ChatGPT и Claude монтируют рилсы и ютуб за меня. Пошаговая инструкция, как заменить монтажеров

0:00Right now , every reel is telling us : "

0:02Now ChatGPT , just like Claude , can edit

0:04your videos . "

0:05I’ve calculated the specific

0:07difference in price and time for

0:08editing reels between a real person and

0:10the Claude AI .

0:11I’ll show you how I make 15 reels in

0:13a single evening .

0:13And in this video , I’ll provide a

0:15step-by-step guide on what you need to

0:17do to start editing your videos with AI

0:19. What types of AI-assisted editing are

0:21there , anyway ? Which ChatGPT and Claude

0:23plans will you need to edit at least

0:24one video ? And in the end , we’ll

0:26conclude whether it’s true that video

0:28editors are no longer needed and AI

0:29does everything for us . My name is

0:31Alexander Croft . I’ve been blogging

0:33for 12 years . I have over 400,000

0:35followers on social media . And in the

0:37last year , I’ve been actively

0:38integrating AI into my business

0:39processes , blog management , marketing ,

0:41sales , and even my daily life . That

0:43includes learning English . Instead of a

0:45human assistant , I now have an AI

0:46assistant . And I will be sharing cases

0:48like these in all my future videos . So ,

0:50I identify four main use cases for

0:52using AI in editing . The first and

0:55simplest one is what we’ll be looking

0:56at today . This is where you do almost

0:59nothing ; you just upload the necessary

1:01materials to Claude or ChatGPT , give

1:03instructions by voice , meaning you

1:05interact with the material minimally

1:07and get an edit . You look at what

1:10turned out , make edits , and then get

1:11the final version . The second method is

1:14the professional one . This is where we

1:15connect the same Claude and ChatGPT

1:17directly to professional software . That

1:19includes Adobe Premiere and , for

1:21example , DaVinci Resolve , where AI

1:23controls your editing software and has

1:25a much wider range of tools . And most

1:28importantly , the editor doesn't just

1:30get a finished result that is hard to

1:32influence ; they can make their own

1:33adjustments and fix micro-details . And

1:35most likely , I will film my next video

1:38about this . There are more and more

1:40editors appearing now where AI is built

1:42directly into the interface from the

1:43start , which implies that we work with

1:45the AI first and then finish things up

1:47by hand . Again , I will film an overview

1:50of such programs in the near future .

1:51The third way to use AI for editing

1:53concerns not the editing itself , but

1:55the creation of B-roll and infographics

1:57in video generators . Such animations

1:59allow you to make your videos much

2:01richer . For this , an image is generated

2:03, usually in ChatGPT or Midjourney . And

2:05then it is animated in Kling , Luma , or

2:07Google Veo . And the last , fourth type

2:09is video-to-video . This is when you

2:11take a finished video file and use an

2:13AI to replace specific elements . You

2:16can change clothing , change the

2:17interior , or even change camera angles

2:19that weren't there originally . In other

2:21words , you create a multi-camera edit ,

2:23even though you only shot with one

2:24camera . But today we are looking at the

2:26simplest , most basic type of editing ,

2:27which is actually necessary for most

2:29people . To do this , first , we obviously

2:31need to install either ChatGPT or

2:33Claude . Next , we move on . If we are

2:35working in ChatGPT , we go to the Code

2:37Interpreter section . If we are working

2:38in Claude , we go to the Claude Code

2:40section . And here , for our models to

2:42learn how to edit , we need to install a

2:44few tools . I will briefly talk about

2:46each tool so you understand what is

2:47under the hood and why these tools are

2:49needed at all . The first one is the

2:50media engine . It is a large media

2:52powerhouse called FFmpeg . It allows our

2:54AI models to work with video material ,

2:56because neither ChatGPT nor Claude can

2:58handle video on their own . And FFmpeg

3:00is the engine for the AI that allows it

3:02to actually edit video , trim it , change

3:04the bitrate , change the aspect ratio ,

3:06and change the frame rate . So , FFmpeg

3:08is actually a very powerful tool that

3:10you will likely need for many other

3:12things in the future . By the way , it is

3:14worth noting that YouTube , Netflix ,

3:15Instagram , Facebook , and many other

3:17social networks and media resources

3:19operate using this engine . The next

3:21thing we need to install is Whisper .

3:22Whisper is a local LLM model that

3:24weighs 3 gigabytes , and it serves one

3:26function . It converts audio into text .

3:28This model is necessary because AI

3:30models are not like humans . They don't

3:32cut the audio track by ear ; they cut

3:34based on the text . First , they get the

3:36audio track extracted from your video ,

3:38transcribe it into text with timestamps

3:40, look at what is extra or where to cut

3:42, and then proceed to edit the final

3:44file . Well , the Whisper model is

3:46actually very versatile , and it allows

3:48you to do things like this . I

3:49practically don't type on the keyboard

3:51anymore ; I just speak everything , and

3:53it all gets converted into text .

3:55Whisper is running under the hood here ,

3:57and with its help , you can convert any

3:58audio or video files into text . For

4:00example , you had a call with a client

4:02and want to get a transcript of that

4:04call . You upload the call to your AI ,

4:06and with the help of Whisper , it will

4:07turn it into text , mark the timestamps ,

4:09highlight key points , and all this is

4:11possible precisely thanks to this model

4:13. And these two tools will allow you to

4:15do the editing when you have your 47

4:17takes , when you have constant stumbles ,

4:19when you have pauses , when you say " um "

4:21— the AI cuts all of that out , and you

4:23get a clean , ready-to-go final video .

4:26And the AI actually handles this task

4:28almost perfectly . In other words , we

4:30can confidently hand over all the rough

4:32editing to neural networks and get an

4:34almost ideal version as a result .

4:35Sometimes I don’t even edit anything

4:37at all and just leave it as it is . But

4:40we need a third tool for creating

4:41infographics , because a clean ,

4:43high-quality video is certainly good ,

4:44but it will be incredibly boring if we

4:46don't have some additional elements to

4:48hold attention . And so , for this , we

4:51need to install one of two elements .

4:53Either Hyperframes or Remotion . I'll

4:55say right away that most people now

4:56prefer working specifically with

4:58Hyperframes , because it was originally

4:59adapted for working with agents . It’s

5:01lighter , and the most interesting thing

5:03is that it was released by the company

5:05HeyGen — the company famous for its

5:06digital avatars . And that is exactly

5:08where the difficulties begin , because

5:10all infographics are , in reality , a

5:12matter of taste , and they will depend

5:14very heavily on how well you can

5:16explain to the AI what exactly you want

5:18: what design , what fonts , what styles ,

5:20and what animations . And all of this

5:23will depend both on the model you are

5:25using , on how good it is , and on how

5:27well and correctly you described your

5:28wishes to the model , and how detailed

5:30and accurate your guidelines were . And

5:33this is precisely where AI messes up

5:34the most — or rather , it's not the AI

5:36itself that messes up , but the AI

5:38operators , because in this case ,

5:40everything depends on how well you

5:41perform your own work . For example ,

5:43look at these two identical videos . At

5:45first , I made this video , and I

5:47didn’t like it because there were too

5:48few elements in it . Well , from an

5:50illustrative standpoint , I needed there

5:52to be as many as possible — more

5:54pop-ups , fly-ins , banners , and

5:56everything else . I provided a new

5:57prompt and got a completely different

5:59result . So , in this case , it will

6:01depend on how you construct your

6:02prompts and then how you turn that

6:04prompt into a cool , powerful skill that

6:06you can combine to ensure you have a

6:08repeatable , stable , and consistent

6:10design time after time , so that nothing

6:12changes from video to video and you

6:13always have a recognizable style . Yes ,

6:15to install all three of these elements :

6:17FFMPEG , Whisper , and Hyperframes , you

6:19actually just need to enter this one

6:21command . I will provide it to you in my

6:23Telegram channel . Accordingly , you type

6:25it in , your AI will ask for your

6:26permission a few times , and it will

6:28simply install everything for you by

6:29default . The next step is to prepare

6:31all the necessary material , put it all

6:33into a folder , and give it to the AI .

6:34So , here is what the material for one

6:36of the reels looks like . And perhaps

6:38you came to this video from one of

6:39these reels . There . So , there are seven

6:41takes here . And the total duration of

6:43these clips is , I think , over 7 minutes

6:45. Although , in the end , this reel only

6:48took up 1 minute . There are logos here ,

6:51there is sound design , and accordingly ,

6:53I then pass this folder on to the AI . I

6:56go to CLDcode or Codex and select the

6:58necessary folder here . In this case , it

7:01is " Editing 2 " . I upload it and then I

7:04write a prompt about what I want to see

7:05and what type of edit I want to get in

7:07this case . And here , besides the model

7:09itself that we choose , as I said , it is

7:12very important what we tell the AI .

7:14Because in this case , a person who

7:15understands editing , who can describe

7:17their desires , convey them to the AI ,

7:19and explain what they want , will

7:21obviously get a much better result . And

7:23, as I say , the final result will

7:25depend on how correctly you compose the

7:27brief , that is , how correctly you write

7:29the prompt . You can grab the prompt I

7:31composed and work with in my Telegram

7:33channel and , accordingly , make your own

7:35adjustments to it later . When you

7:36experiment , when you try it out and you

7:38like something or dislike something .

7:40You can tell the AI directly by voice :

7:41" Look , I want to fix these kinds of

7:43errors in this prompt . This , this , this

7:45, and this were done . I want it

7:46differently , I want other colors , I

7:48want other things . " And you simply

7:50dictate it by voice . Accordingly , the

7:52AI creates a new prompt , you copy it

7:54into a new session , provide a new

7:55folder there , and , accordingly , send it

7:58for editing . After that , the AI starts

8:01working , and actually , the time to

8:03execute and create your video can be ,

8:05uh , very different because it also

8:06depends on the power of your computer ,

8:08since a lot happens on the resources of

8:11the Mac itself in this case , because

8:13things like audio-to-text conversion ,

8:15processing , and subsequent stitching ,

8:17video , and editing — all of that is

8:18done by the computer's power . And

8:22sometimes , actually , even my computer

8:24sometimes turns on the fan because the

8:26load is quite high . Your final edit

8:28will depend precisely on this . And

8:32obviously , the more effects you specify

8:34— for example , in this case , my prompt

8:36included a request for words to appear

8:38behind my head . This means the neural

8:42network has to find the person ,

8:44rotoscope them — yes , create a mask —

8:46and overlay text , infographics , and so

8:49on behind the person . And this also

8:51takes up quite a lot of resources . And

8:53then , in this instance , Claude produces

8:55all these different sketches . It

8:58watches , checks itself , and ultimately

9:00delivers this kind of file . Once you

9:03have received the final edit , you can

9:05make adjustments . That is , you can

9:06explain to the neural network what

9:08needs to be done and how to change it .

9:09For instance , at the thirty-fifth

9:11second , I don't like how the word flew

9:12in . At the fortieth second , I want us

9:14to replace , for example , the types of

9:16infographics and change them to

9:17something else . At the fiftieth second ,

9:19I want to change this , right ? So , you

9:22explain it in a list — 1 , 2 , 3 , 4 , 5 —

9:24and send it back for re-editing .

9:27Basically , this is actually the maximum

9:29amount of control we have .

9:31Unfortunately , we cannot — we don't

9:33have the ability to manually tweak

9:35anything . And as I said , this is a

9:37fairly primitive and the simplest type

9:39of editing . And , if we are talking

9:43about how good it is — if we were to

9:45sum it up — I can actually compare

9:46neural networks to the advent of phone

9:48cameras . That is , while previously , to

9:52be a photographer , you first of all had

9:54to own a camera — and , in essence ,

9:55photographers were distinguished from

9:57everyone else by the fact that they at

9:59least possessed the equipment — when

10:01decent cameras appeared in phones , the

10:03ability to take photos became available

10:05to absolutely everyone . But

10:07professional photographers can still

10:09create amazing photos on a phone , while

10:11those who just use the phone camera ,

10:13naturally , won't reach that level .

10:15It’s the same thing here . In this

10:17case , the neural network is just a tool

10:19. And naturally , if a person is a

10:20professional — if they understand the

10:22basics of editing , typography ,

10:24infographics , how to combine fonts ,

10:26what colors are appropriate , how to do

10:28things in general , and what kind of

10:29effects exist — then , of course , they

10:31will be able to get much more out of

10:33this tool . And if we are talking

10:35specifically about this format — the

10:37most primitive one , where we have

10:38minimal ability to do anything by hand

10:40— then I can say that , overall , it's

10:42already quite decent . And for the

10:45majority , this level is probably above

10:46average compared to what most editors

10:48on the market are doing . Therefore ,

10:50objectively , we can already say today

10:52that neural networks are gradually

10:54beginning to displace video editors .

10:56This is my subjective opinion , but I

10:58see it in the example of other fields ,

11:00including design . The designer I worked

11:02with for many years . Now I reach out to

11:04her much less often . And among other

11:05things , she receives fewer and fewer

11:07orders , because many have already

11:09replaced design . Yes , it won't be

11:10something masterful , but it will be

11:12above average . And so I see the same

11:13thing happening in editing now , because

11:15neural networks are already starting to

11:16push this profession out from the

11:17bottom . But , as they say , neural

11:19networks won't replace experts . Neural

11:21networks will replace those experts who

11:22don't know how to use neural networks .

11:24And in fact , using neural networks for

11:26editing is just one way to optimize

11:28your routine , because now AI allows you

11:30, for example , to analyze competitors

11:32in tens of minutes rather than a week ,

11:34as it really took before . Neural

11:36networks allow you to collect an

11:37infinite number of ideas for your

11:39content in Reels , posts , Telegram , X ,

11:41Threads , create scripts for YouTube

11:43videos , conduct entire large-scale

11:45studies on specific topics , and

11:46automate business routines , all reports

11:48, P & L , cash flow statements , and all

11:50necessary things . If you happen to keep

11:52these , then neural networks handle them

11:54simply perfectly . Creating websites ,

11:56creating carousels , or , let's say , you

11:58have some ideas and you don't know who

12:00to think through them with or how to

12:02develop them . With neural networks ,

12:04using certain skills , you can pull a

12:06lot of different information out of

12:07yourself . You can directly create and

12:11brainstorm with neural networks and

12:12extract your ideas and turn them into

12:14finished content or some business

12:15projects that you might have been

12:17planning to launch for a long time . And

12:19I teach all of this in my monthly

12:21master group . Therefore , if you want to

12:23use neural networks to the fullest and

12:24integrate them into content creation ,

12:26your business processes , and your daily

12:28routine in general , write to me on

12:29Telegram , and I will send you the

12:31details . And lastly , probably the most

12:33interesting part , which plans are

12:34suitable for editing your videos .

12:37Actually , the news here isn't very good

12:39, because for example , the Plus plan on

12:41ChatGPT — the $ 20 plan — isn't capable

12:43of editing even a one-minute video .

12:46That is , I have 7 minutes of raw

12:48material to start with . Yes , I have a

12:52fairly complex prompt where , as I

12:53mentioned , things like text elements

12:55popping up behind my head and so on

12:57require about two , sometimes even three

13:00five-hour sessions , five-hour limits ,

13:02just to edit a single one-minute video .

13:06Therefore , in this case , my setup

13:08handles it , of course , much better . It

13:10edits several videos in a row , but I

13:12have the 200 - dollar plan there . But on

13:14the other hand , as I always say , it’s

13:16not a question of the subscription cost

13:18, but a question of how much this work

13:20is actually worth . Because to edit , for

13:22example , a video like this , it would

13:23already cost , well , at least 20 dollars

13:25. And I certainly make a huge number of

13:27such videos a month on the 200 - dollar

13:29subscription , plus additional work . So ,

13:31everyone has to make that decision for

13:33themselves . And all the information

13:34from this video , all the prompts ,

13:36everything you need to install , is in

13:37my Telegram channel . The link is in the

13:39description . And see you in the next

13:40videos . y