Diffusion Model Visualizer

Generative image and video models like DALL·E, Midjourney and Stable Diffusion don't paint an image in one shot — they learn to reverse a noising process, starting from static and repeatedly predicting a small amount of noise to subtract until a coherent picture emerges. This simulator visualizes that reverse loop directly with a cloud of particles: each one begins scattered at random and, step by step, is pulled toward a position on a target form while a shrinking amount of random jitter keeps it from snapping there instantly. Tune the number of diffusion steps, the initial noise strength and the denoising speed to see how each shapes the path from chaos to structure, and switch the target shape to change what the "model" is generating toward.