meizhong986/WhisperJAV — reverse-engineered prompt
Reverse engineered prompt
Build me a Python subtitle generator for Japanese adult videos that can run from a simple desktop GUI and also from the command line. I want users to add one video or a whole folder, choose a processing mode, choose sensitivity, click Start, and get usable subtitle files with good timing.
The main goal is to handle noisy audio, whispers, breathing, long scenes, and weird hallucinated text better than normal Whisper. Please include several modes like fast, balanced, high accuracy, Qwen ASR, anime Whisper, and transformer based Japanese models, with sensible defaults for normal users. Use scene detection, voice activity detection, optional speech enhancement, and filters that remove obvious hallucinations or non speech markers.
Also add a two pass option that can run two different models and merge the best results. Make installation easy, include clear docs, and make sure it works for both single files and batch processing. Look up current model docs online if you need to.
Want more depth? Deep Reverse