The first image retrieval systems hold the advantage of being fully automatic, and thus scalable to large collections of images but are restricted to the representation of low-level aspects (e.g. colors, textures...) without considering the semantic content of images. This obviously compromises interaction, making it difficult for a user to query with precision. The growing need for 'intelligent' systems, i.e. being capable of bridging this semantic gap, leads to new architectures combining multiple characterizations of the image content. This paper presents SIR1, a promising high-level framework featuring semantics, signal color and spatial characterizations. It features a fully-textual query module based on a language manipulating both boolean and quantification operators, therefore making it possible for a user to request elaborate image scenes such as a "covered(mostly grey) sky" or "people in front of a building".