Hi HN,
I am the creator of ReFlow Studio, an open-source (MIT) desktop "uni-tool" for video creators. I built this because I wanted a way to handle the tedious parts of video editing—transcribing, dubbing, and censoring—locally, without relying on expensive, privacy-invasive cloud APIs.
What it does:
* Local Dubbing: Uses OpenAI Whisper for ASR and Coqui XTTS v2 for voice cloning. It is optimized for "Hinglish" (Hindi + English) content.
* Auto-Censorship: Uses NudeNet (ONNX) to find and blur NSFW content in video frames and "bleeps" profanity in audio files.
* 100% Offline: Once installed, no data leaves your machine. This was a priority for creators working with sensitive content.
Technical Challenges:
Distribution: Packing 5GB+ of AI models into a downloadable installer was a nightmare. I ended up using a split-binary Inno Setup approach to keep the download stable for users with limited bandwidth.
UI Persistence: I just hit v0.3.1 which finally integrated a settings manager so the app remembers your theme (Light/Dark) and processing preferences between sessions.
Performance: Managing the handoff between transcription, translation, and rendering threads without locking the main UI thread in Python.
I would love to get your feedback on the architecture, the offline distribution method, or the performance of the local inference!
GitHub: https://github.com/ananta-sj/ReFlow-Studio
Thanks for looking!