Visually evoked affective states serve as a representative method for investigating human emotion, which is recognized as a foundational substrate for the coordination between behavioral and physiological systems. Prior research has endeavored to establish correlations between subjective affective ratings of visual stimuli and corresponding hemodynamic responses in the human brain, thereby exploring the brain's responses to affective scenes. Existing findings indicate the presence of distributed emotional representations throughout various regions of the cerebral cortex and cortical affective maps that connect categorical and dimensional affective spaces. Nevertheless, these studies are constrained by (1) the exclusion of subcortical and cerebellar regions from analyses, (2) an insufficient understanding of how different brain regions are interconnected in representing diverse emotions, and a lack of explainability methodologies to validate brain decoding outcomes. In the present study, we have devised an affective decoding framework. An atlas encompassing all cortical, subcortical, and cerebellar regions was utilized to incorporate the entire brain into the analysis. By employing a linear decoding approach, we decomposed participants' complex emotional representations into 34 discrete categories and examined activation patterns across brain regions for each category. Ultimately, leveraging generative AI models, we conducted an additional analysis to elucidate the decoding results and thereby enhance interpretability.