Time Blindness: Why Video-Language Models Can't See What Humans Can?

Authors

Ujjwal Upadhyay^* Mukul Ranjan^* Zhiqiang Shen Mohamed Elhoseiny

^*Equal Contribution

Explore how noise animation systems create dynamic visual effects through depth maps and sophisticated masking techniques.

🎨 Try the Animation Studio

Examples from Our Collection

📝 Word Animations

Text-based animations where noise patterns create dynamic visual effects around words and typography.

🔷 Shape Animations

Geometric shapes interacting with animated noise to create mesmerizing patterns and depth effects.

🖼️ Image Animations

Complex images enhanced with depth-based noise animation to simulate motion parallax.

🎬 Video Effects

Advanced video processing with dynamic noise overlays and depth-based animation control.

🔧 Basic Concept

The system creates animated noise patterns that can interact with text, shapes, or images. When depth maps are used, the noise animation speed varies based on the brightness of each pixel in the depth map.

📊

Noise Generation

Binary noise patterns are generated for both foreground and background

🔄

Animation

Noise patterns scroll in opposite directions

🎭

Masking

Content (text/shape/image) creates a mask to separate foreground and background

⚙️ Two Operating Modes

Standard Mode

Without depth map, the system uses a binary mask:

Text, shape, or image creates a mask
Foreground noise scrolls in one direction
Background noise scrolls in opposite direction
Creates visual separation between content and background

// Standard mode (no depth map)
const isFG = maskData && maskData[i+3] > 0;
let val;
if (isFG) {
  val = foregroundNoise[fgIdx];
} else {
  val = backgroundNoise[bgIdx];
}
            

Depth Map Mode

With depth map enabled:

Single noise layer is used
Depth map controls animation speed at each pixel
Brighter areas in depth map = faster animation
Creates illusion of depth through motion parallax

// Depth map mode
if (depthMapData) {
  const d = depthMapData[i];
  const depthVal = d / 255;
  speedFactor = 1 + depthVal * depthScale;
}
const newY = Math.floor((y + backgroundOffset*speedFactor) % h);
            

🔑 Key Technical Components

Noise Generation

Binary noise patterns (black/white pixels) are generated with controllable density and speckle size. These patterns are scrolled to create animation.

Content Masking

In standard mode, text, shapes, or images create a binary mask that determines which noise pattern (foreground or background) is displayed at each pixel.

Depth-Based Animation

When depth map is enabled, the brightness of each pixel in the depth map controls how fast the noise animates at that position, creating a parallax effect.

Background Color Option

The "Remove Background Noise" option replaces noise with a solid color in background areas or in areas below a certain depth threshold.