YT YouTube2Text

YouTube2Text → Transcripts → GPT 6 + Hyperframe = Crazy combo for expert-level videos

GPT 6 + Hyperframe = Crazy combo for expert-level videos — Transcript

by AI Jason · 4,927 words · 683 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Thanks HubSpot for sponsoring this
0:02video. So when we talk about hyperframe
0:04or remote type of AI generated per
0:06launch video, many people's impression
0:08are still those type of PowerPoint
0:10slides type of video that looks really
0:12boring and not interesting. But things
0:14actually changed dramatically for the
0:15past few months ever since model like
0:17fable was introduced and it went even
0:19further with the recent launch of GPD6
0:22Astra. Video editing capability just
0:24went out of roof. For example, I just
0:26feed this video I download from YouTube
0:28to Astra with hyperframe and with just
0:30one shot after 32 minutes and basically
0:33replicate this video pixel perfectly,
0:35which is absolutely insane because
0:36there's so many visual details to get it
0:39right. And I personally has been using
0:41Fable and Astra with Hyperframe to
0:43generate all those super high quality
0:45launch videos for my own product and it
0:47has reached over 1 million views with
0:49those AI generated product content. And
0:51the most crazy example I personally
0:52experienced is this Fable 5.1 launch
0:55video for my product track. So my team
0:57literally just saw the Fable 5.1 release
1:00announcement during the morning and we
1:01decided to do a launch with this and we
1:03have around 4 hours to prepare the whole
1:05launch campaign including this high
1:07fidelity video that looks definitely not
1:09like a powerpoints plus a landing page
1:11like this that looks also again not vibe
1:14coded and everything prepared within 4
1:16hours and we got more than 100k views on
1:19Twitter. And this is a wake when my team
1:21start doing those type of video based
1:23launch and it really drives the growth
1:25of a product out the roof. So if you're
1:27building and distributing products,
1:28mastering your skills of making those
1:30launch videos with hyperframe or reotion
1:32is a must have skill that will give you
1:34huge amount of advantage. And this is
1:36what I want to talk about today. I'll
1:38talk you through how does those
1:39frameworks works and also make a video
1:41from scratch like this together so you
1:44can understand how I make it and what's
1:45my preferred workflow. But before we
1:47dive into this, part of the reason we
1:49can generate amazing videos with
1:51frameworks like hyperframe is because
1:53modern harness like codeex or cloud code
1:55wrap the model with contest use and many
1:57other features to unleash mod's power.
2:00And there's probably a 10x difference
2:02between someone who just simply prompts
2:03the agent and hope for the best versus
2:05someone who fully utilize every single
2:07tip and tricks. And that's why I want to
2:09introduce you this free cloud code
2:11playbook I put together with HubSpot. It
2:13covers core principles I use for
2:15prompting those agents as well as how I
2:18structure my agent.mmd and cloud.md and
2:21project contacts to make the most out of
2:23it as well as different hooks and
2:25guarios that put together to making sure
2:27my agent actually follow SOPs and how do
2:29I keep long running work on track with
2:31task state as well as different layers
2:33of setup for each project when I have so
2:35many parallel sessions running at the
2:37same time even including some of tips
2:39and learnings I got that allow me to cut
2:41the API bills by 70% % but still have
2:44the same quality output. This is
2:45something I wish I had when I getting
2:47started. It's completely free. I put the
2:49link in the description below so you can
2:51go and check out and thanks HubSpot for
2:53sponsoring this video. Now let's get
2:55back to how to make those awesome
2:57hyperframe videos. So cool idea here is
2:59that we're going to make videos using
3:01HTML which is the thing that large model
3:03is really good at and this is possible
3:05because now we have new frameworks like
3:07hyperframe or remotion. before those
3:09frameworks show up. The way human make
3:11video is in software like after effect
3:13and typically in those software the
3:15video is composite of layers effects key
3:17frames and automation is only possible
3:20by those editor's own API which model is
3:22not trained on top of it and more
3:24importantly there's no easy way for
3:26model to understand what our screen look
3:28like at at 1 minute 3 seconds and
3:30hyperframe team initially actually
3:32experimented using JSON or XML to
3:35introduce a data structure that model
3:37can output to describe a video but just
3:39the help model visualize and understand
3:41what does each frame look like and that
3:43makes model ineffective in terms of
3:45editing and modifying the video but
3:47things really change when they go down
3:49this HTML CSS and JavaScript route
3:51because model is really good at writing
3:53HTML. So the way hyperframe works is
3:55that it allow you or model to express
3:58video timeline in HTML. So you can add
4:01those data composition ID and when does
4:03this part of video start how long the
4:05duration will be and wrap the actual
4:07visual under those diff so that model
4:09can still output the design in normal
4:11HTML and CSS and control the animation
4:14using JavaScript which means it can
4:15reuse animation library like gap to
4:18write animations but just bind the DOM
4:20elements to the video timelines. So it
4:22became very easy for the model to be
4:24able to capture any frame at any given
4:27time and this at crux is what hyperframe
4:29or remote does and the only difference
4:31is that hyperframe is optimized for
4:33HTML, CSS and JavaScript versus remotion
4:36is designed for react and CSS. So both
4:38works but I personally found model is
4:40actually more creative or more free to
4:42design things in HTML. Also hyperframe
4:44comes with a whole bunch of different
4:46skills. When you try to install the
4:48hyperframe skills, it probably feels
4:50pretty overwhelming because there are so
4:52many different skills. You might not
4:54know which one to use. But in short,
4:56there are two sets of skills that
4:57introduce. And one must install skill is
4:59just this hyperframes core. It talk
5:02about the core space of hyperframe and
5:04there's some related skills around it
5:06like hyperframe CRI, animations and key
5:09frames. And they also have this
5:10hyperframes registry which allow agent
5:12to search in a big library. They provide
5:14for different view and animation
5:16components. And meanwhile they also
5:17provide a set of use case based skills
5:20that can be loaded on demand like
5:22product launch videos, faceless
5:24explainer, music to video, PR to video.
5:26For example, I just pointed to this
5:28website to video skills and ask it to
5:31make a video based on uber.com. And with
5:33just one prompt after 21 minute, it is
5:36able to make a video like this is fully
5:38unbranded animated on their core UI
5:41elements and user journey. But even
5:43though you do this after you prompt it,
5:45we're still feels very much like a
5:47PowerPoint slides. A bit boring, not
5:49interesting. And this is what I want to
5:50show you next. Let's do step by step
5:52from scratch to make a launch video
5:55looks like this. Something from
5:57professional studio instead. Let's say I
5:58want to make a launch video for the
6:00track plugin that we just published for
6:02Codex. Then I will open up a new session
6:03in our ripple and give a prompt like we
6:06want to make a launch video of Codex
6:07plugin. And here I specifically tell it
6:09don't look at our existing video for
6:11reference. So we can actually simulate
6:13what does it feel like when we make a
6:14video from scratch. Here I'm using the
6:16Astra high as a model as it is extremely
6:19good for video making. So you ask me
6:21this question. I actually do have a
6:23rough idea about the story line. So I
6:24give a prompt. We want to showcase real
6:26codeex app UI simulates a user journey
6:28from search and install plugin to try it
6:31out to showcase power. So the first
6:33thing it did is actually propose me a
6:36kind of story line in text. And if I
6:38open it I can take a look. So codex
6:41plugin diretory cursor enters search
6:44type track makes sense and plug in
6:46detail install state codec composer
6:49select track using the captured apps
6:52actual control and actual progress two
6:54calls focus on search keyword social
6:56complete brief in the real response UI.
6:59Okay, this is pretty good except I do
7:01want to update the story line a little
7:03bit here. The use case to showcase is
7:05more around like a search keyword,
7:07social, which is a good use case. But
7:09one use case that I want to emphasize
7:11more is this people search use case that
7:14we launched on Twitter this week and it
7:16got viral. So I want to double down on
7:18this messaging and use case. So I tell
7:20you the story line looks good and we
7:22want to show both SEO and social. But
7:24let's start with a people search use
7:25case as a hero and we did a product
7:28launch here as a reference. You can use
7:29track to actually get more context from
7:31this because we have pretty good social
7:32scraping out of box. The agent actually
7:35go and find our specific page route as
7:37well for that people search to get more
7:39context which is pretty good. So it took
7:42it like 2 minutes and the story line
7:44looks pretty good now. um except there
7:47are few other points that sometime I
7:49wanted to talk about like a few key
7:50point that we've been emphasized on
7:52previous video is with track you don't
7:53need to pay subscription this no
7:55contract we literally combine all the
7:57service and turn them into usage based
7:59pricing so that's like one big part and
8:02second one is that uh le enrichment is
8:05just one type of use case we actually uh
8:07have more so those are like important
8:09message that I want to get across and so
8:12I would just give a pro a few feedback
8:13there are some core value proposition
8:15message I want to get across. One is
8:17with track you don't pay for
8:18subscription, you just pay for usage uh
8:20with sense per coin instead of very
8:21expensive subscriptions. Uh and second
8:23one is that we support many use cases
8:26like 2,800 different data and tools. So
8:29we want to get a scen to showcasing the
8:31crazy amount of vendors we have. And
8:33third is that track also ranked number
8:34one people search bench. So basically we
8:36are the best people finder for agents
8:38based on the testings. So I want to also
8:41include that messaging and also like in
8:43our last launch people asking whether we
8:45can get phone number and we actually can
8:47in fact we can get phone number from
8:49just 0.044.
8:51So I will give those feedback to you and
8:53ask it to tweak the story line and the
8:55one thing here is that I know based on
8:57my experience normally the video making
8:59is a creative iteration process which
9:00means even though we define the story
9:02line here now after we produce it we
9:04might realize that we want to shift
9:06things around to make things feel
9:07better. The benefit of hyperframe is
9:10that even though you get maybe one
9:12minute video, the whole video will be
9:14broken down into different compositions
9:16which means actually very easy to
9:17shuffle the order. All right, so you
9:19update the story line, people search
9:21prospect table, can you find the phone
9:23numbers too, benchmark proof, scene
9:25review, vendor explosion, SEO demo,
9:28social demo. Okay, makes sense. The last
9:30thing is that I want to end with our
9:33tagline which is open router for data
9:36and tools. Now I'm fairly happy with the
9:38story line. So I'm just going to ask it
9:40to go ahead and let's do next. You can
9:42see what it did next is actually
9:44creating this story line for me. And now
9:47you can like give feedback for each
9:49frame. And at this point probably no
9:51need to focus on the detailed visual but
9:53more like focusing on two things. One is
9:56the overall layout of the screen and the
9:59second one is the copies and you can
10:01give feedback on every single frame.
10:03then you can just click on this button
10:04to copy the prompt and paste back. So
10:07I'm going to do that one by one and here
10:08is where I think at you as founder or
10:11marketers paste actually comes in um
10:14because at default even with model like
10:16uh Astro it still tend to make
10:19PowerPoint slice type of videos out of
10:22box but if you give the right context
10:25and guidance it can make amazing things.
10:27So I'm going to do that one by one. Like
10:29for the first one, I said I don't like
10:31this PowerPoint type of layout. Also, I
10:33don't like the find your next customer's
10:36text on the left since it kind of anchor
10:38track to only that use case even though
10:40that's important, but I don't want to do
10:41that. And also, let's start with the
10:43codeex macap UI then searching stop
10:45plugin. We will figure out the beginning
10:49later. So, this is like what I tend to
10:51do when I make a video. I try to making
10:53sure the crux of the story line good and
10:56later work out like how the first one
10:58second intro should look like and how
10:59the end should look like. For second one
11:01I would say looks fine but make sure the
11:03UI is pixel perfect with real codeex Mac
11:06app UI. And unfortunately I can't paste
11:08image here. Hopefully they add that
11:10later but I'm going to add that image
11:11screenshot in the chat. And this one I
11:13give feedback like the translation of
11:15the whole video should stay in real code
11:17session so it looks like a real demo.
11:19And after the second thing of install
11:21plugin, it should translate to real
11:23codeex macap UI focus on the promp box.
11:25So typing out query also make query
11:27short because people's attention span is
11:30extremely short. So I always prompt
11:33agent to basically simplify like a lot
11:35of work on video making it just simplify
11:37what's showing on the screen because for
11:39example in this one I think the text is
11:42too long and then go to next and this
11:45one again the whole video should be a
11:46continuous codeex code session so
11:49everything just show in the chat session
11:51however you can zoom in and out after
11:53getting the first table and show the
11:55phone number promise followup and
11:57there's one thing I forgot to ask it to
11:59show showcase which is One key part of
12:01track is because we connect so many
12:03different leads and contact database. We
12:05do a waterfall matching which means user
12:07can get results the cheapest possible
12:09way with the best coverage. So showcase
12:12this maybe after agent receives a prompt
12:17track list all the vendors with matching
12:18endpoint and price then scan choose the
12:21right vendor and people search proof. I
12:24don't like the layout at all. I really
12:26don't understand the number it putting
12:27here. And for this demo purpose, I'm
12:29purposely didn't reference any of the
12:31past videos I made. But the beauty of
12:33hyperframe is because everything is in
12:36code. If I open files, you can see all
12:38my past launch videos are just a folder.
12:40And each folder, let's say if I go to
12:43enrichment launch, it has compositions
12:45folder. And this composition folders is
12:47including HML of every single page. And
12:51each HTML files in that folders is
12:53pretty much a video clips like this. So
12:56technically I can just point the agent
12:58to one of those pass frame and say hey
13:00just reuse that. So everything is
13:02actually quite reusable but just for our
13:04demo purpose I'm not going to do that. I
13:06will just build from scratch to
13:08simulate. So for the people search proof
13:11I just give a prompt. Let's start with
13:12compare cloud code score versus cloud
13:14code plus track score on B2B data bench
13:18and you can reference data from our
13:19website which actually have the
13:21comparison data already and then magic
13:24translate to show benchmark of all four
13:26categories. So I wanted to show on B2B
13:28perspective first since that's probably
13:30the most common scenario and then
13:32translate to showing all of them in
13:33terms of recruiting influencer finding
13:35things like that. And for this one I
13:38will give it more detailed prompt. So
13:40here you can see that I'm prompting it
13:41in a way uh saying we want to center big
13:45tax show no subscription wait no
13:48contract wait and you line and pay for
13:50results only. So this is my way of
13:52prompting to get thing like this. No
13:55subure, no contract, wait and zoom out,
13:57show vendors and show price. And I found
13:59this is actually pretty effective way of
14:01explaining what you want. And then zoom
14:03out and move the header text up. Show
14:05the rest of things. Left shows vendor
14:07price. Right side shows track price. And
14:09I want to show like dollar based on
14:11instead of cents to bring some more
14:12contrast. And this one is all right.
14:14Okay. So I already given like a whole
14:16bunch of feedback. And I actually tend
14:18to not giving way too much feedback in
14:21one go. So I'm just going to click save
14:23and copy message and paste here. I can
14:26read the storyboard feedback I save in
14:27hyperframe frame commands.json. So if I
14:30open that folder codeex plugin
14:33hyperframes they have this JSON that
14:35actually log all the feedback that we
14:38made uh bind to the context. So it's
14:40pretty easy way to give feedback.
14:43So now you can see it's been like 13
14:45minutes to process feedback and
14:46basically rebuild most of the pages
14:48based on our feedback and there are more
14:50things I will just go back and forth
14:52like this one initially just show cloud
14:54code versus code cloud code plus track
14:56because it's easier for people to
14:57consume but after that can show all
14:59competitions in four category it should
15:00include all results like XAR let's see
15:03and this one no subscription no
15:04contracts in the same line and also show
15:06a lot more vendors on the left side here
15:09I want to try something like let's
15:11update the out
15:13in codeex chat which track plus uh
15:17vendors it is chosen which is what you
15:20have now and on the right side it open a
15:25window show a Google sheet or CRM type
15:30UI we seeing results of contacts just
15:34flaw in all I will copy and send again
15:39this time it got like story layout out
15:41more or less what I want then I'll just
15:44approve it. You can either copy this
15:45button which will basically send this
15:47message approve the storyboard sketch
15:49and continue to animation or you can
15:51just tell it I approve it. So now you
15:53can see that it actually generate one
15:55version of videos and if I make it
15:57bigger and play it you can see actually
15:59start animate this video based on the
16:01wireframe. Even though some parts you
16:03can tell that it is not correct like the
16:06UI of this smack app doesn't look like
16:08real codeex app at all but this is stage
16:10for you to align with hyperframe about
16:12whether the animation actually looks
16:14correct and give feedback and same thing
16:16we can go back to storyboard and give
16:18feedback frame by frame and what I tend
16:21to do is normally at this point what do
16:24we have is a story line and each scene
16:27in story line is animated so technically
16:29we can just ask codeex to make edit is
16:33scene by scene. So you can spend more
16:35token or attention on individual things
16:37which will normally lead to higher
16:39quality. So what I'm going to do next is
16:41let's just update the first few things
16:44first to making sure it looks exactly
16:47like the real codeex app interface. So
16:49here I'm saying that next let's update
16:5101 discovery and 02 install to making
16:53sure we mimic 100% pixel perfect codeex
16:56app UI to make it almost look like a
16:58real codeex demo. And I'm also going to
17:00add plug-in UI as well.
17:02It is still working but you can see that
17:04it actually turned that image into pixel
17:06perfect UI replication which is
17:09absolutely amazing because it looks
17:11exactly same. I'm also going to just
17:13prompt here can you add a background
17:15where the codeex app is shown. So it
17:17almost looks like a screen studio
17:19recording. All right. So now you can see
17:20it has this background and try to play
17:23it. It will be on this real codeex app
17:25looking interface and it is mimicking
17:28the search. uh click on try uh and then
17:31go to next screen which is not uh
17:32exactly a good-looking one but this is
17:34good enough. I'm just going to like back
17:36and forth uh fine tune a little bit
17:38because I think it's important for you
17:39to get the first few screen right to set
17:42a standard and the style for the
17:44following things. I give you further uh
17:46feedback like after click plugin try
17:48type to search click on plugin things
17:50like that it should all zoom in camera
17:51so we have more clear focus because at
17:53moment might be difficult to see what is
17:55happening so I'm just going to send it
17:57all right with this feedback and I play
17:59again then have this zooming effect uh
18:02that looks like a real recording but I'm
18:04not too big fan of this transition like
18:07search jumping to this zoom out and
18:09zooming again I want it to you know just
18:11be faster just tell you currently after
18:13we get into pluging in detail page if
18:15first they zoom out and then zoom in to
18:18uh click try it and I think this motion
18:19is not necessary remove it just keep the
18:22same zoom in scale so zooming scale is
18:24kind of word you can use to tell it
18:25basically keep zooming scale and just
18:28move camera position okay right now if
18:31we play again it should stay on the
18:34scale here and just go click try now
18:36immediately so you can see this process
18:39we're doing where we just get back and
18:41forth and give feedback once you get the
18:43model's attention on just one or two
18:44seams, it can do extremely good job. So
18:47after this, the process is pretty much
18:49the same. We just going to repeat this
18:50back and forth composite by composite.
18:53And the good thing is that because we
18:55already set up the standard for the
18:57first two compositions, we can just tell
18:58it to do the same to apply same UI
19:01treatment for the next few scenes. And
19:03again, I try to do composition by
19:05composition so you can achieve the best
19:07quality. So I would say great next let's
19:11update 03 composation for people.
19:15There's one feedback I understand which
19:16is um currently after showing the
19:19result. There's a weird zooming
19:20animation in the end and we don't need
19:22it and it just makes the time stall on
19:24the last frame a bit shorter. Okay, for
19:26some reason it is not applying the UI
19:29updates. It just made some change on the
19:31zoom level. Um so I'm going to prompt it
19:33to make a change. So it made a new
19:36version for the composition 3 which
19:38looks something like this.
19:42It's not bad like the UI already looks
19:45very high fidelity though I have a few
19:48feedback I want to give. So I'm typing
19:49out more detailed feedback like firstly
19:51the zooming level for the prompt input
19:53can be a bit more so text will be bigger
19:55and also remove the toolbar above which
19:58is this one and also full access to
20:00remove the visual clutter a bit because
20:02I think one part about this launch video
20:04is people's attention span is very short
20:07so you want to making sure there are
20:08less elements as much as possible and
20:11after click send it should move camera
20:13up we should see the new message pop up
20:15and then the track to also pop up with
20:17more detail animation
20:19Uh cuz currently after it send uh it
20:22just slightly zoom out and I can't
20:23really see the details here. Um so I
20:26want it to add a bit more um detailed
20:28information here. So I click send. All
20:30right. So it got a new version and let's
20:32take a look.
20:40So it's actually not bad but I'm going
20:41to uh make some tweets. So once making
20:45sure you show real logo for each vendor
20:46and second is let's do step by step. So
20:48firstly track show up with some loading
20:52state then vendor popping up then after
20:55choosing the right vendor instead of
20:56zoom out maybe move the camera to the
20:58right side so more clearly we can see
21:00the contacts popping out and each
21:03contact maybe add a real profile picture
21:06and also I'm going to send a second
21:08piece of feedback lab. We want to show
21:09an action card of Google sheet after
21:11vendor list and showing it is fetching
21:13and saving results to Google sheet. The
21:15same card that you use in chat just show
21:18up with loading state while we're
21:19popping out results for the Google
21:21sheet. So basically after vendor is
21:26chosen action card show popping up so
21:31users understand it is fetching and
21:34saving results. Then move camera to the
21:38right side and same results popping up.
21:42All right. Okay. So we got new version.
21:44Let's see it again.
21:48Let's I like the loading state and now
21:50also it shows Google sheet. Then on the
21:52right side everything pop up. So it
21:54feels like a lot easier to follow. One
21:56thing I want to mention like actually
21:58the model plus hyperframe at moment is
22:00actually capable of doing really high
22:02quality things. So the quality of video
22:04is a lot about your attention to
22:06details. For example to be more clear
22:09like all those email are made up. So I'm
22:11going to talk about it uh like while we
22:14popping up the content. It should first
22:16they show name real company this race
22:19and the work email column should show
22:21fashion with different vendors. So
22:23viewers can tell it look like waterfall
22:26fetching for verified email and it is
22:28happen simultaneously for each contact.
22:30So it is all those little detail to make
22:32your video actually representing the
22:35core value of your product. Okay. So now
22:37we get a new version. Let's play again.
22:52Cool. So I'm pretty happy with this one
22:54and I'm just going to ask it to
22:57replicate for the next scene as well.
23:00Let's do composition 4 with same quality
23:02and UI treatment which looks something
23:04like this at moment and uh we will see
23:07what kind of result we get after we
23:08update. All right. So just made update
23:10for the uh composition 4 as well and you
23:12can see because we set the quality for
23:14one two three really well the scene
23:16number four almost just one shot get a
23:19really good result. It follow all the
23:20patterns that we defined earlier. Uh
23:22then it is moving the camera to the
23:25right and the whole thing is really high
23:27quality. Um then I'm going to do next
23:30which is this uh prove screen. Um and
23:33again I just have like lots of detailed
23:36feedback and guidance about the visual.
23:38So firstly the visual looks bad just
23:40align quality we have on the one to four
23:42composition and also start frame with
23:45big header text in the middle with Apple
23:47Nike ad type motion and camera movement.
23:50uh first they focusing on most accurate
23:52weight people finder weight new line for
23:55agents then show codeex cloud code open
23:58claw her logo and uh um I point to loop
24:01hub so loop hub is just a open source
24:04package that has uh almost all the logo
24:07of major AI provider and harness and
24:09then translate to this people search
24:12bench of B2B perspecting where cloud
24:14initially show 43% with a bar chart then
24:17cloud plus track reaching 78% almost
24:20double the amount and again using the
24:23bar chart to contrast then show the four
24:26charts with nice translation. Okay, so
24:29after a few rounds of uh iteration back
24:31and forth u probably for uh like one and
24:34a half hours I got a version that I'm
24:36like pretty happy with. It has all the
24:38animation the story line is correct. Um
24:42it showcase the core features that um
24:44our product actually offer which is you
24:46know connect to also different um
24:48systems and uh you can do waterfall uh
24:51to routing through different providers
24:53so that you can get the data or
24:55information you need with the cheapest
24:57pos possible price. Um and uh you can
25:01see all the animation like feels
25:02actually pretty smooth um and tells the
25:05story right. Um the coursing here is
25:08really you just need to spend time um to
25:10update the details u frame by frame.
25:20So I'm pretty happy with this one. Uh
25:21and lastly let's just add some music. So
25:23I prompt it next I want to choose some
25:26music. H actually comes with a quite big
25:29music library.

About this transcript

This page contains the full transcript of GPT 6 + Hyperframe = Crazy combo for expert-level videos by AI Jason, generated from the public captions YouTube serves with the video. The transcript has 4,927 words across 683 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.