A Survey on the Evolving Journey of Generative AI: From Language Models to Multimodal Intelligence

Generative AI (GenAI) has evolved quickly from basic probabilistic text-generation methods to large multimodal models able to generate text, images, audio and video.In this work, we provide a systematic review of GenAI along its progression timeline, analyzing key approaches and shifts in paradigm in addition to recent trends towards multimodal intelligent systems.For this purpose we review peer-reviewed literature, technical reports, and public model releases.We feature prominent architectures (Markov models, VAEs, GANs, TLLMs, and modern MLLMs), segregate their features/capabilities on modalities, and discuss open issues such as hallucination, data bias, compute/resource limitations and evaluation metrics.The survey ends by outlining potential future directions and implications for research and application.