A novel end-to-end framework is proposed for the detection and removal of hardcoded or embedded subtitles from videos. The framework utilizes two deep learning models, a modified CTPN (Connectionist Text Proposal Network) model, and an E2FGVI (End-to-end Framework for Flow Guided Video Inpainting) model, to achieve its goal. The modified CTPN model is designed to detect multi-scale, multilingual subtitle text while ignoring background or scene text and has been optimized for faster detection, reducing the computational cost of detecting subtitle text from 0.14s/frame to 0.08s/frame. The output video resolution is preserved, ensuring that the final output is of high quality. The proposed framework represents a valuable contribution to the field of video processing, as it addresses the challenges of detecting and removing subtitles while preserving the content of the original video. The entire process of detecting and removing subtitles can be completed in approximately 120 seconds, making it a computationally efficient solution.