Language EN ES

Tutorial: podcast → vertical shorts with a receipt

Canonical first tutorial for Kinocut. Local-only. No uploads. Human review required before publish.

You will produce

One or more vertical clips from a local podcast/interview file
Optional captions if Whisper extra is installed
quality / checkpoint artifacts
A Video Receipt summary (intent, tools, pending human review)

Prerequisites


brew install ffmpeg   # or apt

pip install kinocut

kino doctor

# optional captions:

# pip install "kinocut[transcribe]"

Connect MCP (optional):


claude mcp add kinocut -- uvx --from kinocut kino

Path A — Agent (recommended)

Paste into Claude Code / Cursor with Kinocut MCP enabled:


Local file: ABS_PATH/episode.mp4

Using Kinocut only:

1) Probe duration and streams

2) Trim a strong ~45–60s segment (or start=00:05:00 if unsure)

3) If transcription is available, create an SRT and burn captions; else skip and note why

4) Resize to 9:16

5) Normalize audio (~ -14 LUFS)

6) quality_check + release_checkpoint

7) Summarize a Video Receipt: intent, tool_calls, quality, human_review pending

Do not upload or claim publish-ready.

More prompts: PROMPTS.md.

Path B — Python client


from kinocut import Client



c = Client()

src = "/ABS/PATH/episode.mp4"

clip = c.trim(src, start="00:05:00", duration="00:00:45")

# optional: c.ai_transcribe(...); c.subtitles(...)

vert = c.resize(clip.output_path, aspect_ratio="9:16")

norm = c.normalize_audio(vert.output_path, target_lufs=-14.0)

final = c.convert(norm.output_path, format="mp4", quality="high")

print(c.quality_check(final.output_path))

print(c.release_checkpoint(final.output_path, min_score=50))

Path C — Golden plumbing proof (no private media)


python scripts/golden_path.py

Path D — Workflow engine

Use a job spec (probe → trim → resize → add_text) with workflow-validate / plan / render / inspect.

See WORKFLOWS.md and examples/workflows/captioned-vertical-short/.

Human review checklist

[ ] Hook is clear in first 2 seconds
[ ] Captions accurate (if any)
[ ] Loudness comfortable
[ ] No bad crop on faces
[ ] Receipt matches what you intended