A multimodal large language model trained specifically for facial image understanding. Project page: https://www.idiap.ch/paper/facellm