A single conference talk is easy to deal with. A course is the thing that defeats me.
Forty videos in a playlist, and by video six the notes have stopped. The material isn't the problem. The bookkeeping is. Which ones have I done. What did video three establish that video nine now assumes I remember. Where did I put any of it.
I've spent the last few months building SummarizeVideoToText and its Obsidian plugin, and I wrote up the plugin internals a while back — markers, tokens, the review rules. This post is the other half. Not how it's built, but what the two halves are for when you point them at a whole course instead of one video. The annoying parts are still in.
The playlist becomes a queue, not a folder
Paste a YouTube playlist URL into the web app and it imports as a collection, up to 500 videos.
You get a preview step first: the title, the count, the first handful of videos, and then a confirm. That step exists because an import that silently grabs the wrong 200 videos is worse than no import at all, and you don't find out until you're three summaries deep.
Everything lands marked "to learn". Nothing is summarized yet.
That last part is the whole design. Summarizing 40 videos up front gives you 40 documents nobody opens. What you want is a queue with a cursor in it. The collection row on the dashboard carries one button — Continue · Part 7 — and that button is the only decision you have to make when you sit down.
One video becomes one note
From inside Obsidian: one command, one link, one note.
What lands in the vault is a TL;DR, key insights, timestamped chapters, frontmatter shaped for Dataview, and an embedded player at the top so a timestamp seeks in place instead of throwing you back to YouTube. Optional extras are the full transcript in a collapsed callout, a quiz, and a longer summary in one of several templates.
Two things about it that matter over a long course rather than a single video.
The note has a generated region between %% svt:start %% and %% svt:end %%, and a "My notes" section outside it. Re-running the command replaces the generated region and leaves your writing alone. On video nine, when you realise your frontmatter tags were wrong on all eight previous notes, you can re-run all of them without losing a word you wrote yourself.
And re-opening a video that's already been processed never costs a credit. Refreshing a note you made last week is free.
If you don't use Obsidian, the same note exports to Notion as a real page, or downloads as Markdown with the timestamps still clickable.
The quiz becomes cards
There's a command that turns the quiz in a video note into Spaced Repetition cards in a separate note.
Run it again after a refresh and it only appends the new questions. Cards you already have, cards you wrote by hand, and the <!--SR:...--> review scheduling the SR plugin writes back are all left untouched. This sounds like a detail and it is the reason the command is safe to press. A "regenerate my flashcards" button that resets three weeks of review intervals is a button you press exactly once.
A quiz across the whole collection
Per-video quizzes test whether you were awake. The thing I actually wanted was a quiz drawn across the collection, because that's where a course either connected for you or didn't — video three's definition showing up inside video nine's argument.
That one runs on the web side, not in the vault, and it's the feature I'd point at if you asked me what a "collection" is for beyond grouping.
Here's one of mine, published: 3Blue1Brown's neural network series — eight videos, eight notes, one page. That's the shape the vault ends up in too, minus the Obsidian chrome.
What it costs
- No account at all: 2 summaries a day, videos up to 15 minutes, plus one TikTok / Instagram / X video a day up to 5 minutes. Transcripts are readable, copyable and downloadable without signing in, five a day. No email, no signup wall to get a look at the output.
- Free, signed in: 10 credits a month, which is roughly ten videos, with videos up to an hour, plus saved history, collections, and the Obsidian and Notion exports. Playlist import needs this tier — 500 videos of state has to live somewhere.
- Pro: $9/month (launch price, normally $12) or $54/year. Unlimited summaries under fair use, videos up to 6 hours, and 300 credits a month for the transcription cases. The 6 hours is the reason this tier exists at all: lecture recordings and conference livestreams are absurdly long.
- Top-ups, if you don't want a subscription: $1 for 20 credits, once per account, or $5 for 100. Bought credits don't expire, and they're only spent once the monthly ones run out.
TikTok, Instagram, X and uploaded files cost more per video than YouTube does, because those have to be transcribed rather than read.
Output comes in 14 languages, and the output language is independent of the video's. Watching a Spanish lecture and keeping notes in English is the normal case, not a workaround.
What it doesn't do
Everything downstream is derived from the words. Captions when YouTube has them, speech-to-text when it doesn't. That means a lecture where the argument lives on the whiteboard and the speaker says "so this equals that" produces a summary that is technically accurate and completely useless. I have no good answer for that one yet. Diagram-heavy talks are where I still take notes by hand.
The plugin also holds no model keys and does no inference locally. It's a client for an account on the server, and it needs the network. If your requirement is that your vault never talks to anything, this is the wrong tool and I'd rather you know that from the third paragraph than from the settings screen.
It needs Obsidian 1.13.0 or newer. The declarative settings API the community review pushed me toward doesn't exist before that.
Links
- Web app: summarizevideototext.com
- A finished collection, if you want to see the output before installing anything: 3Blue1Brown, 8 chapters with notes
- Obsidian plugin: community.obsidian.md/plugins/summarize-video-to-text
- Source, MIT: github.com/diaozxin007/obsidian-summarize-video-to-text
- Setup walkthrough: summarizevideototext.com/obsidian-plugin
- If your material is lectures specifically: summarizevideototext.com/ai-lecture-summarizer
The playlist-as-a-queue idea is the part I'd keep if I rebuilt all of it. Summarizing everything up front feels productive and produces an archive. Summarizing the next one produces a course you finished.
Top comments (0)