05
//!gui:slider var=fovRadius label="FOV Radius (px)" min=10 max=400 default=120
//!gui:slider var=minConfidence label="Min Confidence" min=0 max=1 default=0.6 step=0.05
//!gui:endsection

//!gui:section label="Smoothing" mode="expanded"
//!gui:slider var=smoothness label="Smoothness" min=0 max=1 default=0.6 step=0.05
//!gui:checkbox var=showFov label="Show FOV circle" default=1
//!gui:endsection

// --- Visualization ---
if (settings.showFov) {
  drawCircle(center_x, center_y, settings.fovRadius, "#ff7b85", 1);
}

// --- Gates ---
if (confidence < settings.minConfidence) {
  return (0, 0);
}

dx = detection.x - center_x;
dy = detection.y - center_y;
dist = hypot(dx, dy);

if (dist > settings.fovRadius) {
  return (0, 0);
}

// --- Initialize smoothing state ---
if (!isset(state.smoothX)) {
  state.smoothX = 0;
  state.smoothY = 0;
}

// --- Compute raw offset, scaled by strength ---
rawX = dx * settings.strength;
rawY = dy * settings.strength;

// --- Exponential smoothing ---
// Higher smoothness slider → slower response → more human feel
// Rate is normalized to dt so it stays consistent at any FPS
rate = (1.0 - settings.smoothness) * 30;
alpha = 1.0 - exp(-rate * dt / 1000);

state.smoothX = state.smoothX + (rawX - state.smoothX) * alpha;
state.smoothY = state.smoothY + (rawY - state.smoothY) * alpha;

// --- Debug monitor ---
monitor("dist", round(dist, 1));
monitor("conf", round(confidence, 2));

return (state.smoothX, state.smoothY);

#How it works

#Section grouping

The script opens with two //!gui:section blocks, which group related sliders into collapsible categories in the GUI. The first holds "Targeting" sliders, the second holds "Smoothing".

#FOV visualization

drawCircle() paints a circle on the preview window so the user can see the FOV gate. Toggling the Show FOV circle checkbox hides it.

#Gates

Two guards short-circuit the script when conditions aren't right:

• Confidence gate — bail out if confidence is below the slider threshold
• FOV gate — bail out if the detection is further than fovRadius pixels from the crosshair

Both return (0, 0), which produces no aim adjustment.

#Smoothing

The smoothing block is the part that makes the assist feel human instead of snapping. We maintain state.smoothX and state.smoothY across frames and update them each frame by a fraction of the raw error:

state.smoothX = state.smoothX + (rawX - state.smoothX) * alpha;

The alpha value is computed from dt (delta time) and the user's smoothness slider. This makes the effective response rate frame-rate independent — the assist behaves the same at 60 Hz capture or 240 Hz capture.

A smoothness of 0 means alpha ≈ 1 (instant, snappy) and a smoothness of 1 means alpha ≈ 0 (very slow, smooth).

#Debug monitor

Two monitor() calls expose live values to the monitor panel, so you can watch what the script is seeing while tuning.

#Tuning tips

#Extension ideas

• Soft falloff — instead of a hard FOV cutoff, scale strength by distance: strength * (1 - dist / fovRadius)
• Hold-to-aim — only run the script while a button is held: if (!makcu.right) { return (0, 0); }
• Pose-aware aiming — for YOLO-Pose models, target a specific keypoint: target = detection.keypoints[KP.NOSE]; (see Detection)
• Track velocity — predict where the target will be next frame: predX = detection.x + detection.velocityX * lead