IndexTTS2 is a next-generation text-to-speech model developed by Bilibili, officially open-sourced on September 8, 2025. The model achieves major breakthroughs in emotional expression and duration control, being hailed by the community as 'the most realistic and expressive TTS model.'