Diffusion Model Visualizer

Diffusion Model

Diffusion steps 60

Noise strength 0.35

Denoise speed 1.0×

Step

–

Noise σ

–

Fidelity

–

Generations

–

How it works

A real diffusion model (DALL·E 2, Stable Diffusion) learns to reverse a gradual noising process: starting from pure noise, it repeatedly predicts and subtracts a little bit of noise until a coherent image emerges. This visualizer follows the same shape with a toy rule instead of a trained network — every particle starts at a random point and, each step, moves partway toward its assigned position on the target form while a shrinking amount of random jitter is mixed back in, so the cloud visibly "denoises" from chaos into structure.

x(t−1) = x(t) − η·(x(t) − target) + σ(t)·ε
σ(t) = σ0 · (1 − t / T)

Diffusion steps (T) — how many denoising steps the generation takes; more steps trace a smoother path from noise to form, mirroring the quality/speed trade-off real samplers make.
Noise strength (σ0) — how much random jitter ε is mixed in early on; higher values keep the cloud chaotic for longer before it resolves.
Denoise speed — playback rate of the step loop, so you can slow down and watch a single step's pull-toward-target + jitter clearly.
Target shape — the "prompt": the fixed point cloud (torus / sphere / spiral) every particle is guided toward, standing in for whatever a trained model would predict.

Real-world relevance: this same predict-and-subtract loop, run thousands of times on trained noise predictors, is what actually powers today's image, video and audio generators — the toy version here keeps the geometry but swaps the neural network for a fixed target.

Denoised particle High-noise particle

⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.