Images and music share deep and mysterious links that drive audio-visual playlist composition. In this paper we first explore the possible descriptors for each medium, and their difference of nature that makes the mixing of both media a difficult operation. In order to fill the gap between both worlds, we then propose a double method for indexing music titles and pictures on a common mood basis. The evocation process that triggers a picture when a music title is played or conversely that calls for a music when looking at a picture is modeled through a visual map metaphor that blends picture semantics and music moods. Then music moods are propagated to the pictures. The results of this indexing method are assessed and different improvements are proposed.