In this work, we address the problem of amodal content completion and propose a model that reconstructs the content of the occluded region of a partially visible object by explicitly attending to its visible pixels. Our method utilizes a weighted mask that guides the network to prioritize features from the visible portions of the occluded instance, and series of enhanced gated convolution layers that preserve and reinforce this guidance throughout the encoder. We train our model on the COCOA dataset and a subset containing only animal instances by implementing a self-supervised method to synthesize occlusions in the training samples. The experiments show that our model generates richer semantic and textural details than existing baselines, achieving competitive or superior performance across multiple quantitative metrics.