AI import
The video ends.
The map is what you keep.
An hour of someone's best thinking arrives as a timeline you can only move along — and when it's over, what's left of it is a line in your watch history. Paste the link. Atlas Red pulls the video's transcript and lays the argument out as an editable mind map, with the video still linked from it.
How it works
Copy the link
Any shape of it — the address bar, the Share button, youtu.be. Tracking parameters and playlist ids are ignored.
Paste it, choose the depth
Atlas Red fetches the transcript itself. Pick overview, balanced or detailed — how much structure you want kept.
See it laid out
The whole shape at once, and the video linked from it. Rearrange, prune, export to any of 14 formats — or hand it to your assistant over MCP.
Nothing is downloaded and nothing is transcribed, so a two-hour talk usually finishes in well under a minute.
What it needs — read this first
One condition decides whether a given video will work, and it's worth knowing before you paste:
The video must have captions. Atlas Red reads the caption track YouTube already holds — it doesn't listen to the video — so an upload with no captions in any language can't be imported this way. That's checked first, for a fraction of a cent, so you're told immediately rather than after paying for a job.
In practice this is rarely the blocker: conference talks, lectures, interviews, tutorials and news almost always carry captions, auto-generated at minimum. Music videos and small unlisted uploads often don't.
Two more limits, both stated up front: videos are capped at six hours, and Shorts, clips, playlists and channels are refused — each by name, so you're told which one you pasted rather than that your link looks broken. A normal watch link that happens to sit inside a playlist is still just that video, and imports fine.
The progress bar is not an outline
The only structure a video hands you is a timeline. It tells you precisely where you are and nothing whatsoever about what's there — which is why finding the one part you needed again means scrubbing, and why a chapter list, when there is one, feels like such a relief for something so thin.
The thinking inside a good talk isn't linear anyway. It has a main claim with two or three supporting lines under it, an objection raised and answered, a digression the speaker flagged as a digression. All of that reaches you at the same speed, in the same voice, spread along the same bar — and the structure has to be reconstructed from memory, if it survives at all.
A map lays it back out. Everything sits on one surface, in relation to everything else: the main threads side by side, the supporting detail folded under the point it supports, the aside visibly an aside rather than four indistinguishable minutes. You take it in the way you read a diagram, spatially and at a glance, instead of playing it back.
Where the words come from
Atlas Red doesn't transcribe the video — it takes the transcript YouTube already has, which is why this route is fast and costs almost nothing. Two things about that are worth knowing:
- The uploader's own captions are preferred over YouTube's auto-generated ones wherever both exist. Uploader tracks are punctuated and spelled properly; machine transcription is neither, and punctuation is a surprising amount of what tells an argument apart from a word stream.
- The language is whatever the video actually has. Your browser's language is used as a preference, not a demand: if the video doesn't carry it, Atlas Red falls back through the common languages and finally takes the first track available — and then tells you which one it used, rather than leaving you to work it out from the map.
Auto-translation is deliberately not requested. A machine translation of a machine transcription is a quality floor not worth paying a model to summarize; if you want a talk in another language, an honest map of what was actually said is the better starting point.
The map remembers where it came from
The finished map keeps the video's title and a link back to the watch page, stored with it rather than in a note you'll lose. Six months on, the map is what you'll actually search — and the source is one click away when a node turns out to need the original context.
The transcript is kept alongside it, which buys something else: you can rebuild the map from that same source at a different depth without re-fetching anything. Import at overview, decide you wanted more, rebuild detailed — the video is never touched a second time.
And because it's an editable map rather than a generated artefact, the version you end up with is yours: collapse the branch you already knew, merge two points once you notice they were the same point, cut the half that wasn't relevant. What's left is the talk as you understood it — which is the version you'll still recognise next month.
The structure, plus whatever else you want on it
Beyond the shape itself, you can name up to 20 fields before importing — labels of your own choosing, in your own words. There's no fixed set and nothing is required. For a conference talk you might ask for:
A field is a request, not a guarantee. The AI fills one in only where the transcript genuinely supports a value — it's instructed never to invent one — so a point the talk didn't cover comes back with that field simply empty. That's deliberate, and it's the part worth trusting: an empty field is information, where a confidently wrong one would be a trap.
The other dial is depth. Overview gives a roughly 15-node take on the whole video, balanced around 40, detailed around 100 — and a sentence of focus narrows the map to the one part of the talk you're actually here for.
What it costs
$1 trial credit
Every new account starts with it — no credit card required, valid 30 days.
AI credit packs
Top up when you like — no subscription, no expiry. Failed jobs are never charged.
Bring your own key Pro
Connect your own account for the map building — no markup, ever.
This is the cheapest AI import Atlas Red has, and it comes in two parts. Fetching the transcript costs under half a cent, because no model is involved in it. Building the map is the part that costs anything — the ordinary text price, so a full-length talk usually lands around 10 to 30 cents at the balanced setting and a short video a few cents.
Taken together, the $1 of trial credit every new account gets is good for roughly a dozen videos — fewer if they're multi-hour talks, more if they're short. No card, valid 30 days, and failed jobs are never charged.
One honest detail about the fetch: it's always billed as credits, even for Pro users running the map build on their own key. It goes through Atlas Red's caption vendor rather than a model you can bring an account for. Two calls per video, under half a cent all in — so a bring-your-own-key account does still need a credit balance above zero to import from YouTube, though a dollar of it goes a very long way.
Questions people ask
How do I turn a YouTube video into a mind map?
Copy the video's link and paste it into Atlas Red's import menu. The caption track is fetched, read for meaning, and built into an editable map — usually in well under a minute, since nothing has to be transcribed. The map keeps a link back to the video.
Does the video need captions?
Yes. Atlas Red reads the transcript YouTube already holds rather than listening to the video, so an upload with no captions in any language can't be imported this way. That's checked first and costs a fraction of a cent, so you find out immediately. Talks, lectures, interviews and tutorials almost always have captions; music videos often don't.
Can I import a Short, a playlist or a channel?
No — and each is refused by name, not as a broken link. Shorts carry no transcript. A clip addresses a segment rather than a video, so open the full video. A playlist or channel is many videos with no single answer. A normal watch link that sits inside a playlist works fine.
How long a video can it handle?
Up to six hours, checked before anything is fetched. Past roughly three and a half hours of speech the transcript hits a 200,000-character cap and is truncated rather than refused — and both the import and the saved source say so. A partial map of a five-hour stream beats a refusal you can do nothing about.
Which language does it use?
Whichever the video has. Your browser's language is a preference, not a demand: if the video doesn't carry it, Atlas Red falls back through the common languages and takes the first available track — preferring the uploader's captions over auto-generated ones — then reports which language it used.
What does a YouTube import cost?
The cheapest import there is. The transcript fetch is under half a cent; the map build is the ordinary text price and is the part that costs anything — around 10 to 30 cents for a full-length talk at balanced, a few cents for a short one. The $1 of trial credit every new account gets is good for roughly a dozen videos, valid 30 days, no card. Failed jobs are never charged.
Can I use my own API key instead of credits?
For the map building, yes — Atlas Red Pro lets you connect your own AI account with no markup. The transcript fetch is the exception: it runs through Atlas Red's caption vendor, so it's billed as credits even for key holders. That's two calls per video, under half a cent — enough that a key-holding account needs a balance above zero, not enough to be worth thinking about after that.
Does it download the video?
No — only the caption text, never the audio or the video file. The link you paste is reduced to its eleven-character video id before anything leaves Atlas Red, so your URL is never forwarded anywhere.
Is it just a summary, or something I can work with?
A real mind map, not a picture of one: every node editable, movable and deletable, exportable to any of 14 formats, and open to your AI assistant over MCP. Overview gives a roughly 15-node take, detailed around 100 — and a sentence of focus narrows it to the part you came for.
No captions on it? Upload the recording and it's transcribed for you. Already have the text? Transcripts and caption files. Working from a document? Word, PDF and photos.