YT YouTube2Text

YouTube2Text → Transcripts → Ingeniería inversa para editar vídeos con Claude

Ingeniería inversa para editar vídeos con Claude — Transcript

by Lorena Bordonaba - Inteligencia Artificial · 2,639 words · 375 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Look, these are 15 videos and they are
0:02all made with AI, including music,
0:04animations, images, sound effects, and
0:07all without opening a single editor.
0:09Today I'll show you how to do this with
0:11Claude, even if you don't know how to
0:13edit. You just need to have access to
0:15Opus 3.5. And to the methodology I’m
0:18going to teach you in this reverse
0:20engineering video. It will work for any
0:22type of editing style. Also, I’ll
0:25leave everything I use in the
0:26description so you can copy and paste
0:28and have all the tools. There are seven
0:31steps, and at the very end, we’ll
0:32turn it into a skill so you can reuse
0:34it. Let’s start with the first step:
0:36the style of the videos we are going to
0:38edit. And for this, my recommendation
0:40is to go into a tool that already has
0:42videos, like Pinterest. You can search
0:45for examples with these words here, "
0:47Real Motion Graphics." Based on this,
0:49ideas like these here will appear. My
0:52recommendation is to save the ones you
0:54like and create your own board to have
0:56different styles and different examples
0:58to use. And you can select as many
1:01styles as you want. Once you have them
1:04chosen and found, we go into Claude,
1:06although you could carry out the same
1:08process from ChatGPT or Claude. And in
1:12this case, we are going to come into
1:13Claude, because here we can always
1:15execute many more functions. And what
1:17we are going to do is open a new folder
1:18. In this case, I name it "motion," I
1:20click create, and we open it. We go
1:23into the chat, select a short reasoning
1:26model so it doesn't consume credits,
1:28and we write a message like this: "
1:31Install gallery-dl and download the
1:33following videos." With this tool, we
1:36will be able to download reels from
1:38Pinterest, Instagram, or any other
1:40visual tool. Next, we paste the links
1:43of the three videos. It’s important
1:46that in this case, they are three
1:48similar styles. You will be able to do
1:49this process for each video style you
1:51are going to ask it to replicate. In
1:54this case, these three follow these
1:56style patterns where many fonts appear,
1:58some images, and they explain some kind
2:00of story. And then we add this here.
2:04Decompose the three videos into parts
2:06and create a context to replicate them.
2:08We press enter and let it do its work.
2:10It has finished, and here we have the
2:13context. In my case, it told me it
2:15already had it; I told it to open the
2:17preview, and here we can see how it has
2:18broken down each of the editing
2:20elements. This is the step that allows
2:24you to break down every element that
2:26makes this video work: how the words
2:28appear, which images show up, at what
2:30frame they are placed, and it is what
2:32will give you that "cloud" style for
2:34your future edits. Importantly, the key
2:37here isn't to copy this video, but to
2:39understand what makes it viral,
2:41understand what's working, how the
2:43video is being generated, and then use
2:45that to replicate it and generate our
2:47own. Once we have that, we just need to
2:50recreate it. But to carry out this
2:52recreation, we must first install some
2:55dependencies and tools to help us do it
2:57. The first, minimal, and necessary
2:59step is to install hyperframes and the
3:01different skills that make up the
3:03hyperframes package. This will help us
3:05with the edits. This is the bare
3:08minimum, but we can expand a bit
3:10further so that the videos aren't just
3:12like this one here which has animations
3:14, but can also include images generated
3:16on the fly and audio generated on the
3:18fly. For images, we can either generate
3:22them on the fly or load them from a
3:23folder. If we want to load images
3:26directly from a folder, all we do is
3:28click "add folder" and browse to the
3:30folder on our own computer where we
3:32have all those saved images we've
3:33generated and want to use for these
3:35Reels. If, for example, you have
3:38recorded a video and in this case you
3:40want to edit yourself, now is the time
3:41to upload it as a resource for
3:42generation. If you are going to
3:44generate everything from scratch, you
3:45can do it in different ways. You can go
3:47into any image generation tool like
3:49Gemini, ChatGPT, or any other, generate
3:52each of the images, download them, and
3:54upload them. You can look for
3:57connectors via API, like this one here,
3:59or you can also look for switch tools
4:01where you pay a subscription and also
4:03connect via MCP. In this case, I am
4:06going to do it via MCP, so what we
4:08would do is make that connection. To
4:10connect it, you would go to "more
4:12connectors,""manage connectors," and
4:14here you would click "add." Whatever
4:16connector you are going to use, it
4:18always works the same way. We paste the
4:21URL, in this case mcp.gfil.eai, give it
4:23a name, and once we have that, we click
4:25"continue." Once we have clicked "
4:28continue," the next step would be to
4:30click "connect." In this case, it
4:32already shows up as connected for me,
4:33and it will ask you to authenticate
4:34with your own account. This process is
4:36always the same. Now, we move on to the
4:39third step and it's time to ask it to
4:41recreate it. For this, you just have to
4:43give it the context we’ve already
4:44generated, our script, and the next
4:46step would be to provide the tools
4:48we've already connected. In this case,
4:50I’m going to ask it to invent the
4:51script to do it quickly, recreating the
4:53style of the analyzed video by creating
4:55it in Spanish with this legendary story
4:56that is very well known. Create the
4:58images, and I tell it to use Nano
5:00Banana based on the Hixfiel MCP.
5:02Perform the editing with Hyperframes,
5:04which we also have, and add sound
5:06effects with media. Which, in this case
5:08, is one of the skills that comes
5:10within the Hyperframe package. Once we
5:12have it, we hit enter and let it
5:14generate. Once it starts, the different
5:17steps it will follow appear here, and
5:19you just have to let it work. Once
5:21it’s finished, Hyperframes has an
5:23interface where it shows us the design
5:25so we can edit it. This interface opens
5:27by default in the browser, and here we
5:29can see the different editing layers it
5:31has created. We can see how the images
5:34and texts appear, and furthermore, we
5:36can tweak, move, or change the text of
5:38each of the different tracks, or any
5:40adjustment we don't like. Once we have
5:44it, we can click the export button up
5:46here or do it directly within Cloud. It
5:49is most common for it to render it and
5:51give us the completed video. In my case
5:53, it didn't ask for anything; it gave
5:55me the video directly, and here it is.
5:57Here we can see how it appears and how
5:59the different elements pop up. It looks
6:01a lot like one of the examples we
6:03provided. And up to this point we have
6:05the videos, but we are missing
6:06something important, which is the voice
6:07. And for the voice, just as we did
6:09with the images, we can record our own
6:12audio, upload it, or we can also
6:13connect a tool to generate it. And this
6:16is where the sponsor of this video
6:17comes in, which is Fish Audio, and it
6:19is the one I used to generate the
6:20voices for these videos. This is the
6:22Fish Audio platform, and what's
6:24interesting is that you can register,
6:26and once we are inside their interface,
6:28you can use it directly from here. You
6:30can clone your own voice or choose one
6:32of the available voices, as there is a
6:34super extensive catalog for all
6:35languages, including Spanish. And here
6:37you can use your voice or search for
6:40any of the voices that are already
6:41registered. There are many, and what's
6:44interesting is that there is a large
6:45marketplace in Spanish. Once you have
6:48it, you can generate text-to-speech
6:50directly here, or what I like and get
6:52the most out of, which is this
6:54developer section here, where you can
6:56get an API key and connect it within
6:58Claude. Once you are inside Claude, you
7:01just have to ask it to do the
7:02generations for you and it does
7:04everything for you. This is one of the
7:06interesting things, it adds all the
7:08tags so it sounds completely authentic
7:10with my own voice. Artificial
7:12intelligence is not magic. It is key to
7:15understand its infrastructure and not
7:17get carried away by the hype. And now,
7:19my honest opinion is that the
7:21generation speed it has is incredible.
7:23You click add tags, it adds them in 2
7:25seconds, you click generate speech, and
7:28the speech is right here below. If you
7:30do it from Claude, it is exactly just
7:32as fast. And if we talk about cost, we
7:34are talking about prices that are super
7:36low. In this case, ElevenLabs has a
7:39cost of $ 10 per million characters,
7:41which if we translate it to having made
7:4315 videos, the 15 reels are less than 7
7:46cents. And in this case, they all have
7:48a duration of more or less between 40-
7:50something seconds. Let's look at this
7:52last case. Here we can see the cost of
7:55a single video, and in this case, it
7:57was 0.01. The truth is that it is a
8:00ridiculous price, it costs practically
8:01nothing. And in this case, it is in
8:03dollars, so if we are talking about
8:05euros, it is even less. Also, if you
8:07want to try it, now is the time because
8:09you can sign up totally for free. You
8:11have my link down here and you get 50%
8:13off the ElevenLabs model for the
8:15different Starter, Creator, and Pro
8:18plans. Additionally, you can also use
8:20it directly via API, as I have done,
8:23and pay exclusively for the cost of
8:25what you consume, which in this case
8:27was 0.07 cents. All you have to do is
8:30top up dollars and from there start
8:32using them from Claude, ChatGPT, or any
8:34other tool. I'll leave the link to the
8:37application and the registration
8:38directly down here in the comments and
8:40in the description. Let's continue with
8:42the video. And once we have the image,
8:44we have the style, and we have the
8:46audio, we can move on to packaging all
8:47this together. And so you can see some
8:50more examples, I'll show you two very
8:52different styles.
8:53>> I asked Claude to correct two mistakes
8:55for me. I've been trying for 40 minutes
8:57to stop it from changing my life. The
8:59message was, "Hi, Ana, I'll send you
9:01the quote tomorrow. Let me know. Only
9:03spelling, please." This right here is a
9:06style where I've recreated HTML
9:08animation, and in this case, it's pure
9:10motion graphics generated with
9:11Hyperframes. This means we haven't used
9:15any image generation model. I’ll show
9:18you the other example, where I uploaded
9:20a video of myself and asked it to do
9:22the same thing, but based on an editing
9:24style. In any case, the important thing
9:27is that we land on this reverse
9:29engineering methodology: find the
9:31reference, extract the contact, and
9:33from there, start editing by giving it
9:35the necessary tools. But now, let’s
9:37move on to the fifth step: we can
9:39package all of this into a skill to
9:41have consistency and go from one video
9:43to fifteen, making it all more
9:45mass-produced. And for that, we can do
9:47it with a prompt. Package the process
9:50we have carried out as a skill to be
9:52able to generate more videos from a
9:53script. with this style and methodology
9:56. So, based on this, we will be able to
9:59have the skill ready to simply name it
10:01and generate it. An important nuance:
10:03it would be appropriate for the
10:05Anthropic Skill Creator to generate
10:07this skill for you, as it will help you
10:08in this process and help you package it
10:10much better than if we did it ourselves
10:12. Once we have the skill created, we
10:15can go into the folder, and here we can
10:17see within Ocean what we have generated
10:18, the references, and we also see that
10:20inside the cloud point we have the
10:22skill. Important: this skill is saved
10:24by default in this project scope, in
10:27this folder we have started. If we want
10:29it to be installed locally, that is,
10:31for all devices, all folders, all
10:33projects, we have to tell it this here.
10:35Install the skill in a local scope.
10:38From here, it will read it and install
10:40it. Once we have it installed and it
10:43gives us the check that it's ready, we
10:45can just do this: write the name we
10:47gave it, in this case, Motion Graphic
10:49Reel, and from there we would provide
10:51the script for our next video and just
10:53click generate with the script, and it
10:55would start the generation with the
10:57same style. I’ll show you some of the
10:59styles I’ve created. We’ve finished
11:02the six steps so far, but we have one
11:04last step left, which is the
11:05difference-maker and helps us achieve a
11:07slightly more professional result.
11:10Because when we start generating all
11:12this in bulk, there are also small
11:13elements that we have to keep in mind
11:15along the way, such as sound effects or
11:17some music libraries. If, for example,
11:20we don't want to generate every single
11:22image, that's where these tools come in
11:23. To generate sound effects, my
11:26recommendation is to use media.io. In
11:29this case, it's a specific skill from
11:31deeper frames that helps us add various
11:33click or swoosh effects, or other
11:34things that help make the video much
11:36more engaging. The next thing we need
11:39to keep in mind is that we have already
11:41installed this tool here, gallery-dl.
11:44This tool allows you to create a folder
11:47like this one here—I call it "reels
11:49resources"—where you can save
11:51different images, icons, and elements
11:53that you like and want to add to your
11:55creation. So, once you have everything
11:58generated, you just have to copy this
11:59link right here. Important: the board
12:02must be public. You come into Claude,
12:05paste it, and ask it to download all
12:07the content. You can also tell it to
12:10save them in the project folder where
12:12we're working and start reusing them in
12:14all the mass videos you're going to
12:15generate, so you avoid using the exact
12:17same things repeatedly. You can do the
12:20same with audio libraries. And in this
12:22case, one of my recommendations is
12:24Pixabay, which is this one here. We can
12:26copy and give the link to Claude and
12:29ask it directly to use their API, which
12:31is completely open, to download the
12:33different audios that fit our videos,
12:35effectively delegating that entire part
12:38of audio editing and generation to it
12:40as well. By giving it these additional
12:43resources, we can now make the leap
12:45toward greater autonomy and delegate
12:47more of the execution. What helps
12:50Claude become much more creative and
12:52write better scripts is for us to do
12:54this reverse engineering process
12:56ourselves, package it into a skill, and
12:59give it to the AI directly so it does
13:01exactly what we want. Alright, that's
13:04it for this video. You now have the
13:06seven steps and all the tools we've
13:07been looking at. I’ll leave
13:09everything for you down here in the
13:11description. And if you liked the video
13:13, don't forget to subscribe and hit
13:15like to stay up to date with more
13:17content like this.

About this transcript

This page contains the full transcript of Ingeniería inversa para editar vídeos con Claude by Lorena Bordonaba - Inteligencia Artificial, generated from the public captions YouTube serves with the video. The transcript has 2,639 words across 375 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.