Skip to content

Add yt-infographics: video transcript + infographic extractor - #32

Open
technicalpickles wants to merge 2 commits into
mainfrom
add-yt-infographics-script
Open

Add yt-infographics: video transcript + infographic extractor#32
technicalpickles wants to merge 2 commits into
mainfrom
add-yt-infographics-script

Conversation

@technicalpickles

Copy link
Copy Markdown
Owner

Summary

  • Adds bin/yt-infographics, a self-contained uv run --script (PEP 723 inline deps, no brew install needed beyond uv and a logged-in claude CLI) that pulls everything worth summarizing out of a video into one chronological full_context.md
  • Spoken transcript comes from yt-dlp captions (manual if available, else auto-generated)
  • On-screen infographic/slide text comes from ffmpeg scene-change detection → RapidOCR first-pass OCR → a word-frequency heuristic that triages real slide prose vs incidental HUD/logo/clothing noise → dedup of near-identical consecutive frames (same slide re-triggering a scene cut) → a clean Claude vision re-read of whatever survives, since OCR alone still mangles colored/underlined/stylized text
  • Adds a bin/CLAUDE.md entry documenting it alongside the repo's other bin/ utilities

Test plan

  • Ran end-to-end via uv run (letting uv resolve all deps fresh) against a real YouTube video, verified transcript.md, clean_transcript.md, triage_report.md, and full_context.md all get written correctly
  • Spot-checked transcript de-duplication against YouTube's rolling-caption VTT format
  • Confirmed --skip-download reuses an existing video/transcript for iterating on later stages

🤖 Generated with Claude Code

technicalpickles and others added 2 commits September 3, 2026 09:16
…actor

Self-contained uv script (PEP 723 inline deps) that pulls everything worth
summarizing out of a video: spoken transcript via yt-dlp captions, plus
clean on-screen infographic/slide text via scene-change detection + OCR
triage + Claude vision re-reads for anything OCR mangled. Merges both into
a chronological full_context.md ready to hand to an LLM for summarizing.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Reflects what's actually on disk under --out-dir rather than an assumed
layout, so it stays accurate across --no-transcript and other flags.
Collapses the frames/ jpg listing to a count instead of listing all 35+.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant