Shapiro, Smith and Knight (2007) showed that most lightness illusions can be generated from the input images by removing their low spatial frequencies content. They suggested that the visual system eliminates spatial frequencies lower than the fundamental frequency of the area of interest. Here, we extend this account in three ways: 1. We show that removing the low spatial frequency energy is algebraically very similar to many existing contrast models (i.e., contrast computed over a region, followed by a two-branch non-linearity). With appropriate specifications, contrast models can account for many well-known lightness illusions, including Adelson's snake and checker-shadow illusions, the white shadow illusion, Anderson and Winawer illusions, Bressan's Dungeon illusion, gradient-gradient illusions, and test spots placed on natural images. 2. We present lightness demonstrations that cannot be explained by an anchoring rule that ascribes "White" to the highest global luminance, but can be explained by anchoring "White" to the maximum output of an array of contrast filters. The demonstrations are based on Adelson's checker-shadow illusion with the cylinder removed (see www.shapirolab.net). We show that when the luminance of the square in the shadow is increased, the square is perceived as white even when other lights in the global environment have higher luminance levels. 3. We present demonstrations that contain test areas of different spatial scales; a model based on a single cut-off spatial frequency cannot account for lightness variations at all spatial scales simultaneously We therefore propose a model with tunable high-pass cut-off or in which contrast is calculated over a tunable integration area. We discuss whether the filters adjust to local organizational factors (edges, grouping, scission and attention factors) or whether the model can be driven by a process that favors the spatial frequency channel that gives the maximum response within a local region (a winner-take-all model).