Turn a YouTube video into newsletter notes with cited quotes

If you write about talks, interviews or podcasts, the slow part is finding what to quote. The summary preset does the first pass: a short summary, the key points, the most quotable lines and the people in it, with a link to the moment behind every quote.

You still choose what goes in. You just don't scrub through the video to find it.

Run it #

Paste a link into the playground and choose summary, or call the API:

# pip install cleanscript-ai
from cleanscript_ai import CleanScript

client = CleanScript()
result = client.extract("https://www.youtube.com/watch?v=UF8uR6Z6KLc", preset="summary")

result.data has summary, key_points, quotes and people. For Steve Jobs' 2005 Stanford Commencement Address, the summary begins:

In his Stanford commencement address, Steve Jobs recounts three stories from his life: dropping out of college and later seeing how calligraphy influenced the Macintosh, being fired from Apple and finding renewed creative purpose, and confronting a cancer diagnosis that led him to reflect on mortality.

The quotes:

"quotes": [
  "Again, you can't connect the dots looking forward; you can only connect them looking backwards.",
  "Your time is limited, so don't waste it living someone else's life.",
  "Stay Hungry. Stay Foolish."
]

And the citation for the second:

"quotes[1]": [{
  "quote": "Your time is limited, so don't waste it living someone else's life.",
  "source": "speech",
  "start": 752.15,
  "end": 757.29,
  "url": "https://youtu.be/UF8uR6Z6KLc?t=752"
}]

The link opens the video at 12:32.

Draft the section #

This prints a newsletter section with the summary, three key points and the quotes with timestamp links:

def stamp(seconds):
    m, s = divmod(int(seconds), 60)
    return f"{m}:{s:02}"

post, data = result.post, result.data
lines = [f"## {post.title}", f"By {post.author.name}", "", data["summary"] or "", "", "**Key points**", ""]
lines += [f"- {point}" for point in data["key_points"][:3]]
lines += ["", "**Worth quoting**", ""]
for i in range(len(data["quotes"])):
    cite = result.citations[f"quotes[{i}]"][0]
    lines.append(f"> {cite.quote}\n> [{stamp(cite.start)}]({cite.url})\n")

print("\n".join(lines))

The last two quotes come out as:

> Your time is limited, so don't waste it living someone else's life.
> [12:32](https://youtu.be/UF8uR6Z6KLc?t=752)

> Stay Hungry. Stay Foolish.
> [14:28](https://youtu.be/UF8uR6Z6KLc?t=868)

Before you send #

• Quote only the citations. Each citations["quotes[i]"] quote is the speaker's words, checked against the post. The summary and key points are paraphrase, so keep them out of quotation marks.
• Check the summary against the video. It's a fair first draft, in the model's words. Your take should be in yours.
• Click the links for anything you quote.
• Credit the creator and link the video: post has the author and URL.

Your own format #

If your newsletter has fixed parts, such as "the one idea", "the number to remember" and "what to try", describe them as fields in your own schema. See extract fields from a video.

For another language, pass language: values are written in it, and quotes stay in the words that were spoken.