Gaussian blur is a digital image processing technique that softens an image by averaging each pixel's colour value with those of its neighbours, weighted according to a bell-shaped statistical distribution known as a Gaussian function. The result is a smooth, natural-looking softening effect that closely resembles the out-of-focus appearance produced by an optical lens. The technique is named after the mathematician Carl Friedrich Gauss, whose work on normal distribution underpins the mathematical formula that determines how each surrounding pixel contributes to the blurred result.
The Gaussian blur works by passing a kernel ( a grid of numerical weights based on the Gaussian curve ) over every pixel in an image. Pixels closer to the centre of the kernel contribute more to the final averaged value than those at the edges, which is why the resulting blur looks smooth and graduated rather than harsh or blocky. The size of the kernel, known as the radius, determines the degree of softening: a small radius produces subtle softening, while a large radius creates a heavy, dreamlike blur.
In post-production, Gaussian blur is used for a wide range of purposes: smoothing skin in portrait retouching, creating depth-of-field effects when footage was captured without sufficient lens blur, obscuring sensitive information such as faces or licence plates, and generating soft glow effects by blending a blurred layer over the original image. It is also a fundamental step in many compositing workflows, particularly when integrating digitally generated elements into live footage to reduce the sharpness mismatch that makes CGI look artificial.
In AI image and video generation, Gaussian blur appears in multiple forms. Diffusion models use blur-like noise processes during training, and many AI tools offer Gaussian blur as a post-processing adjustment within their pipelines. Understanding Gaussian blur helps creators prompt for intentional softness, atmospheric haze, or cinematic focus effects, and makes it easier to refine AI outputs through standard compositing software when the generated image requires further polish.
