Open-Source Artificial Intelligence Avatars - Technologies, Architectures, and Multimodal Language Integration

The creation of digital avatars based on artificial intelligence (AI) has gained relevance across multiple domains, particularly in the public sector, where the automation of communication with citizens requires transparent, secure, and ef-ficient solutions. This paper presents a critical analysis of the available open-source tools for the development of AI avatars, combining a literature review with an experimental evaluation of practical solutions. Image generation (Fooo-cus), facial animation (SadTalker, GAIA), and voice synthesis (Coqui TTS, OpenVoice) tools are discussed, considering parameters such as latency, visual quality, lip synchronization, hardware requirements, and license compatibility. The analysis emphasizes the applicability of these technologies in Portuguese public services, including municipal portals and gov.pt platforms, highlighting the potential for integration in institutional contexts and the challenges related to digital sovereignty, privacy, and accessibility. As proof-of-concept we pro-vide a prototype to encourage future research in this field, the code and pre-trained models are available in a public GitHub repository.