Reel Sloth: I saved 'for later' reels for years. Now a local Gemma sorts them while I sleep 🦥

Description and use

Reel Sloth automatically extracts recommendations from saved Instagram reels, classifies them, and produces a to‑do list that can be exported to Markdown, CSV, JSON or synced with Notion.

Problem

Users save Instagram reels as a to‑do list but never revisit them, leaving valuable recommendations trapped inside videos.

Outcome

Reel Sloth converts 6 sample reels into 15 clean backlog items (movies, books, AI tools, a place, a product) in under 35 minutes on a 16 GB CPU machine, with overnight mode enabling large‑scale, offline processing.

Workflow

• 1. Provide reel URLs or an Instagram data export (saved_posts.json).
• 2. Reel Sloth downloads videos and captions using yt‑dlp, respecting Instagram rate limits and retry logic.
• 3. Audio is transcribed with faster‑whisper (small model) on CPU.
• 4. Overlay text is extracted from keyframes or carousel slides using Qwen2.5‑VL via Ollama.
• 5. Gemma 4 E2B classifies the combined text into categories and actions, returning JSON.
• 6. Custom Python cleanup normalizes titles, merges duplicates, and removes junk.
• 7. Results are cached in SQLite for incremental runs; overnight mode throttles downloads and resumes on interruption.
• 8. Output is exported as Markdown/CSV/JSON or synced to a Notion “Reel Backlog” database.
• 9. Users check off items manually in Notion or exported files.

Evidence

Reel Sloth does the "later" part for you. You give it your saved reels and it hands back a tidy, checkable backlog:

Source

The first run downloads the Whisper small model (~460 MB). After that it works offline, except for downloading the reels themselves.

Source

My machine is a Windows PC with 16 GB of RAM and no GPU, so most of the work was making open models behave on modest hardware.

Source

Everything is cached in SQLite, so re‑running costs nothing for reels you’ve already processed.

Source