OCR Subtitle

1. New Whisper Voice Recognition Mode Integrated the Whisper speech-to-text framework. When videos lack on-screen text or have highly cluttered backgrounds, users can select this mode to directly transcribe audio tracks into time-coded standard SRT files. Supports automatic audio language detection. 2. Refactored Functional Modes (Four Independent Options) To better accommodate varied post-production workflows, core features are now structured into four distinct, mutually exclusive modes: Manual ROI Mode: Users manually define the Region of Interest (ROI) for targeted visual OCR scanning. Auto ROI Mode: The system automatically detects and locks the subtitle area for visual OCR scanning. Voice Recognition Mode: A new mode that bypasses visual scanning entirely, performing voice-to-text transcription directly on the audio stream. Playback Only Mode: A new mode that invokes neither OCR nor voice recognition engines. It functions strictly as a standard video player, allowing users to preview media or overlay existing subtitle tracks. 3. Core Workflow Integration Subtitles generated via either OCR scanning or the new voice recognition mode integrate seamlessly with the pre-existing Seamless AI Translation Workflow and AI Video Copilot prompt templates. 4. Free Tier Limits & Weekly Ultra The baseline evaluation cap applies uniformly across both OCR and voice recognition engines. Free access is limited to transcribing or scanning the first 60 seconds of any single video file. Subscribing to Weekly Ultra fully removes all duration caps and enables unlimited batch processing. 5. Sandbox Privacy & Performance Optimization Strictly operates within the Apple Sandbox framework. Both audio and video streams are processed on-the-fly inside active memory with zero local copying, caching, or remote server exposure. Optimized dynamic memory release logic to improve stability when handling long-duration footage.