Home▸Computer Science▸Diffusion Model Visualizer (2D)

Diffusion Model Visualizer (2D)

2D canvas companion to the diffusion-model visualizer: a cloud of dots denoises step by step into a circle, ring or spiral, with steps, noise strength and denoise speed tunable in real time.

Computer Science2DModerate60 FPS📱 Mobile-adapted⇄ 3D version

This 2D companion drives the same denoising update rule as the 3D diffusion-model visualizer on a plain canvas: a cloud of dots starts scattered at random and, step by step, is pulled toward a position on a target outline — circle, ring or spiral — while a shrinking amount of random jitter keeps it from snapping there instantly, making the "predict a little noise, subtract it, repeat" loop that trained diffusion models run directly visible.

⚙ Under the hood

2D canvas denoising visualizer: a particle cloud is pulled toward a target outline each step while noise shrinks over time, following x(t−1) = x(t) − η·(x(t) − target) + σ(t)·ε.

diffusion modelsgenerative aimachine learningdenoising

2D · HTML5 Canvas 2D · 60 FPS target · runs fully client-side, no install

What does this 2D diffusion visualizer show?

A cloud of dots starting at random positions on a flat canvas, pulled step by step toward a target outline (circle, ring or spiral) while a shrinking amount of random jitter keeps them from snapping into place instantly — the same predict-and-subtract loop that trained diffusion models like Stable Diffusion use to turn noise into images.

Is this the same simulation as the 3D version?

It runs the same denoising update rule, x(t-1) = x(t) - eta*(x(t) - target) + sigma(t)*noise, but in a flat 2D canvas instead of a rotating 3D point cloud, so the same steps/noise/speed controls are easier to read against a fixed 2D outline.