Seamless is a family of open-weights generative AI models developed by Meta's FAIR group for multilingual speech and text translation. The family includes SeamlessM4T v2, SeamlessExpressive, SeamlessStreaming, and unified Seamless models that support real-time translation across multiple modalities while preserving vocal style and prosody. These models cover up to 100 languages and employ unified architectures to reduce error propagation compared to cascaded translation systems.