YT YouTube2Text

YouTube2Text → Transcripts → How to Audit ComfyUI Generative AI Workflows for Film & VFX

How to Audit ComfyUI Generative AI Workflows for Film & VFX — Transcript

by Victor Perez's NODE BY NODE · 4,718 words · 718 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:05Hello and welcome back to generative AI
0:07for visual effects, ComfyUI for Nuke.
0:10This is Victor Perez and this is a bonus
0:12class that goes between class number
0:15five and class number six of the unit
0:18four. So, we are discussing ComfyUI.
0:20But, I wanted to bring this particular
0:23class at this point to celebrate the
0:25inclusion of Article 50 for the European
0:30Union legislation of AI. So, on the 2nd
0:34of August of 2026,
0:37the Article 50 is going to enter to
0:39become law. And maybe not all the
0:43implications are going to be directly
0:45engaged at that date, but we are going
0:48to start getting into this new way of
0:52regulating the transparency of the use
0:55of AI. So, we are not just regulating
0:59the technology, we are regulating how we
1:02are going to develop that technology.
1:05So, it's all about transparency. It is,
1:08in brief, people has the right to know
1:12that you use AI for the content they are
1:16watching. This is for many reasons,
1:18especially to protect, you know, the
1:21identities and all of that, but also, in
1:24the long run, is to protect what we love
1:26the most, our industry and our jobs. So,
1:30it's a way of deploying the information,
1:34for declaring the transparency of how we
1:37use that in any project. So, of course,
1:41right now, this is legislation of the
1:45European Union, but we can expect other
1:48countries to start joining the same
1:50current, maybe with different
1:52variations, but the purpose is going to
1:56be always the same. We just need to
1:57disclose what's going on in here because
2:00the technology is so good that sometimes
2:04things can look real when they are
2:06actually not. Um this power can be
2:10in the wrong hands. So, that's why um is
2:13going to affect the whole chain from
2:15distribution to the content creators and
2:19everybody must adopt this way of
2:22transparency. And I think it's something
2:24very, very healthy. But,
2:27once that is going to
2:29add a new layer of uh elements to keep
2:33into account uh like for instance, how
2:37to create an audit. And I know it may
2:41look like overwhelming just to create
2:43and to enter into a legal system and all
2:45of that. And yeah, for sure we are going
2:48to have a rise into new figures, new
2:51professions into the AI world like
2:54lawyers that specialize in this kind of
2:57uh
2:57disclosure that we have to that we have
3:00to create. So, uh but in the meantime,
3:03it is important that you understand how
3:07this thing works. And to be honest, it's
3:09not even that difficult once you know
3:12what you need. So, for this class, what
3:15I'm going to do and I'm going to release
3:17this publicly because I think this is at
3:20the service of everybody. So, uh you can
3:23find this class on on YouTube and it's a
3:25bonus. So, yeah, it is it's something
3:28that everybody can benefit. Um I'm going
3:31to use as an example as I was saying um
3:34maybe one of the most restrictive which
3:37because it's a very, very big
3:40uh channel of distribution that is uh
3:43Netflix. That was one of the pioneers
3:45into declaring
3:46uh how to use the AI by providing uh you
3:50know, a guide uh, for generative AI for
3:54content production that is going to be
3:56used uh, for any content uh, distributed
3:59by Netflix. So, this is uh, a page that
4:03is actually uh, available. So, you can
4:06access the the URL is in here, but I'm
4:10going to give you this in the
4:12uh, in the content materials for for
4:14this class. Um, but in here you can
4:18read, you know, uh, what what they were
4:21expecting you to understand
4:24uh, that uh, you need to keep into
4:27consideration when you are creating
4:29content using generative AI either for
4:33visuals or for audio, for music, voice,
4:37representing uh, the likeness of someone
4:39else. So, you have in here very very
4:43clear. It's It's not like super super
4:46complete, but we are also calling for
4:49the common sense as well. So, uh, this
4:51is This was a very practical tool and a
4:54very easy way to start introducing the,
4:58you know, some sort of uh, transparency
5:01vehicle for for the content. So, I'm
5:04going to leave this in here. Um, I have
5:06been basing also the the contents that
5:09I'm going to show in in here that I'm
5:11going to distribute for for you as well
5:14uh, in the class materials, but I was
5:16basing uh,
5:18the
5:20the disclosure uh, agreements and all
5:22the all the paperwork we are going to we
5:24are going to have in here
5:26uh, on both the EU laws that you have in
5:30this article that you are going to find
5:31as well in the in the contents for for
5:34this class, but in here in the
5:35bibliography you are going to find uh,
5:37different uh, websites where you can
5:40enter to see everything in detail. Maybe
5:44too much detail for for you at this
5:46particular point, but the whole content
5:48of the law, the text is going to be in
5:51there. So, I have been combining both.
5:53So, we have uh you know, what is the
5:55articulation of the law, but in
5:58combination with the practical
5:59application of something like Netflix
6:02that is very well known uh and is one of
6:05the
6:05highest
6:07regarded platforms for content
6:09distribution. So, you are going to have
6:12both of them. So, have a look at this.
6:14This is just a uh an article I created
6:16just very brief just to showcase the
6:19meaning of this Article 50 uh for the
6:22regulation of uh generative AI, which is
6:25more about, as I said, regulating the
6:27transparency more than the technology in
6:30itself. Um with this,
6:33um
6:33I have also uh another more practical
6:37thing. So, I'm going to close this. So,
6:39I'm going to
6:40uh
6:41free a bit of materials.
6:44We are going to get to that in a bit. Um
6:47but, I wanted uh you to have, which is
6:50the main main thing, is this guide that
6:54I have created to showcase the whole
6:57process of disclosure um
7:00just with a legal validity. So, it's
7:02something that you can
7:04uh adopt as a practice for an audit. And
7:07if you know how to use it, it's going to
7:09be really, really easy once you are
7:12required to provide an audit that can
7:14happen at any point uh with with your
7:17client. It's a right of your client just
7:19to understand that you are complying
7:22with the the laws basically because it's
7:25their responsibility as well. So, uh I
7:27put together this guide, and I have in
7:29here the table of contents, so you can
7:32see. You can go directly to to those
7:34pages, but what I what I'm explaining in
7:37here is the whole process of how to
7:41create the audit. And this is going to
7:45explain the whole thing. Of course, I'm
7:47not going to enter into all those
7:49details because it's much easier just to
7:51get the the the document, but I'm going
7:54to to give you also
7:56uh a little trick.
7:59Which is
8:00I have created this guide in a way that
8:04if you load it to your preferred uh LLM,
8:08for instance, um ChatGPT or Cloud,
8:12is is going to uh understand the
8:15instructions and is going to guide you
8:17through the process. Why that thing?
8:20Well, it's because it's actually very
8:22programmatic. So, you just need an
8:25assistant, so the LLM is going to act as
8:28the assistant with this set of rules,
8:30and it's going to be asking you to
8:32provide information, to provide the you
8:36know, the materials to be analyzed, and
8:39is made in order to automate a lot of
8:42the process. Of course, you you have to
8:44disclose manually different things, but
8:47many of the fields that you have to
8:49compile in in the document uh are
8:52actually extracted by the workflows in
8:56themselves. Because remember that the
8:58workflows, and I'm going to show you as
9:00a uh as an example, so let me open this.
9:04This is a workflow from class number
9:07five, I think. So, I'm going to open
9:10with
9:12uh the notepad.
9:13And this is a JSON file. So, it's a
9:15workflow that I exported from um
9:18ComfyUI.
9:19So, just once,
9:22and I'm going to show you in here that
9:24this is the the workflow from ComfyUI
9:28that is uh human-readable. I mean, you
9:30can see what is in there. You can see
9:32that there is a resize image uh node
9:36that is in this position. If I mean, if
9:39you pay attention, is very similar to
9:41the scripts in Nuke. So, you can read
9:44and see the the node. So, something that
9:47that you can that you can see in here is
9:49like there are checkpoints. And in the
9:52checkpoints is going to to give you
9:54informations of models, of links. So,
9:57there are so many things that are
9:59embedded in this file. So, all the nodes
10:02that you're using with all the models
10:05for instance, in this is telling me that
10:07I'm using this Juggernaut XL Ragnarok
10:11safe tensor. So, the LLM can also go to
10:16the
10:17to the internet and to search about what
10:20is the meaning of this model. Can I
10:23find? So, is going to check based on the
10:25instructions that are provided in here
10:28and is is able also to check licenses.
10:31So, if you can comply, if you are using
10:34a
10:35a model that is actual for for the
10:38production. If if not, is going to ask
10:41you, "Hey, do you have the license for
10:43this? And if can you update that? And
10:47I'm provide with that." So, is going to
10:48be a way of assistance because
10:52filling all this documentation is a lot
10:55of bureaucracy, is a lot of paperwork
10:57and is just to extract and to copy data.
10:59So, that's why I created this system
11:02that is just going to help you uh in
11:05order to
11:07assist with
11:08filing paperwork, papers. That is
11:11something that as an artist is not that
11:13the most fashionable part of
11:15of the work.
11:17However, it is important because we need
11:20to comply and because of this tool is,
11:22you know,
11:23can be controversial sometimes. So, we
11:25need to declare and also to show the
11:28authorship of the work. Okay? Up to the
11:31certain extent, understanding as we
11:33discuss in in unit three, the
11:36implications, the legal implications,
11:38and of course the implications moral and
11:41ethical implications, but I am going to
11:44assume that you are going to be the
11:47author, you are controlling what you are
11:49doing, and you are not just copy and
11:51pasting because that is I mean, there is
11:54no authorship in in there. There is only
11:56operation. So, for that is a different
12:00separate argument that I'm not going to
12:02touch in here. What I'm going to discuss
12:05is how
12:06we can do this
12:08this audit in a very easy and practical
12:13way. So,
12:14there are many pages and there are
12:17so many things that it would be really
12:20helpful if you just read it once just to
12:23understand what you are talking about
12:25when we are discussing audits because it
12:27can be scary, but it's actually just a
12:30way of demonstrating that yeah, you are
12:33following
12:34the rules and and everybody is safe. So,
12:37for for that, something that you have to
12:41fill out, there is no way I mean, you
12:44there is a way of doing with the with
12:46the LLM, but it just easier if you go
12:49and open this template that I created
12:52for you. It's a Word document, very very
12:54easy that you just fill in here the
12:56fields. You just put the you know, all
13:00the data that is requesting.
13:02Something that you are going to read in
13:04the guide is that
13:06you have
13:07sometimes these three options. You have
13:11yes, no, and unknown. And trust me, it's
13:17much better if you don't know something
13:20to declare that you don't know.
13:23The program in itself is going to assist
13:26you to get to know, if possible,
13:29[snorts]
13:30um
13:31about that particular question.
13:34But,
13:35be honest, first of all, with yourself.
13:38It's much better to declare something
13:41that you don't know that to put yes or
13:44no without really knowing what's going
13:47on. Because you can just
13:50get in trouble. And you can get in
13:52trouble, get your client in trouble, get
13:54you can confuse a lot and maybe it's
13:56actually not a big deal or actually, by
13:58contrary,
13:59is a big deal. So, it is important that
14:03you are honest when you are answering.
14:05And if you don't know, you don't know.
14:08Um upload this uh document to the LLM
14:12after you train with the with the
14:14instructions that I'm going to to to
14:16give you. Um it's completely fine. It's
14:19going to uh help you to assist you to
14:21fill out.
14:22If at the end of the process, still you
14:26don't know,
14:28no worries. Declare
14:31you don't know.
14:32And of course, maybe someone is going to
14:35get upset because you are working
14:37without knowing an important aspect.
14:41But, it's better to get someone upset
14:43and then maybe deleting one workflow or
14:47adapting one workflow than
14:50getting sued because you did something
14:52that was wrong. It's much better to be
14:55clear.
14:56It's fine. Just That's why I'm always
15:01telling you to check the licenses of the
15:04models you are using so you are
15:06completely safe when you are using
15:08anything. And models are not just the
15:10motors, I mean, the engines of the of
15:13the generation. Also, ControlNet's have
15:17engines, I mean, they have their own
15:19models. Uh LoRAs, everything that has
15:22been drained, everything that is
15:24generating something, you must declare
15:28if you have the license, if you can use
15:30it, and if you have you have to provide
15:32for for that information. So, that is
15:37that is in in in a nutshell what this
15:39document is is for. So, this document is
15:42something that you are expected to fill
15:44by hand. Um, again, this document is
15:47provided with the content materials. So,
15:51let me close this. Um, this file, that
15:54is a text file,
15:56this is the actual prompt.
15:59This is what you are going to tell your
16:02LLM uh,
16:05to provide with every steps
16:09filling the the document that that you
16:11are going to that we are going to to
16:13create the audit. Um, of course, this
16:17must be provided in combination with the
16:22JSON files. So, it depends on on the
16:25platform you're using. I'm going to use
16:27for this example ChatGPT just because
16:30it's the most popular and and it's easy
16:33just to understand the the concept, but
16:36uh, this is supposed to be in the
16:38instructions
16:40of the uh,
16:42the input in the project. So, it must be
16:45embed
16:46uh, in the root of the project you are
16:50creating. So, for this, just to have
16:53like a quick uh, introduction to the to
16:56the system, I'm going to do it just with
16:58a simple chat, but you can implement it
17:02uh, any way you want with any other LLM,
17:05but this is just a way of showing you.
17:08So, let me bring in here
17:12uh, ChatGPT. So, I'm going to close the
17:15guide.
17:18And I'm going to use uh chat GPT in
17:20here. I'm going to
17:22uh
17:23copy and paste this.
17:27And I'm going to create in here, yes, a
17:29very simple uh prompt that I'm going to
17:31say
17:33I am
17:35going to
17:36create
17:38an audit
17:42using
17:44these
17:46instructions.
17:50And I'm going to uh
17:52go below.
17:56So, here.
18:01And I'm pasting the the text. You can
18:04even put the
18:06uh
18:06the text file in in there, whatever is
18:10is more convenient for you. And I'm
18:12going to execute. So, now, in my case,
18:16right now, I have the
18:18high model, so it's just going to think
18:21a lot. Um but it's just going to to take
18:25into account the
18:28all the information that that we have
18:30provided. So, it's now telling you that
18:33we are going to audit each JSON
18:36separately. Uh well, just read. It's
18:39just that simple. That's the purpose of
18:42of this thing. So, I'm going to put it
18:45here, and I'm just going to
18:46uh add this um
18:49this JSON. It's not now in the naming
18:53convention that should be applied that
18:55you are going to have in the in the
18:57guide, but this is just for showing you
18:59how how it works in a in a nutshell. So,
19:03uh I'm going to say start
19:05with this
19:09JSON
19:13file.
19:17So, now it's going to open this JSON
19:20file, it's going to capture the data,
19:23and it's going to start elaborating all
19:25that data according to the instructions
19:28that we provided. So, right now it's
19:31implementing the the workflow, and it's
19:33doing, you know, all the due diligence.
19:36So, this is the the work of of the
19:38assistant. So, that's why
19:41AI in itself is actually something very
19:44helpful because as an artist, this is
19:46part of the job that that I have to do.
19:50As a professional, being an artist
19:51doesn't mean that you are going to be
19:53painting all the time or doing just
19:55creative work. Being a professional
19:57means also administration, and means,
20:00especially if you are a freelance, um
20:03you need to wear many hats. And in here,
20:06this legal thing, imagine doing this by
20:09hand, or just by
20:11it's going to be very time-consuming.
20:13So, the cool thing that I really love
20:16about this is I'm using AI to audit AI.
20:21So, it's a process that I seriously
20:24don't mind that AI is going to steal my
20:28job
20:29>> [laughter]
20:30>> as an administrator of my audits and the
20:33legal stuff. So, I am really happy that
20:36I can synthesize all of this in a way
20:40that is just going to make my life
20:42easier. So, with this, I can focus on
20:45the part of the job that I like the
20:47most, the creative part. If not, you can
20:50have a ratio that you are going to be
20:52working in administration, in in any
20:54other parts,
20:56way more than the actual,
20:58you know, the typical thing that
21:00everybody thinks that you are an artist
21:02and you are all day long creating. I
21:04wish. So, that is that is the thing.
21:07He's like, this is one of those uses of
21:10AI that I really love. So, I don't
21:14agree at all that
21:16AI is just here to kill us all, to
21:19remove my job. Well, it depends on who
21:22you are.
21:24I like to to automate my processes in
21:28order to be able to have more time for
21:31for creation. Of course, it requires
21:34time and investment man, and but at the
21:37end of the day, you have to put things
21:38in the balance, and you need to
21:40understand the benefit that is actually
21:43on on this. So,
21:45uh going back to the the workflow, as
21:48you can see,
21:49it's taking a while because he's he's
21:52doing an assessment quite deep because
21:54the instructions were very very precise.
21:57And this is done
21:58right now for one particular workflow.
22:02So, it's much better to apply to one
22:06workflow at the time because you're
22:08going to have way more control. If you
22:11apply just
22:13the same process but to a whole batch,
22:16the risk is like it's going to treat the
22:20the batch as a unity. So, maybe you're
22:23going to have things that are not going
22:25to be compliant with things that are
22:28compliant, and at the end, due to the
22:30limit of tokens, I mean, the memory that
22:32you're going to be allocated in all of
22:33these, maybe you're going to have um I'm
22:37not going to say errors, it's just like
22:39it's not going to overview certain
22:41things. So, it's always much better to
22:44go one by one. Um you know, right now
22:47I'm just doing this alone, but I could
22:50be doing this while I am actually
22:54performing
22:55you know, I'm not going to say composite
22:57theme on comfy UI, but I will say doing
23:00generations.
23:01I can be using comfy UI in the meantime
23:05this audit is is done.
23:07For a studio format, if you have a
23:11person that is designated to doing this,
23:13you know, it's just going to be doing
23:15this one at a time. If you have
23:16different
23:18different models, you can do sequential,
23:21whatever whatever is your your pipeline.
23:24I'm not entering in there. This is more
23:27operative than
23:28you know, in the instructions of what is
23:31the best way. So, the important is like
23:34at the end you are going to have
23:36a long list of of the audit that you
23:39have just to copy and paste basically.
23:41Or you can even download from from here
23:44if you have like me for instance I have
23:47connected my Adobe Acrobat so I can
23:51generate
23:52a PDF from from the result of this audit
23:57and you can have the you know, the
24:00result in in one document. So you can
24:02put all documents together and you are
24:05going to be sure that the elements are
24:07in there. Something to keep in mind is
24:10like
24:11comfy UI has something that is actually
24:15really cool which is the the content
24:17that you create directly from comfy UI.
24:22It means that it cannot be filtered by
24:24third-party softwares. For instance, if
24:26you
24:27just transcode that with any software
24:31like even Nuke, it's going to lose the
24:34metadata that contains the workflow. So
24:37something that is also interesting,
24:39depending of course on on the agreement
24:42with the client, is that to provide the
24:46actual raw creations, the raw
24:49generations from comfy UI. That is just
24:52going to ensure the JSON file just is
24:57matches with the JSON file embedded in
25:00the image. And
25:02you know, there is a visual audit that
25:05you can have just to prove something.
25:07This would be like a double layer of
25:10verification just to see that is not
25:12just any workflow that created anything.
25:16It's just like the actual images contain
25:19the workflow in itself. So, that is
25:21another layer. It depends, of course,
25:23and that is something that needs to be
25:25agreed. So,
25:27at the end of the day, this is not even
25:29removing the work of
25:31the lawyers, I would say, as licitors or
25:33any any uh legal adviser because it's
25:37their job to understand
25:40everything about the law, but when you
25:43have to fill the paperwork, what you are
25:45doing is just executing instructions.
25:47So, instead of just doing
25:49that yourself,
25:51it's actually the AI doing the
25:54I like to say the the self-audit, even
25:58if it's not self.
26:00But it's it's actually really really
26:02cool that is just visualizing and
26:05getting the data from from that
26:07particular file. So, right now, after 6
26:11minutes and 35 seconds in my case,
26:15this very simple
26:17workflow has been audited.
26:20And you have in here everything
26:23that
26:24was retrieved from from there. It's
26:27giving you already the the ability to
26:30download
26:32a markdown,
26:33which is like a
26:35text kind of a text file that you can
26:38that you can download, and you're going
26:39to have the
26:41the result.
26:43It's telling you this is the most useful
26:46evidence is the the PNG, what I was
26:49mentioning before. It's like if you
26:50provide with that particular image, that
26:54is also going to to provide the the
26:57output, so it can be inspected visually.
27:00So,
27:01it audits can come in very different
27:04ways. I will say
27:07to be open. Be open. If you have nothing
27:11to hide, what is the problem? So, be
27:13open. Provide as much as you as you can,
27:17of course, without overwhelming the poor
27:19people that can that needs to review all
27:22of that, but it is a good practice just
27:25to go and give them
27:27the images, the inputs, the outputs, the
27:31workflows, the audits, the
27:34everything is in there just because at
27:37the end of the day is a collaboration
27:40between you and the machine. So, the
27:43part that is the machine is not yours.
27:46Nobody is attacking you, so you are not
27:49uh the author of the models or anything
27:53that is in there. You are only the
27:55author of the configuration of that
27:57workflow. So, it's how you put
28:01and you assemble the ideas of the
28:04workflow, that is yours.
28:06Uh but the actual content, well, that is
28:10depending on what is the content, but
28:12what I can what I can assure in my case
28:15is like I know very well what is my part
28:19of IP in here, what I am creating, and
28:23what is the part that is not mine, and
28:25what is not mine, I can donate it to the
28:28audit, of course,
28:30uh because it's not mine, and depending
28:32on the license, I can even give it for
28:34free because it's actually free. So, you
28:37are not uh if you are not stealing, they
28:40cannot steal from you.
28:41>> [laughter]
28:41>> It is it belongs to the depending on the
28:44license, it belongs to who it is belong,
28:46but for sure
28:47is not you.
28:49Uh that that is that is something that
28:51we already discussed and of course it
28:53can have
28:54more complicated articulations, but
28:57the most important part that is the
29:00reason I decided to create this bonus
29:02class is that
29:04we are seeing a movement from
29:07governments, from the institutions
29:10that want to guarantee transparency.
29:13And transparency is something that is
29:16beautiful. And I believe transparency is
29:19never enough. So,
29:22don't be afraid of disclose everything
29:24because at the end of the day, this is
29:26what we do as artists. We are sharing
29:29even techniques of all of that. Now we
29:31are in a new era where we have to share
29:34also the honesty of what was me and what
29:37was the machine. And there is no shame
29:39in that because at the end of the day,
29:41we will find a a balance. We will find
29:44an agreement between between society and
29:48the use of of AI in certain
29:51articulations of the industrial
29:54application of art. And I want to
29:57underline the industrial application.
29:59So, is we are creating something that
30:01must be explored at at a industrial
30:04level. So, that's why we are just using
30:06heavy machinery for for generating. And
30:09there is a component of art, of course.
30:11Um that is the part that I love the
30:14most. But for everything else, yeah, we
30:16have we have LLMs and we have generative
30:19AI models and
30:21all those things that there is also a
30:23lot of people working in order for us to
30:26be using that. But that will never
30:28substitute your critical approach and
30:32thinking. So, that's why I see
30:36there is a lot to talk about this and I
30:39just wanted to use this as a
30:40conversation starter. If you have other
30:43ways of doing the the audit, if you have
30:46to something to improve, just share with
30:49the rest of the world, so we can all
30:51guarantee safety by transparency. So,
30:55that's all.
30:57Um so, yeah, I hope you like this
31:00this way that I'm that I'm doing. I'm
31:02putting those materials, so you can use
31:05them. Um
31:07Yeah, if you want to put any comments or
31:10or something in in here, I appreciate
31:13it. If you are interested, the rest of
31:14the classes are already there. You can
31:16even double-check. This is very related
31:18to unit three that we have discussed
31:21plenty of small classes. We have
31:24discussed all the nuances of the legal
31:28the legal framework and also the ethical
31:31implications of all of these, but you
31:34know, once you know what you are doing,
31:36yeah, there is no problem at all and no
31:38fear, and that is the status of grace
31:41that you can act when you are creating.
31:44So, with this I'm going to leave this
31:47class in in here.
31:49I just have to thank you for for
31:51watching this this class.
31:54I see you in the next one. Thanks a lot.

About this transcript

This page contains the full transcript of How to Audit ComfyUI Generative AI Workflows for Film & VFX by Victor Perez's NODE BY NODE, generated from the public captions YouTube serves with the video. The transcript has 4,718 words across 718 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 translate the transcript into 100+ languages.

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, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.