krystophny/slopcast

slopcast

Automated lecture videos from Beamer PDFs with pdfpc speaker notes plus a reference voice WAV. Pipeline: slide.pdf → segments.json → audio/ → clips/ → final/lecture.mp4. Voice cloning via Qwen3-TTS, optional split-screen livecode segments.

Use this skill when the user asks to render a lecture video, re-render selected slides, splice clips, or author Beamer notes for slopcast.

Repo: ~/code/sloppy/slopcast. Driver: ./generate_lecture.sh.

Generate

./generate_lecture.sh path/to/slide.pdf path/to/reference.wav

Output goes to slopcast/<lecture-key>/{segments.json, audio/, clips/, final/}. Clips are content-hashed: re-running only redoes what changed.

Useful flags

--slides 36-38                 # work on a slide range
--pages 1-20                   # PDF page range
--rerender-clips 3,5-7         # re-render selected clips, reuse audio
--patch-into lecture.mp4       # splice the rendered range into an existing video
--merge-clips                  # produce final/lecture.mp4 (default stops after clips)
--pacing natural|flat
--ref-text "transcript of the reference WAV"
-o out.mp4
--lecture-key foo

./generate_lecture.sh --help lists the full set.

Authoring (Beamer)

Use the \autonote macro (preamble in the slopcast CLAUDE.md) instead of bare \note<N> — it auto-captures the current overlay number and avoids a known duplication bug.

\autonote{[moderate] Die Sprungfunktion springt um eins, wenn x positiv wird.}
\pause
\autonote{[simple] Formal: Theta von x.}

Complexity tags drive pause timing:

| Tag | Pause after | Lead-in | |---|---|---| | [trivial] | 0.5 s | 0.5 s | | [simple] | 1.5 s | 0.8 s | | [moderate] (default) | 3.0 s | 1.0 s | | [complex] | 5.0 s | 1.5 s |

End-of-slide bonus: +1.0 s. Write umlauts as UTF-8 (ä ö ü ß), not LaTeX escapes.

Livecode

\note{[livedemo:livecode/accumulator.py]}

Renders a split-screen segment: real python3 execution, terminal typing animation, mechanical keyboard sounds.

Dependencies

sudo pacman -S ffmpeg poppler python    # or apt/brew equivalents
pip install -r requirements.txt pygments keyboardsounds

QWEN3_TTS_PYTHON should point at a venv with Qwen3-TTS installed (default: ../slopshell/tools/qwen3-tts/.venv/bin/python). A 5+ minute reference WAV of the target voice is required.