The main advantage of this technique over conventional spatial anti-aliasing is that it does not require large amounts of computing power. It achieves this by smoothing undesirable jagged edges ("jaggies") as pixels, according to how they appear on-screen, rather than analyzing the 3D model itself, as in conventional spatial anti-aliasing. Since it is not based on the actual geometry, it will smooth not only edges between triangles, but also edges inside alpha-blended textures, or those resulting from pixel shader effects, which are immune to the effects of multisample anti-aliasing (MSAA).
The processes of FXAA are listed as follows:
Finding edges is typically a depth-aware search, so that pixels which are close in depth are not affected. This helps to reduce blurring in textures, since edges in a texture have similar depths.
Smoothing is applied as a per-pixel effect. That is, there is no explicit representation of the edges. Rather, the first step is a depth-aware edge filter, which marks pixels as belonging to edges, and the second step filters the color image values based on the degree to which a pixel is marked as an edge.