The paper presents a methodology to derive seismic vulnerability of existing reinforced concrete buildings by using mechanical models derived from data obtained by transfer learning technique. In detail, the methodology aims to exploit the combination of some structural building features obtained from the analysis of photographs within the VULMA framework, and other general information obtained from other sources, such as year of construction, localization and covered area. Joining and matching the above data, the input data for simulating an ideal mechanical model are provided, which is elaborated and analysed under seismic actions. The method has been firstly presented and after applied to a couple of existing buildings, with the aim to derive seismic vulnerability. The results of the analyses show how the methodology can be applied in a view of defining risk mitigation strategies in a focused area, relying on two automatization processes.