MARS5 TTS is an innovative open-source Text-to-Speech (TTS) model developed by CAMB.AI, designed to replicate speech performances across 140+ languages with remarkable accuracy, even in challenging prosodic scenarios such as sports commentary, movies, and anime. This advanced model utilizes a unique two-stage AR-NAR pipeline, featuring a novel NAR component that significantly enhances its ability to generate high-quality speech from minimal audio references. With just 5 seconds of audio and a snippet of text, MARS5 can produce speech that is rich in prosody and tailored to specific linguistic contexts. The model's architecture is engineered to handle raw audio and byte-pair-encoded text, allowing users to guide the prosody of the generated output through simple text manipulations like punctuation and capitalization. For instance, adding a comma in the transcript can introduce a natural pause, while capitalizing a word can emphasize it. MARS5 supports both shallow and deep cloning inference methods. Shallow cloning is faster and does not require the transcript of the reference audio, while deep cloning, which involves providing the transcript, offers higher quality output albeit at a slightly slower pace. This flexibility makes MARS5 an ideal solution for a wide range of applications, from content creation to accessibility tools. Installation and usage of MARS5 are streamlined for ease, with options to install via pip or Docker, ensuring compatibility and scalability across various environments. The model is available on GitHub, where users can access detailed technical documentation, sample outputs, and an online demo to experience its capabilities firsthand. CAMB.AI actively encourages community contributions and feedback, fostering an environment of continuous improvement and innovation. MARS5 TTS is not just a tool but a dynamic platform that adapts to the evolving needs of its users, promising to make every voice count in the digital landscape.