AutoReelsAI a reels factory and transcription tool with correct Arabic captions and real-estate video enhancement
From a YouTube link or phone video to vertical reels with synced Arabic captions, or a full meeting transcript in seven formats.
- 38
- minutes to transcribe a 1 h 52 min meeting
- 7
- transcript formats
- 6
- caption styles
- 0
- keys needed to run locally
The problem
Arabic captions in most video tools come out with words in the wrong order, transcription breaks on long recordings, and apartment videos shot on a phone arrive over WhatsApp shaky and compressed.
Overview
A production tool I built in Python and Streamlit, a companion to Maestro that runs on the machine itself. It has three pages: reels with Arabic captions, transcription of long recordings, and enhancement of apartment walkthrough videos. It works even without any AI key, using a local transcription engine.
What I built
- Multiple sources: uploads up to 2 GB, a YouTube link or a local path, with up to 10 clips cut per run
- Captions in the spoken language, translated or bilingual; verbatim, corrected or cleaned of filler, with the spoken word highlighted
- 9:16 conversion with face tracking, split screen for two speakers, automatic best-clip detection and music that ducks under speech
- Transcription with on-device speaker labels, exported to TXT, Word, SRT, VTT, JSON, MP3 or all in one file
- A real-estate page: two-pass stabilisation, light correction, sharpening, upscaling WhatsApp video to 1080p and 7 cinematic looks
- Spoken dialects: Egyptian, Saudi, Gulf, MSA and English, with translation into several languages
The engineering call
Multi-colour Arabic captions came out with the words in the wrong order, because every colour change restarted a left-to-right run inside the subtitle renderer. The fix sets the encoding in the caption style and starts every line with a right-to-left mark, with a test that checks the order at the pixel level.
