Reel Transcriber Bot

Problem

I consume a lot of short-form video content for research and learning. Watching every reel fully is slow, and rewatching to catch a specific detail is slower. I needed a way to turn any reel into text I could read, save, and reference later.

Reading a transcript takes seconds. Watching a 90-second reel twice takes three minutes.

Stack

Make.com orchestrates the entire pipeline. A RapidAPI service downloads the reel from Instagram. Gemini API processes the audio and generates a structured transcript with analysis. The bot only processes audio, not video frames. Input and output both happen through Telegram Bot API.

Flow diagram

7 steps — triggered instantly

Webhook → HTTP (RapidAPI download) → HTTP (process media) → HTTP (download file, conditional filter) → Resume (error handling) → HTTP (Gemini API transcription) → HTTP (send to Telegram)

~30 seconds per reel. No scheduling.

The output has four sections: Transcript, Summary, Tone and audience, and Takeaways.

Prompt iterations

First Gemini prompt: raw unformatted text. Round 1 defined explicit output sections (transcript, summary, tone, audience, takeaways). Round 2 tuned handling of overlapping audio, background music, and unclear speech to flag gaps instead of guessing.

Hardest part: getting consistent output format across different reel styles (talking head, voiceover, interview).

Failures fixed

Result

Used multiple times a day. Send a link, get structured transcript in ~30 seconds. Replaced watching, rewatching, and manual note-taking. Running in production on v2 with no interruptions.