Developed multilingual support for the Qwen3-TTS model within the vllm-project/vllm-omni repository by integrating language configurations from fine-tuned model checkpoints. This feature introduced dynamic language validation and fallback mechanisms, enhancing the reliability of text-to-speech outputs across multiple languages and reducing language-specific errors. The implementation leveraged Python for API development and machine learning, with a focus on robust testing and code traceability through detailed commit and pull request linkage. The work established a foundation for future stability and performance improvements, ensuring that multilingual capabilities are both extensible and maintainable as the project evolves to support broader language coverage.