YT YouTube2Text

YouTube2Text → Transcripts → إزاي تسبق ٩٩٪ من الناس في الذكاء الاصطناعي وتعمل فلوس مع وائل

إزاي تسبق ٩٩٪ من الناس في الذكاء الاصطناعي وتعمل فلوس مع وائل — Transcript

by Wael El-Selouky · 12,509 words · 1,908 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00I'm late to AI, meaning today, at the
0:01moment I'm watching this video, I'm
0:03just now going to learn AI, while a
0:04million people are ahead of me in AI.
0:06>> That's what people think. This whole
0:07thing is like the stock market. I know
0:08people who do it, I do it myself. I
0:10mean, I'm making money from this stuff.
0:11I would have been sitting with you
0:12right now. So I'm telling you, this
0:14exists. I taught my friends about this.
0:16Before I taught them, they used to tell
0:17me, "That doesn't exist.""No one gets
0:18those numbers.""No one makes that kind
0:20of money.""No one is going to pay for
0:21those things.""It's not a real thing,
0:23it's impossible." They started working.
0:24With me. And none of them want to work
0:26with anyone else but me. They are now
0:27convinced that, yes, this is the future
0:29. That’s the trick right there.
0:30That’s why you should learn AI.
0:31There’s a huge misconception among
0:33many people, thinking that. Claude is
0:34replacing people. And Claude does this
0:36and that. And ChatGPT. Look guys, this
0:38is literally a masterclass in image
0:39generation. If you apply what I just
0:41said, you will generate images just
0:42like mine, and even better. Ready to
0:43learn AI?
0:44>> Let’s go learn AI.
0:45>> I was just sitting with Wael. And he
0:47taught me everything I need to know. So
0:49I can start in AI, and get ahead of 99%
0:52of everyone else. And if you don't know
0:54Wael, he's one of the first people who
0:56started with AI in the entire Middle
0:58East. Three years, a million...you know
1:00Wael from the video in front of you
1:02right now. And this was his course
1:03announcement video, which just came out
1:05very recently. The same video went
1:07viral, and was able to get millions of
1:09views in just a few hours. And what you
1:10don't know about Wael is that he has
1:12worked with more than 200 clients.
1:14Doing AI-powered work for them. And
1:15these clients include many big names,
1:18like Al-Shifa Honey, Saudi Al-Hilal
1:20Club, 1980, High Point Furniture, and
1:22many other names. This is what made
1:24Wael one of the people who was able to
1:25make millions in his very first year in
1:27AI. And just as he is also famous for
1:29giving workshops to companies, to teach
1:31them AI. So today, he's doing the same
1:33for us. After I saw his course, I
1:35called him and told him to come
1:36immediately. With the intention of
1:38making a video to teach you how to make
1:39videos with AI just like he does. But
1:41after we sat down together, I
1:42discovered that he was giving me a full
1:44masterclass in AI. In short, he
1:46explained to me. Everything about AI
1:48from A to Z. With all the models, and
1:50everything I need to get started.
1:52Regardless of what I’m going to
1:53execute. So if you feel like you're
1:55late to AI, and you want to start
1:56learning it in detail. And to get ahead
1:58of more than 90%of the people already
2:00in the field, you have to watch this
2:02video. Before we start the video, I
2:03want to tell you that we are holding a
2:05competition. You could potentially win
2:07about 200 dollars. This competition
2:09will be among people who know how to
2:11make videos using AI. Quite simply, we
2:13will give you a specific topic. And you
2:15have to make a video of up to 30
2:16seconds using AI that covers this topic
2:18. After that, we will make a video on
2:20YouTube watching all the people who
2:21submitted their work. We will bring in
2:23an expert to decide which video is best
2:25and who will win. And we will send 200
2:27dollars to the winner. Keep in mind, we
2:28will try to repeat this competition
2:30every week or every two weeks. The
2:31competition rules are very simple.
2:33First, you must be subscribed to the
2:34channel, and you must have liked this
2:36video. Second, go down to the video
2:37description, and you will find a link
2:39to register for the competition. In
2:40that link, there will be all the
2:41information you need to start working
2:43on your video. Work on the video using
2:45AI only, and upload your work. And
2:47hopefully, look forward to the day of
2:48the competition. If you want to start
2:50making money, join our AI competition.
2:52>> Today I will tell you how to get ahead
2:54of most people and even ahead of AI
2:56itself in this era of AI.
2:57>> That’s great, and I will explain it
2:59in six steps. The first step is the
3:01importance of AI: why you should learn
3:04AI. The second step is the pillars of
3:06AI, whether it’s tools, what it
3:08generates for you, what you use it for,
3:10use cases, all that stuff.
3:12>> Great.
3:12>> That sounds good. The third thing is
3:15how to use AI for prompting and
3:16references. Okay. The fourth step is
3:19choosing the right models and merging
3:21between them. This is something many
3:22people don’t do and don’t know
3:24about. The fifth point is how to bypass
3:26the limitations of AI. How AI has
3:27limits. How to get past them with
3:29workarounds. Well, number six: AI is
3:31expensive and costs money. So I will
3:33teach you how to save credits and spend
3:35less money while using AI.
3:36>> This is the most important thing.
3:38>> This is the most important thing
3:39because its cost is very high.
3:40>> And the seventh and final point: how to
3:42actually make money with everything you
3:44learned, not just talk. I’ve learned,
3:46but how will I make money? How to
3:48monetize; I won’t say "make money,"
3:50I’ll say "how to monetize." Because
3:52you are learning a skill and how to
3:54make money from it. How do you monetize
3:56that skill? You’ve got me excited to
3:58start this topic.
3:58>> Let’s go.
3:59>> Let’s go.
4:00>> It will be an intensive video, but I
4:01will explain everything in it.
4:03>> Let’s go.
4:03>> Let’s go.
4:06>> I have a question now: why is it
4:07important at all for me to learn AI?
4:09>> That’s a great question. Many people
4:11want to learn AI but don’t know why
4:12they are learning it in the first place
4:13and what it will benefit them. What
4:14does AI do? AI makes it so that instead
4:16of you, Wael, doing, for example, X
4:18amount of work. It lets you produce 10X
4:19. So, consequently, you are amplifying
4:22your existing skills. You are
4:23amplifying the work you are doing. So
4:25you can achieve an equation. What is
4:26this equation? The well-known business
4:28equation for this triangle is that in
4:29any business you do or any service you
4:31provide, you can achieve two out of
4:32three: either your quality will be high
4:33, and you'll deliver it in little time.
4:35Therefore, your price will be expensive
4:36. Or, for example, your price will be
4:38low, and the quality is high, but. You
4:39will provide it in a very long time. Or
4:41it's a low price and a short time. So
4:43the quality will be the lowest. That's
4:44your normal work without AI. But what
4:46does AI do? It comes to break this
4:49triangle. Why? Because first of all, AI
4:53lets you do your work in much less time
4:55, okay?
4:56>> Because if I were doing my work
4:57manually—I used to be a 3D artist,
4:59for example—the video that would take
5:01me a month, I can now do in an hour.
5:03>> Much less time. Do you know how much
5:05you've saved? I'm not talking about
5:07saving an hour or two; you're saving
5:09days and months. That's an insane
5:11amount of time.
5:12>> In every field.
5:13>> In every field. In any field: photos,
5:15videos, graphic design, video editing,
5:17content creation, automation. If you
5:19work in anything, AI will help you.
5:21Sales, anything, okay, that's great
5:23regarding time. But what about quality?
5:25A huge problem people fall into is "No,
5:27that looks like AI,""No, what's that
5:29quality?"
5:29>> Honestly, with my mediocre use of AI,
5:31the things that come out, I'm not
5:32always satisfied with them.
5:34>> True, that's real, and I faced this in
5:35the beginning when I submitted work and
5:37people would say, "No, this is AI, it
5:39looks like AI, and so on." I invested
5:41all my time over the last two years
5:42into learning how to use AI so the
5:44output doesn't look like AI.
5:45>> So that people can pay me money,
5:47they're getting value. And I want to
5:48compete in the market; I want to put
5:50out very high-quality work. I love my
5:52work, so my idea isn't just AI; my idea
5:54is that Wael puts out really strong
5:55work. But I want to use AI to be able
5:57to make more money. There are
5:58workarounds that we will get into in a
6:00later point on how to use AI to make
6:01work that doesn't look like AI.
6:03>> Meaning if the quality coming out of
6:04the AI is bad, is it the model's fault
6:06or my fault?
6:07>> No, if the thing coming out of AI is
6:09bad, the mistake is on your end.
6:11>> Okay.
6:11>> AI doesn't make mistakes.
6:13>> Nice.
6:13>> It's true that AI has context and
6:15memory. But if you give it everything
6:17correctly and know how to use it, it
6:18will give you what you want. AI has a
6:20fixed principle: Garbage in, garbage
6:22out. As you give, you receive. Well
6:25said. Okay, what about the price?
6:27>> You can't convince me that the price of
6:29AI is cheaper than anything else out
6:31there.
6:31>> It's very, very expensive.
6:33>> Right, and to use it.
6:34>> Look, the pricing part is very tricky
6:36and has a big question mark over it.
6:38>> Because we are comparing it in the
6:40Egyptian market, or the Arab market in
6:41general, which isn't used to the prices
6:43of normal work. Especially the things
6:45that are done on a laptop, the virtual
6:46stuff. I mean, let's compare right now,
6:48for example. Let's say you're going to
6:50make a product animation video. Would
6:51you do it in 3D, like CGI for instance?
6:54Or would you do it with AI? Sounds good
6:57. So, what’s the normal price for
6:58this one? And what’s the normal price
7:00for this one?
7:01>> Um.
7:02>> Usually, the AI video price will be
7:05cheaper for the client to pay. If they
7:09pay you $ 500 for this, they might pay
7:11you $ 300 for that.
7:13>> Okay.
7:13>> For the same video and quality.
7:15>> The client, the client.
7:16>> The client, yes, will pay me that price
7:18. That’s great.
7:19>> People think, "Hey, in 3D I used
7:22Blender, which is a free program."
7:24>> Or, "I used Maya or 3ds Max, I
7:26downloaded the cracks, so there’s no
7:28cost."
7:29>> So it costs nothing.
7:30>> Yeah.
7:31>> Is that how you think?
7:32>> Right.
7:32>> But the cost isn't calculated that way.
7:35How is it calculated? This 3D video
7:37took me how long? It took me a month.
7:40>> Good.
7:40>> And this AI video took me how long? One
7:43day. It took me one day.
7:45>> Okay.
7:46>> Who is going to pay you for the month?
7:47That time you wasted, your time is
7:49money. Something many people don't
7:51understand, guys. You have to realize
7:53your time is money. Even if you are
7:55providing a service, your money isn't
7:57just the final video. The time you put
7:59into learning and the time you spent
8:01creating the final product is what
8:02drives up your price.
8:04>> Yeah, because in that month, if I spent
8:06a month using free tools, I spent money
8:09on food, drink, and electricity, and I
8:11don't have time to get another client.
8:15>> Unlike AI. I might have paid money for
8:17the tool, reducing my profit. But
8:19it’s only a day. Meaning I can get
8:21another one tomorrow. Exactly, that’s
8:24the point. Another point, actually. 3D
8:26programs and Adobe programs—which
8:27you’ll edit on anyway, like Premiere
8:29and After Effects. To make a clean 3D
8:31video. You will use tools.
8:33>> Adobe subscriptions alone are $ 100. Or
8:35$ 200 or more. Hundreds of dollars a
8:37year. Regardless of what you pay per
8:39month. Maya and Autodesk subscriptions.
8:41These subscriptions are very, very
8:42expensive. They cost a lot of dollars.
8:44So, in reality. The software you are
8:47using costs a lot of money.
8:49>> But because we are in the Arab world,
8:51people download cracks and think it's
8:53free. These things aren't free at all.
8:55When you compare the price of these
8:57tools to AI tools, you'll find that AI
8:59tools are actually cheaper.
9:01>> Okay.
9:01>> You see, the difference is that the AI
9:04tool charges you per use.
9:05>> Yeah.
9:05>> But this one charges you by the month,
9:07and you use it as much as you want.
9:09>> That’s the difference.
9:10>> So here is the tricky question. How do
9:13I price this? What if it doesn't come
9:14out right the first time? Then the cost
9:16increases for me. That’s where the
9:18craftiness comes in. That’s why you
9:20learn what? How to reduce the cost. So
9:22that you earn more from it. And you
9:23know how to produce something better.
9:25>> Meaning there are two people working in
9:27AI, both know how to do it, both are
9:28doing the same thing, but one does it
9:30cheaper than the other.
9:31>> They do it as a cost.
9:32>> It's cheaper in terms of cost and
9:34credits because in AI, it's cheaper
9:35since they understand the tool.
9:37>> Let me give you a great example.
9:39There’s a contest called the Global
9:41Film Festival. This contest was on a
9:43platform. Everyone knows it, and you
9:44surely know it too. The green platform
9:46called Luma Dream Machine. Okay. This
9:49platform tells you to make a short
9:50movie. Meaning, in a range, for example
9:53, from three. Minutes up to twenty
9:54minutes. And enter this contest. The
9:56first place gets a million dollars. So
9:58many people applied, about 6,000 films
9:59were submitted. So when you look. The
10:01average is from five to 20 minutes.
10:03People making films and such. It shows
10:05you how many credits they used to
10:06produce this film.
10:07>> Every project is open to everyone, so
10:09you see every generation. I did some
10:11good research, and I looked at all the
10:13films released. To see how many credits
10:15they used. I’ll tell you the average
10:17cost in money, because people won’t
10:18know how to calculate credits, since
10:20every platform has a different price. I
10:22, for example, participated in this
10:24contest; I made my film 14 minutes long
10:27.
10:27>> 14 minutes is a lot, by the way, for an
10:30AI video, 3D video, or anything; this
10:33video cost me how much? It cost me
10:363,000 EGP.
10:37>> 3,000 EGP.
10:38>> To make a 14-minute short movie.
10:40What’s in this short movie? There are
10:42aliens. There are serious VFX. There is
10:44consistency in everything. There is
10:46music, songs, and sound effects. There
10:48are different worlds. There are
10:49locations. There are actors. There are
10:51emotions. There are so many things.
10:53Well, normally, to shoot a film with
10:54actors for those same 14 minutes, how
10:56much would that cost with this quality?
10:58>> A single location would be over 30,000.
11:00We could show shots from trailers to
11:02see the quality you produced; normally,
11:04something like this would cost you
11:05millions of dollars.
11:06>> I achieved quality, for instance,
11:08similar to Avatar quality. In certain
11:10points of the film and such. With
11:11something like this, what are you
11:13talking about? You are talking about
11:15millions, millions. Of dollars, you
11:17make it for how much? 3,000 EGP.
11:20Egyptian pounds, not dollars. So, which
11:23is cheaper or more expensive? The
11:25regular way is you’re talking about
11:27millions of dollars, but this is 3,000
11:29Egyptian pounds.
11:30>> So AI in high productions is much
11:32cheaper. But you need to know how to
11:34save on it. There are people who made a
11:36five-minute film, for example; I made a
11:3715-minute one. People made five minutes
11:39, five minutes, how much did it cost in
11:41credits? When I calculate it in
11:42Egyptian, it cost them 100,000 EGP. So
11:44you made about a third of the time I
11:46made. Yes, exactly a third. A third of
11:49the time I made, and paid triple the
11:50money I paid. Can you imagine that
11:52number? So, this depends on how clever
11:55you are.
11:55>> If you know what you want and know how
11:57to use the right AI, know the right
11:58models, know prompting techniques and
12:00workarounds, you can significantly
12:01reduce the cost of credits for yourself
12:03.
12:03>> Nice.
12:04>> Great. Now back to our main topic: why
12:07you need to learn AI to break the
12:09triangle of price, quality, and time.
12:12>> Good.
12:13>> So what does this let you do? It lets
12:15you—see, this is Wael, right? And
12:18these are what? These are the other
12:20people. People available plus Wael with
12:23AI alone, multiply him by 100,
12:25literally, how much is he worth? He's
12:28worth a company of, say, 30 or 40
12:30people. You get what I mean? He's
12:34bigger than that, so you alone are
12:36worth all those people. "I'm late to AI
12:41.""I mean, today, at the time I'm
12:43watching this video, I'm just now going
12:45to learn AI.""There are a million
12:46people ahead of me in AI, right?"
12:48>> Very good. People think this is like
12:50the stock market.
12:50>> How so, for example? You find a stock
12:53that was going up like this, okay. Then
12:55it dropped really, really, really low.
12:56So it's known that it will go up nicely
12:58. So it rose up to here. The stock is
13:00here. I am someone who started this
13:01from the very beginning, started from
13:03...Three years ago, back when AI had no
13:04control at all. The topic, so I started
13:06here at this point. And because I did
13:07that, I am sitting with you now. I knew
13:09how to make money, and I knew how to
13:10work in this field. Why? Because when
13:12it grew, I developed along with it. And
13:13I took advantage of that exploitation.
13:15Okay, where are we now in 2026? It’s
13:17October, we’re at this point,
13:19basically. I, I took advantage of this
13:21part. This is the part I took advantage
13:23of. But, Wael, should I learn AI now?
13:25Do you think we are going to drop again
13:26? Or do you think it will go back like
13:28this or go down? That is wrong. It
13:30won't happen. So what will happen? It
13:31will keep going up. It will rise. It
13:33will rise. It will rise.
13:35>> We still haven't reached the peak of
13:36intelligence.
13:37>> We haven't reached the peak at all.
13:39This peak is something called AGI,
13:41Artificial General Intelligence. After
13:44we reach it, you and I and everyone
13:46else won't find work, and there will be
13:48unemployment in the whole world. This
13:50is real, seriously.
13:52>> But we are still far from that point.
13:54>> Not that far, not that far, but we are
13:56far.
13:57>> Okay.
13:57>> I am now, I am now offering you an
13:59opportunity. I said, "Wael, if you
14:02learn AI now in this window, within two
14:04years you will make 2 million pounds."
14:06>> Um.
14:07>> Would you refuse it?
14:08>> No, of course not.
14:09>> You wouldn't refuse it, but I’d say,
14:10"I don't know what the world will be
14:12like after two years." So, for you, is
14:13it better to make 2 million pounds and
14:14see what happens after two years, or
14:15not make any money at all and just see
14:17what happens in two years?
14:18>> No, let's work.
14:19>> Take advantage, right? When I started,
14:21I was working in 3D. I spent three
14:22years learning it. AI came in; I got
14:24depressed. I felt like I learned
14:25something for nothing. And people won't
14:27ask for it. Because AI does it much
14:28better, faster, cheaper, and all that
14:30stuff. What did I do, man? I'm going to
14:32exploit AI and I started. To get into
14:34AI and I made money from it. If I had
14:36closed my mind and just put an X over
14:38my brain. And I said no, I learned 3D
14:40and this and that. I didn’t, I
14:42didn’t adapt to the market. So
14:43consequently, I won’t be able to make
14:45money. Why would people come to me?
14:47>> Yes.
14:47>> For you to make money, it’s literally
14:50one sentence. Opportunities. I seized
14:53opportunities before others, so I made
14:55money. I am telling you, whoever is
14:57watching this video, you are in. An
14:59opportunity exists right now. It’s
15:00available for up to two years. There is
15:02no such thing as being too late. Take
15:03advantage of it during these two years.
15:05And after the two years, we’ll see
15:06where AI goes. But we are still in it.
15:08>> Very nice, alright, so come on,
15:09introduce me and teach me AI now,
15:11let’s go.
15:12>> Let’s go learn AI together.
15:13>> Ready to learn AI?
15:14>> Let’s learn AI, let’s go learn AI
15:16together and understand the basics and
15:18fundamentals first.
15:19>> Nice.
15:20>> Understand its foundations, and from
15:21there we move to the advanced stuff.
15:23>> So let’s understand its foundations.
15:24How to write here like this. AI pillars
15:27. What can AI do for me? AI is
15:30something that runs on a computer. Or
15:32on a laptop, it can. Currently do
15:34everything. So what comes out of the
15:36laptop? What are the outputs that can
15:38come from a laptop? First thing it can
15:41produce is text, right?
15:43>> It can produce images. Right?
15:48>> Nice.
15:49>> Okay, what else can it produce? You
15:51tell me.
15:52>> Videos.
15:52>> Videos, right.
15:54>> And audio.
15:57>> It can produce sound, like music, sound
16:00effects, or voiceovers, right, any kind
16:03of sound, is that it?
16:07>> I mean, I could say programs, websites,
16:09all that stuff, the whole internet
16:11world.
16:12>> Code. This code, but this code is a big
16:14word, what falls under it?
16:18>> Many things fall under it, for example,
16:21files; any file is basically code,
16:23right? Websites, correct.
16:27>> Sites, programs.
16:28>> Applications, apps, fall under it.
16:30>> Nice.
16:31>> Right?
16:32>> Is that all?
16:33>> Certainly not.
16:34>> Are there more things? Anything can be
16:36done.
16:37>> Right.
16:37>> Okay, great, so now I know the outputs
16:39that AI gives me.
16:41>> Nice.
16:41>> If there is one that gives me all these
16:44outputs and has infinite memory, a
16:46memory that doesn't end, then we have
16:48reached something called AGI, which is
16:50what they say in two or three years,
16:52God knows, by then there will be no
16:54jobs.
16:54>> It’s one place that does everything.
16:56>> It does everything, Artificial General
16:58Intelligence. Something that just does
17:00everything. So, is there a model that
17:03does all these things at once? And has
17:05infinite memory and produces amazing
17:07things? No, there is no such thing. But
17:09there are models that excel in
17:11everything. For text, you’ll find a
17:13very good model. For images, you’ll
17:15find different, very good models. For
17:17videos, same thing. For music, songs,
17:19code, and all that.
17:21>> Um.
17:21>> We hear a lot of things. We hear "
17:23Claude,""ChatGPT,""Claude is connected
17:25to my washing machine and washes my
17:26clothes,""The internet says Claude is
17:28the best one and ChatGPT is trailing
17:30behind."
17:30>> There is a misconception among many
17:32people who think Claude replaces people
17:35, that Claude does this and ChatGPT
17:37does that. Look, guys, anything Claude
17:39can do, ChatGPT can also do.
17:41>> Okay.
17:42>> Why do they keep saying Claude is
17:43better and not ChatGPT? It’s just
17:45marketing, nothing more—provocative
17:47things and marketing.
17:48>> Didn't they say Claude is specifically
17:50for coding?
17:50>> No, there’s no such thing.
17:52>> No, that’s nonsense. Both of them do
17:54the exact same thing. Every once in a
17:56while, a company releases a model that
17:57is stronger than the other. So the
17:58company releases a better model. They
18:00both do the same thing, but there are
18:01differences in features. But before we
18:02get into that point, let’s take it
18:04one step at a time. What are Claude,
18:06ChatGPT, Grok, Gemini, and those things
18:08anyway? All these things fall under
18:10something called—let's make a box
18:12here and call it "LLM." What is this
18:15LLM? Large Language Model. This Large
18:17Language Model, what should you
18:18consider it? Consider it the brain.
18:20It’s a brain. It is the brain of AI.
18:22It can think, setting aside how it
18:23works, token-based, prediction models,
18:25and so on. It’s not important to know
18:26those technical details that aren't
18:27necessary. The important thing is to
18:28know that this is the brain. Inside
18:30this brain, there are many models.
18:31There are models better than others.
18:32ChatGPT, Claude, OpenAI, Anthropic;
18:34these are companies offering the same
18:36thing, they offer it to this day.
18:37Claude is better in one model, and the
18:39other is better in a different model.
18:40So what do they both do? They do the
18:42exact same thing.
18:42>> Okay.
18:43>> Cool. My personal choice is that I use
18:45ChatGPT for a reason.
18:46>> Really?
18:47>> For a reason. I am subscribed to both,
18:49but most of the time I use ChatGPT.
18:51There is a reason for this: ChatGPT
18:54comes as a package. What is this
18:56package? It comes with an image
18:58generation model inside, a model that
19:00makes images.
19:01>> Okay.
19:02>> Claude doesn't have that part.
19:03>> It doesn't do that at all; there is no
19:05image model in Claude.
19:06>> Claude does not make images.
19:07>> Hmm.
19:08>> And if you ask it for an image, it will
19:09make it for you in code, yeah.
19:10>> Meaning, it will output a 2D image like
19:12one you’d make in Paint or programs
19:13like that, you know what I mean?
19:15>> ChatGPT outputs actual images for you.
19:17And it works on images. Everyone knows,
19:18surely you’ve used it before. I mean,
19:20we see people on Instagram now, sitting
19:21in their homes. And they post a picture
19:22of themselves at the beach and all that
19:23. They aren't at the beach, nor
19:24anything like it. And it’s very
19:25obvious that it’s AI. Nobody should
19:27do that, guys.
19:27>> Just one I put there.
19:28>> Nobody should do these things; there
19:30are much better ways to make an image
19:31not look like AI, and ChatGPT isn't one
19:33of them. Okay, so these are the LLMs.
19:35>> Cool.
19:35>> I chose ChatGPT because it's the same
19:37price as Claude but offers extra
19:38features, so that's the choice. Both do
19:40the same thing. But is Claude better
19:42than ChatGPT at certain things?
19:43Claude's model is actually better than
19:45ChatGPT when it comes to coding; it
19:47really is better at coding.
19:48>> Okay.
19:48>> But not as much as you think. You know,
19:51it's like talking about two students
19:54who both graduated top of their class;
19:56one got 99.9 and the other got 99.89.
19:59>> It’s not a meaningful difference.
20:01>> A difference that isn't worth it at the
20:02point we are at. Before, it might have
20:04been worth it. But right now, at the
20:06point we're at. The point that. We're
20:07at now, it's not worth it at all. Both
20:09will build you the same website. Both
20:11will build you the same application.
20:12Both will give you the same automation.
20:13>> So why are you subscribed to both if
20:15they are the same?
20:15>> I'm subscribed to both because I teach
20:17people, and I need to show them there
20:18is no difference, so it’s forced upon
20:19me. But I don't want to subscribe to
20:21both; normally, I would just subscribe
20:22to ChatGPT.
20:23>> Okay.
20:23>> Fair enough. But will I subscribe to
20:26just ChatGPT forever? No, AI evolves
20:28every week. So, as it evolves every
20:31week, I see what's new.
20:32>> When Anthropic releases a new model and
20:34says Claude is much more powerful than
20:36ChatGPT, I'll switch to Claude.
20:37>> Okay.
20:38>> The week after, if a better model is
20:40released for ChatGPT, I'll switch to
20:41ChatGPT. I subscribe based on what's
20:43released.
20:44>> Whichever is best; I know what to
20:45subscribe to, when, and where.
20:47>> Great, so the first element in AI is
20:49the LLM.
20:50>> Exactly. And what I'm teaching you now,
20:52Wael, will benefit you generally,
20:53because the things I'm saying today
20:55might be outdated in a month in terms
20:57of models. That's why I'm not giving
20:58you specific model names now; I'm
21:00explaining concepts.
21:01>> Later on, I'll tell you the top models
21:02and everything. But for now, I'm
21:04explaining concepts. A month later,
21:05there will be a new model. It will
21:06overshadow these current models. So you
21:08don't go about it like that. Meaning, "
21:09I will only subscribe to this." No. AI
21:11is not like that. Part of learning AI
21:13is understanding AI research. The new
21:14model that is released and its features
21:16. Searching on YouTube or Google. All
21:17of that stuff.
21:18>> Nice.
21:19>> That's fair. So our first pillar is the
21:21"LLM." This is the brain, as we said.
21:24GPT, Claude, Gemini, Grok, all of those
21:27things. Image generation models, right.
21:32Image generation models are well-known.
21:34You'll find that everyone has used what
21:36? Midjourney. And you have GPT Image 2
21:38and you have, you have many. People
21:40think that image generation models are
21:42just models like Midjourney or GPT.
21:44That is wrong. That is wrong. Image
21:46generation models are much bigger than
21:48that. Let's. Open a new page. Talk
21:50about image generation. And then go
21:51back to this paper. We have image
21:53models that fall under two things. The
21:56first. Thing is diffusion models. What
21:58are these diffusion models? These AI
22:00models are diffusion models. Not
22:02ChatGPT. Not Nano-Banana. Not those
22:05things, these are dumb models. Very
22:06dumb. I mean, literally, they're dumb.
22:09They can't generate your own likeness.
22:11They can't generate your same outfit.
22:12They can't generate the same product.
22:14They can't generate the same location.
22:16They can't do these things.
22:17>> Um.
22:17>> But they have an advantage that other
22:19models don't have, the diffusion models
22:21. Their images are unnaturally
22:23realistic. Very realistic, you see. You
22:25see the Hexfield ads. You see the
22:27influencers on Instagram, the foreign
22:28ones. The ones that show, for example,
22:30a girl taking a selfie. A foreign
22:31influencer taking a selfie. And a
22:33foreign male influencer. And they are
22:35actually AI.
22:36>> They are made with diffusion models.
22:37>> They don't look like AI at all, they
22:39don't look like AI in the slightest.
22:41>> But if I uploaded a picture of myself
22:42and asked it to convert it to AI, it
22:44would look like me.
22:44>> That's not how it works; that's what
22:46these diffusion models are.
22:47>> They just create something.
22:48>> They output an image from a prompt.
22:50>> But you can't send them your face;
22:52there's no control.
22:54>> You only use text, write a prompt, or
22:56get an image.
22:58>> That's as far as you can go. But you'll
23:00get a very realistic image. And beyond
23:03being very realistic, you'll get an
23:06image I call very aesthetic. What does
23:08very aesthetic mean? Very aesthetic
23:10means it has great visual appeal. As in
23:12design principles, composition, and
23:14basics, the image is very beautiful. In
23:16terms of colors and color grading, the
23:17image is very nice.
23:18>> Okay.
23:19>> So what do people do? People use these
23:20diffusion models to impress you. Look
23:22at how beautiful images AI can generate
23:23. Yes, it's very nice for an image. But
23:25the process is very easy. Write a
23:26prompt, it will give you an image. You
23:28didn't really do anything. You just
23:29used diffusion models. So what are
23:30these diffusion models like? There is a
23:32model on Hexfield called Soul 2.
23:34>> Um.
23:35>> There is another one called Soul Cinema
23:37, and there's one very famous one
23:39called Midjourney.
23:40>> It was the first one to generate images
23:42, from the very beginning.
23:43>> Are there only these? No, there is
23:45Recraft and several others, but these
23:46are the three most famous, the three
23:48most important ones. These two models
23:50are generally exclusive to Hexfield
23:51only.
23:51>> Okay.
23:52>> They aren't available on anything else.
23:54One of the reasons I subscribe to
23:55Hexfield is the availability of these
23:57models. I can't subscribe to anything
23:58else because I need these models. Fair
24:00enough. Meaning Soul 2, Soul Cinema,
24:02and Midjourney on its own. The platform
24:03on its own. That's great. But Wael, why
24:05are there two? What's the difference
24:07between this and that? When do I use
24:08which? It's in the name: Soul 2, Soul
24:10Cinema. I expect Soul Cinema is used
24:12for what kind of images?
24:13>> Cinematic.
24:13>> Exactly, images that feel like they're
24:15taken from a movie frame.
24:16>> Ah.
24:16>> And Soul 2?
24:17>> It feels like images straight out of
24:19Pinterest, clothing brands, people;
24:21Soul 2 is very good for Pinterest-style
24:23images. Okay.
24:24>> Fair enough. So you can produce a very
24:26beautiful image. Some people stop right
24:29here. No, that’s not it. We have
24:32another type of model called
24:33autoregressive. What are these
24:35autoregressive models? They are models,
24:38unlike the others, that are very smart.
24:41This model understands, thinks, and
24:42knows what it’s producing for you.
24:44Its prompt adherence—what does prompt
24:46adherence mean? It means the rate at
24:48which the image turns out exactly as
24:49you wrote, which is much higher than
24:50diffusion models. With a diffusion
24:51model, I might describe things, and
24:53you'll find it gets a bit messy.
24:54>> Um.
24:54>> Autoregressive isn't like that. It
24:56produces what you want very easily.
24:58That’s why most people use it. It’s
24:59easier for me to get what I want and
25:01everything. And it can edit images for
25:02me. So, for example, it takes a picture
25:04of you and me. I tell it, "Let's swap
25:06seats." So it swaps our places. "Change
25:08his t-shirt to black.""Make him look at
25:10me.""Make me, for example, punching him
25:12.""Do any of these things." So it can
25:15perform edits. It can edit. This one
25:17used to take just text prompts. This
25:20one takes prompts plus references. What
25:22does reference mean? I can upload my
25:24photo, and it edits it. I can send it
25:26an image of a t-shirt to put on someone
25:27. All that kind of stuff. So, what is
25:29the problem with these autoregressive
25:30models? Why not use them all the time?
25:32Well, it's...
25:33>> Their problem is that the images look,
25:35well, not realistic.
25:36>> Is Nano Banana Pro going to come up now
25:38?
25:38>> I'm getting there. I will tell you
25:39their names. Not realistic, they look
25:41like AI. When someone makes an image
25:43with ChatGPT, I know it's ChatGPT.
25:44It’s very obvious, that’s their
25:46problem. So now I have these two models
25:48. Like what? Like what, for example?
25:50Examples of autoregressive models,
25:52well-known examples, like Nano Banana,
25:54just as you said.
25:55>> Because that description you gave is
25:57the description of all the images it
25:58produces.
25:59>> Anyway, you have GPT-Image. That’s
26:02the latest one released. But whatever.
26:04Guys, anyway, there's GPT-Image 2.5. By
26:07the time this video is released, maybe
26:083 will be out. Maybe 4 will be out.
26:10It’s the same principle. Are there
26:11other things besides them? There are
26:12other things. These have advantages and
26:14disadvantages. These have pros and cons
26:15. There are other things that just came
26:17out that are amazing. Like one called
26:18Flux 3, from a company called Black
26:20Forest Labs.
26:20>> That company is amazing, I love them.
26:22There’s another one called Ideogram
26:244.5; they do the same thing.
26:26>> Okay.
26:27>> But there is a downside, and I will
26:28send you a video to use as b-roll here:
26:30Nano Banana and ChatGPT, when you edit
26:32an image more than twice, the quality
26:33starts to drop. You must have noticed
26:35this.
26:35>> Try editing an image 10 times.
26:37>> You know. You know Van Gogh’s Starry
26:39Night paintings and things like that.
26:40You'll find the image turned completely
26:42into a Van Gogh. The texture is all
26:44wrong. And everything is ruined. Your
26:45face changed. All that stuff. This is a
26:47problem with these models. Flux 3 and
26:49Ideogram were released a week ago. What
26:51are they doing? They solve this problem
26:52. You can edit the image 200 times and
26:54the quality will stay as is.
26:56>> Okay, I actually stopped using the same
26:58image output from AI as a reference
27:00again because it makes me look old; you
27:03only have two tries at most.
27:05>> Anything more than that, the quality
27:07will be ruined. These ones, no, you're
27:09free to do as you please. They do it
27:10using a new technology that prevents
27:12the quality from dropping. So now, Wael
27:13, what about image generation? We know
27:15this is a masterclass in image
27:16generation. If you apply what I just
27:18said, you will get images just like
27:19them. And even better than them, too.
27:20What should I do, Wael? Logically.
27:22Based on what I explained. What do you
27:24think is the best way to get a
27:25realistic image and control it?
27:27>> I will use two.
27:28>> You'll combine them, plus...
27:30>> I will use both. I'll get the nice
27:32image using diffusion models, and I'll
27:35take that nice image to an
27:36autoregressive model like Flux 3 or
27:38Ideogram.
27:39>> Um.
27:41>> So this nice image, it's beautiful,
27:43I'll work on it; it will be beautiful
27:45and under control.
27:47>> I'll start it there first, then...
27:49>> I'll generate a beautiful image from
27:51Flux 2, then send it to Flux to swap
27:53the character to you and change the
27:55outfit and such. So, I have a beautiful
27:57image with my appearance, the outfit,
27:58and everything—realistic, not looking
28:00weird, and the quality hasn't dropped.
28:02Do you get the process?
28:03>> That's a great idea.
28:05>> To merge the models. I know the pros of
28:06this and the pros of that. I know the
28:08cons of this and the cons of that. So I
28:10get exactly what I want. If I'm doing
28:11something cinematic, I get it from Flux
28:13Cinema. The same principle. We arrive
28:15at our important point. That we must
28:17know two things about every model. Its
28:19pros and cons, its strengths and
28:21weaknesses.
28:22>> From that, I merge them; that’s what
28:24makes you good at AI—merging them.
28:26Another use that people might not know
28:27about—and this is a trick you won’t
28:29find in any YouTube video, nor has
28:30anyone explained it before, this is
28:32exclusive—these autoregressive models
28:33, for image generation, have very cool
28:35uses.
28:35>> I personally use them for presentations
28:37. Sometimes when people make a
28:38presentation, they do it manually. Or
28:40when they tell Claude to do it, Claude
28:41does it with code. So it makes you a
28:43presentation with code. A presentation
28:44made with code has limitations. What
28:46are its limitations? That it won't be
28:47able to draw anything for you. If it
28:49puts an image in, it won't know how to
28:50add effects to it. It won't be able to
28:52do that. Because as code, there are
28:54limits. It's not an image; it's code.
28:56>> Yeah.
28:56>> But as image generation, no, it...
28:58>> It has no limits in drawing for me. It
29:00can make the presentation look like a
29:02sketch. It can create any shape.
29:04Therefore, I can create a presentation
29:05from it. Because it is a very smart
29:07model. If I told it, "Look, I'm doing
29:09an episode with Wael El-Solouky talking
29:11about AI." Make me a presentation about
29:12topics one, two, and three. It will say
29:14, for example, "I'll make it 30 slides.
29:16" These 30 slides talk about this, this
29:18, and this. Each slide has this in it.
29:19I'll say, "That's very good." I like it
29:21, or I change what I see fit with it.
29:23I'll say, "Alright, that's great, make
29:24it." But before I tell it to create it,
29:25I send it a reference image, for
29:26example, a presentation from Pinterest
29:28that I liked. I tell it I want it in
29:29the same style and colors. And these
29:31fonts I provided to it. I give it the
29:32fonts, their names, and so on. And I
29:34give it the reference and tell it. It
29:35tells me fine. I tell it to start
29:37creating it with image generation, not
29:39code generation. Complete control, so
29:41you'll find the presentation comes out
29:42with a design that is awesome, really
29:44awesome. So it starts, what? As an
29:48image? This is talking about Payload.
29:52It can generate 10 images in a row.
29:54People think it only generates one
29:56image, but no. You can generate in
29:58batches. If you tell it to do a batch,
30:00it will generate 10 for you at once.
30:02Its maximum is 10; you might say, but
30:05my presentation is 30 slides? My dear,
30:07it's fine; generate 10, then 10, then
30:1010; it will stay focused, it will
30:12remember. I mean, finish the 10 and
30:13tell it to continue with the next 10.
30:15It will do it very normally.
30:16>> All of them, and even the page
30:18numbering will be accurate and
30:19everything; you take all these images,
30:21assemble them into a PDF, you've made a
30:22presentation, you did it in how much,
30:24design-wise, and you finished it in
30:25half an hour, less than half an hour.
30:27>> Um.
30:27>> This presentation manually would take
30:29two days.
30:30>> At the very least.
30:31>> See?
30:31>> So when we talked about quality, time,
30:34and price, it's cheaper for me, faster
30:36for me, and higher quality.
30:38>> Certainly higher.
30:39>> You understood it; so it became Wael
30:41times 10.
30:42>> Image generation is done; with the same
30:44principle we discussed in image
30:46generation, we'll do the same in video
30:48generation; are there types and such?
30:50No, there aren't.
30:51>> There are good models and bad models;
30:53forget about the bad models, don't
30:55waste your time and money on bad models
30:57; the best model available for videos
30:59right now is called Seedance 2.5.
31:01>> That's the best model; how do the
31:03videos work, boss? I provide it with
31:05your image as a character sheet.
31:06Meaning, I don't just put your face. I
31:08put it from the front, from the side,
31:09and a full body. So it will have
31:10context for everything. That's the
31:11problem with videos. It says my face
31:13looks right from the front but wrong
31:14from the side. Because you didn't give
31:15it your face from the side. Or, my
31:16t-shirt is right but the print is wrong
31:18. Because you didn't give it an image
31:19of the print up close. Give it the data
31:21and the location of this studio. And
31:23then I tell it I want Wael El-Sokky
31:24sitting on this chair. And this one
31:25sitting on this chair, and they are
31:26talking to each other. And then Wael
31:27El-Sokky stands up and punches him.
31:29Write exactly that in your prompt. So I
31:31put a reference, which is this person
31:32sitting in this place with this guy.
31:34And I wrote a prompt for one, two,
31:36three to happen. The idea is just that
31:37the video prompting is a bit stronger
31:39or more detailed. You describe several
31:41things. So what do you describe? You
31:42describe animation. Okay. Animation of
31:44what, then? Animation of several things
31:46. Camera and people. Each one, I say
31:48what the camera is doing. Is the camera
31:50fixed, or does the camera go with me to
31:51the right? The camera moves.
31:52>> But that requires something from me too
31:54; it wants me to be a director myself
31:56so I can describe how the sun looked,
31:57where the lighting came from, and what
31:59the camera view was like. That's a lot
32:01of work.
32:01>> Right, see what you just said? It wants
32:03me to say what the lighting was like
32:05and what the camera was like—you're
32:06imagining it in your head, meaning
32:08you're visualizing how you want the
32:09video to turn out.
32:10>> Okay.
32:11>> If you imagine it, can you write it?
32:13>> Write down every detail then.
32:14>> How do you write it? I'll tell you; you
32:16won't need to write every detail. We'll
32:18go back to what? We'll go back here to
32:20the brain, the LLM.
32:21>> I'll talk to it just like I'm talking
32:23to you. "Look, I want to make a video,
32:24and here are the references.""I'll make
32:26it 30 seconds.""I want this, this, and
32:28that to happen.""And the camera will be
32:29one, two, three." I'll talk to it in
32:31Arabic, and it will write it for me in
32:32English. When I give ChatGPT or Claude,
32:34for example, certain skills involving
32:35direction and camera angles, it
32:37understands them. It already
32:38understands them anyway. So, I take my
32:40simple, non-professional phrasing—
32:41which lacks good direction—and it
32:42turns it into something very clean. It
32:43becomes very good direction. So, I get
32:45a great prompt out of it. Then I take
32:47the prompt from the LLM and type it
32:48into the video model. And generate the
32:50video; it gives me the video I want.
32:52What if it doesn't come out right?
32:53Everything in it is correct, but
32:54something went wrong. I go back to it
32:56and say, "Everything happened fine, but
32:57one, two, three happened, and I want to
32:58fix it." It adjusts the prompt, and I
33:00go back again.
33:00>> Do I adjust it based on the first video
33:02reference that came out, or do I start
33:03over?
33:04>> You start over and regenerate. Don't
33:05send the bad video; the bad video is
33:07just bad. But this is video generation;
33:09naturally, the better you get at video
33:10generation, the more you need to have
33:12ideas, understand direction, and all
33:13that stuff. You need to...
33:14>> The person themselves needs to be
33:15skilled, it's not just the AI tool
33:17doing the same thing.
33:17>> No, the AI tool produces amazing things
33:19.
33:19>> Um.
33:20>> If the video turns out bad, the problem
33:22is with you, not the AI. It needs
33:23practice and it needs learning. That's
33:25why you have to learn AI. Okay, same
33:26thing applies beyond videos. Then
33:29there’s music, voiceover, and all
33:31that. For music, we have something
33:33famous called Suno. Did you see the Red
33:36Bull Salsalona episode with Amr Mostafa
33:38when he used Suno? He was using Suno
33:40for the songs. You go in and describe
33:42the song you want. And the instruments
33:44you'll use. If you were a musician,
33:46you'd know how to describe it better.
33:47>> Certainly, of course.
33:48>> If you don't know how to describe it as
33:50a musician, I'll go back to the LLM and
33:51say, "I want to make a song like this,"
33:53and it will describe it better for me.
33:54What if I still don't know how to
33:55describe it, but there's a song I like?
33:57I'll send it to it and say, "Make a
33:59song like this."
33:59>> And it will write me a prompt.
34:01>> So you won't get stuck if you have a
34:02reference and you know how to talk like
34:04a human being—you know how to talk,
34:05I'm not telling you to write prompts,
34:06just talk, you know how to describe
34:08what you want, and it will give you
34:09what you want. And with practice,
34:11you'll build up a repertoire of
34:12information and terminology you can use
34:13.
34:14>> Nice.
34:14>> That applies to music, and to voiceover
34:16. There’s another model called 11Labs
34:18, but just for very simple things. What
34:19about coding? Coding lets you create
34:21website files and applications. You
34:23have ChatGPT or Claude. You tell it,
34:25for example, "I want to make a website
34:26that does one, two, three." It will
34:27generate it for you. But where is the
34:29problem? What will the website look
34:30like? That is point number one.
34:31>> Very true.
34:32>> Number two, now I’ve made a website
34:34that appeared on Claude; how do I send
34:35it to people, or how will people access
34:37it? I don't...
34:38>> So, to make normal websites and
34:40applications, you must already be
34:41someone who understands these things.
34:44>> You have the basics of web development;
34:46for example, you understand what
34:48frontend is, what backend is, what a
34:50database is, and what deployment is.
34:52>> AI won't replace the need for knowledge
34:54; it replaces the need for execution.
34:56>> Exactly, yes.
34:57>> But the knowledge must be there.
34:59>> I must have the knowledge to guide it.
35:00>> If I don’t understand web dev, how
35:02will I guide it? I wouldn’t even
35:04understand what it’s doing.
35:05>> But are you required to understand how
35:07every programming language works? Do
35:08you know how Python works? Do you know
35:10how to write Python? Do you know how to
35:12write Next.js? HTML? You’re not
35:14required to. But I understand that the
35:15frontend—the outer appearance—is
35:16the frontend. It consists of things
35:18like HTML and CSS. And there are other
35:19languages for it, like Tailwind. I just
35:21understand that information. The
35:22backend, which is the function; when I
35:24press a button, it takes me somewhere
35:25else to work. Okay, I understand that
35:27point. The database, which is: now
35:28someone filled out a form on the
35:30website, where will it be saved? It
35:32must be memory; it is my memory, my
35:34storage, and deployment is getting it
35:36displayed on the internet so people can
35:38access it.
35:38>> And I know deployment services like
35:40Vercel or Netlify, and these things I
35:42can connect to ChatGPT or Claude, so I
35:44tell it to automatically do them
35:46together, but I understand what is
35:48happening.
35:49>> So I can direct them.
35:50>> As long as I understand, I can do
35:51whatever I want. Not "understand deeply
35:54," just understand the basics. And the
35:55more you work, the more you learn and
35:57the better you get. Do you get the idea
35:59? What does AI do? It allows the person
36:00just starting, the beginner, to jump
36:02straight to becoming a senior. You know
36:04what I mean, to execute ideas.
36:05>> Because they know how to execute right
36:07away and much faster, and that’s the
36:08whole idea, you know what I mean? And
36:10it makes the senior person much
36:11stronger, too.
36:12>> Yes.
36:12>> You know what I mean? But it makes the
36:14person who does "donkey work"—the one
36:16who just executes without purpose—
36:17that person will be 100%replaced
36:18because I don't need you.
36:19>> I get it. In summary, AI won't take my
36:23place if I reallocate my execution time
36:26to studying, to delve deeper into the
36:28knowledge itself, so I know more
36:30details and can use AI better for
36:32execution.
36:34>> Exactly, what makes you good at AI, as
36:36you said, is the integration between
36:38models, combining diffusion models and
36:40auto-regressive ones. We know the pros
36:41and cons of each model, and we know
36:43them all, so we can blend them together
36:45.
36:45>> Nice.
36:45>> And the workarounds—remember when I
36:47told you in the videos that I took the
36:48first two shots from a bad generation
36:50and then attached them to the second
36:51generation? So I used two generations
36:53instead of throwing it in the trash.
36:55>> There are many of these workarounds.
36:56For instance, something very important
36:58regarding workarounds: many people say,
37:00"But Wael, AI doesn't make videos in
37:02Arabic, or if it does..."
37:04>> Or when it does make videos in Arabic,
37:06it does it......in a way where the
37:07voice is very fake. Like Google, Google
37:09Omni and things like that, it's the one
37:10doing it. Because Google has a higher
37:12database. So it knows how to speak
37:13Arabic, but it's fake. It's not good
37:14Egyptian, and it speaks in a weird way.
37:16And all that. So, I...Right now, I want
37:17to make a movie like the one you made.
37:19But with Arab actors speaking Arabic.
37:20How do I do that? It's very difficult,
37:22right? That's where the workarounds
37:23come in. A person working strictly
37:24would say, "No, honestly...""
37:25Unfortunately, AI doesn't do that." And
37:27AI truly doesn't do that.
37:28>> But are you relying on the AI and
37:29shutting off your brain? Or are you
37:31thinking of ways to solve the
37:33limitations of AI? I have a model
37:34called Kling 2.5. It produces very good
37:36videos, but it doesn't speak Arabic.
37:38But this model can take reference audio
37:40. It can take audio.
37:42>> Nice.
37:42>> Very nice, so I can give it an audio
37:44file? Yes, and what happens? The audio
37:47file takes the speech being said in
37:49that audio, and the character in the
37:51generation lip-syncs to the words.
37:52>> Okay.
37:53>> But when you add the voice, it will
37:54take the same voice that was in the
37:56audio. Meaning, if I recorded it on
37:57this mic, let's say my character is in
37:59a war, for example. And that character
38:01in the war, there's sound around him—
38:02gunfire and so on. I'm speaking into
38:04the mic; the sound here is very clean.
38:05So you'll find that in the generation,
38:07the sound is very clean, it's not...It
38:08doesn't make sense for the war he's in.
38:09>> That right there is the problem. But I
38:11managed to make it speak Arabic with
38:12the same lip-sync. So what am I going
38:14to do, sir? I have a video where the
38:16guy is speaking Arabic. But the audio
38:18is bad. I'll take this video and put it
38:20into any video editing program. I'll
38:21extract the audio from it. I'll take
38:23that audio file and send it to another
38:25tool called ElevenLabs. ElevenLabs is
38:27an AI model specialized in......
38:29>> in sound.
38:30>> So I upload my audio file to it. And I
38:33tell it, "Replace the voice." And I
38:35choose a voice from the available
38:36voices. Or I create a new voice.
38:38Someone, for example, tired or whatever
38:40. So it says the same thing but in a
38:42different voice. And I tell it, "Make
38:43it sound like war." And all that. So it
38:45gives me......a new audio file. This
38:47audio file says the same thing that was
38:49said. But the sound is fixed. And there
38:51was a war, and it had another man's
38:52voice, and all that stuff. I take it
38:54back to my video editing software. So
38:56my video turned out great, and it's
38:57talking about a man. So that's what you
38:59call a workaround.
39:01>> I’ll give you my failed experience
39:03with this; I was making a video
39:05showcasing an AI model, and the goal
39:07was to make an ad for my shop. I own a
39:09coffee shop, and...I wanted to make an
39:12ad for it. I had problems. The first
39:14problem was that I was using SeaArt 2.5
39:16. The problem is that when it sees my
39:18face, it immediately gives me an Indian
39:20accent; it says you look Indian, so it
39:22turns me Indian. The second thing is
39:24that I tried to explain a lot to it,
39:25not in boring detail, of course. I get
39:28bored. But its problem was, for example
39:30, if I told it that I am the hero of
39:32the video. And then I told it once that
39:34there is another person in the video
39:36doing something else. It realizes that
39:38I am the second person and makes me do
39:39the other thing. So I told it that
39:41I’m entering the shop to order coffee
39:42. And then the barista will make coffee
39:44. So it made it look like I walked in
39:46and ordered coffee. I felt that was the
39:49problem. I kept looking for solutions,
39:52and I realized that the process is
39:54exhausting and expensive. If I have 20
39:56characters in my video, I need to
39:58create the 20 characters and the
40:00objects themselves. It told me this
40:02coffee cup, create it as an object and
40:04place it. So I felt, wait, this is
40:06exhausting. It really is very
40:07exhausting; it’s not like what I
40:09expected—just write a prompt and be
40:10done with it.
40:11>> As I told you, good videos are
40:13exhausting. Bad videos are easy. Just
40:15like he told you. You have to provide a
40:16character sheet for yourself. A
40:18character sheet for the barista and
40:19what the coffee cup looks like. And a
40:21location sheet, what the location looks
40:22like. And you write, and then your
40:24problem becomes a prompt problem. That
40:25you have to explain in the prompt what
40:27the importance of each reference is. In
40:28your prompt, you have to describe that
40:30from second this to that, such and such
40:31will happen.
40:32>> Exactly, I saw someone doing this and I
40:34wasn't convinced that this is the way;
40:36I said I'll spend two or three days
40:37teaching this monster just to get one
40:39proper video.
40:40>> That's if you're writing the prompt by
40:41hand.
40:42>> Great.
40:42>> I don't do that.
40:43>> I have Claude now, or I have ChatGPT. I
40:46trained it. I gave it a skill on how to
40:48write very good SeaArt prompts. I spent
40:50a day doing it, or I got it from the
40:52internet. I downloaded it, I trust
40:53someone who made it. And I read it,
40:55found it good. And put it in. Sir, I
40:57hit the mic button next to the text, I
40:59press it, I speak, I tell it I want to
41:00make a video, one, two, three, with
41:02these characters. Here are their
41:03pictures, this will be reference 1,
41:05this is reference 2, reference 3 is the
41:06barista, and so on. I talk, I don't sit
41:08there writing a lot, I just talk about
41:09what I want to happen. I say the camera
41:11should be a wide-angle shot. Then it
41:12goes in and such and such happens. So I
41:13walk in, walk in to order coffee. So he
41:15does it for me. I take it and absorb it
41:16. And I say how beautiful it is. And
41:18the camera cuts, like a drone, for
41:19example, flying in the sky. I told him
41:21that, he takes my words and from. His
41:22skill starts writing me a detailed
41:24prompt. A very long prompt. Exceeding
41:2610,000 characters, for example. A very
41:27long prompt. I take this prompt and my
41:29references to generate with. The video
41:31comes out just right. It didn't come
41:32out right? I go back and tell him the
41:34problems with it. Just as we said a
41:35while ago, he fixes it for me.
41:37>> Can I get the creative direction itself
41:39from Claude and ChatGPT?
41:40>> You can, but it’s not good.
41:43>> Use AI in creativity and vision, don't
41:46use AI in execution.
41:47>> Don't come and tell me the AI gave me
41:49this business idea, the AI gave me this
41:51video idea, or the AI directed this
41:53video. It will look like it was
41:54directed—it will look very bad.
41:56>> But if you use your brain, if you are
41:58creative, if you understand and have a
42:00vision, and guide the AI in execution,
42:02then you are working correctly.
42:04>> Good.
42:04>> So no one should turn off their brain
42:06with AI.
42:06>> The AI is an employee of mine; I think
42:08of everything, and it executes. I am
42:09the senior, and the AI is my employee.
42:11That’s how it is, nothing else. Very
42:13good.
42:14>> That's it.
42:14>> Another point, as you said, AI is
42:16expensive. You work, the video doesn't
42:18come out right, you iterate once, twice
42:19, three times, the 50-dollar
42:21subscription is gone. It’s very
42:22expensive, and I don't have the luxury
42:24of experimenting and losing money.
42:25>> That’s why you must learn AI. To get
42:27anything from AI, you have to spend on
42:29it. To learn it, you have to spend on
42:31it. But after you have learned, you
42:33will learn how to save credits. You
42:37will first learn what companies to
42:38subscribe to. Should I go to Runway?
42:40Should I go to Magnific? Should I go to
42:42Hexfield? Should you go to OpenArt?
42:43There are so many platforms. Look at
42:45how many credits this one gives you.
42:46And how many videos can those credits
42:48produce from SeaArt? How much is it
42:49here and how much is it there? What are
42:51the advantages here that are not there?
42:52So you start looking at all the
42:53platforms and decide which ones to
42:54subscribe to.
42:55>> Do you recommend specific things to us
42:56or not?
42:57>> I recommend Hexfield. The reason is
42:59what I told you a while ago, try to
43:00remember it regarding image generation.
43:01>> That it has all the models?
43:03>> No, all of them have all the models,
43:04other tools have all the models too.
43:06>> It has exclusive models.
43:07>> It has two models available only there,
43:09which they created themselves. That’s
43:10a plus. Another plus is it's the
43:12cheapest, cheapest platform based on my
43:13research.
43:13>> Even though it's expensive, even though
43:15it's not cheap.
43:15>> It’s the cheapest platform.
43:17>> Wow.
43:17>> Just imagine, because the Max
43:19subscription, which is 200 dollars on
43:22Hexfield, gives you 5,400 credits.
43:24>> These 5,400 credits are equivalent—if
43:27SeaArt 2.5, 30 seconds, 720p quality—
43:31to 210 credits per generation.
43:33>> So on average, calculate it, it will
43:36make you, for example, 27 or 28
43:38generations of 30 seconds.
43:40>> Oh, I see.
43:41>> Great. Other things, no, they max out
43:43at 17, so this is 10 more than them.
43:46>> Okay.
43:46>> So this is the cheapest one, and with
43:48the exclusive models, it’s the best.
43:50>> And I like to subscribe to ChatGPT with
43:52it. Because it includes the image model
43:54. And it has all the things I might use
43:56. So that’s 20 dollars. Subscribe to
43:58ChatGPT on your own. Don't go buying
44:00from people on Facebook. Shared
44:02accounts and things like that. For two
44:04reasons. First, you will get scammed.
44:06I’ve been scammed so many times.
44:07That’s number one. Number two, the
44:09chatbot, the more you talk to it.
44:10>> The more it understands you. When you
44:12have 100 people on the account, the
44:13next day it’ll say "Hi Sawsan" and
44:15stuff, because there’s someone on the
44:16account named Sawsan. I’ll try to
44:17tell it to make me a prompt, and it
44:19says "How was your shopping yesterday?"
44:20And I don't talk to it about those
44:21things.
44:22>> So that’s the problem. So 20 dollars
44:23is more than enough. ChatGPT is...and
44:25it will handle itself. There are many
44:27plans. You subscribe according to your
44:28needs. Because it is expensive. And
44:30guys, videos are expensive. Don't make
44:32videos. Unless it will be cheaper for
44:33you to make them with AI than to film
44:35them. Or you are making money. Or you
44:37want to learn. Otherwise, you’ll get
44:38lost. I mean, don't just spend and get
44:40lost.
44:41>> Spending a lot while you learn.
44:42>> Exactly, that’s something to know
44:44about subscriptions. But besides
44:46subscriptions, Wael, you made that
44:47coffee video and it turned out bad
44:49after 10 generations; you spent 100
44:51dollars.
44:52>> I have custom skills on ChatGPT that
44:54write me very good prompts. I’ve
44:56tried many times, I know where it makes
44:58mistakes and where it doesn't, so I
45:00know how to prompt.
45:01>> Right, I know how to place the
45:03references so it looks better; I have
45:05experience because of trial and error.
45:07>> This comes from experience or someone
45:08teaching you directly, someone who has
45:10already tried it. In that case, I can
45:12output the video in how many
45:13generations? I can do it in two. You
45:15did it in 10; you spent 200 dollars to
45:17output it, and it turned out bad in the
45:19end.
45:19>> If you came to me, I would ask you for
45:21the same 200 dollars to do it for you.
45:23I’d ask for 200 dollars, this is an
45:25example, right? It will come out on the
45:28second generation.
45:28>> So it cost you less.
45:30>> It cost me much less. And it turned out
45:32much better than yours. So you paid the
45:34same 200 dollars. But I made money.
45:35Because I understand the subject. And
45:37you took it. And saved yourself the
45:39trouble. You got it in faster time.
45:40That’s why people pay those who use
45:42AI. You might say, "Let me go do it
45:43myself." Go ahead and do it yourself.
45:45It will cost you.
45:46>> It will cost you more and it will turn
45:47out bad. You’ll pay me the same money
45:49, get a better result, and it will cost
45:50me less, so we both win; it’s a
45:52win-win situation.
45:53>> Um.
45:53>> Either you learn AI properly from the
45:55start, and as I told you, AI is
45:56expensive to learn. Now, I want to
45:58learn AI, I have two paths in front of
46:00me.
46:00>> Um.
46:01>> Two small paths. The first path is
46:03self-learning, where I go alone, open
46:05YouTube, there are many free videos,
46:07and I sit there and learn and
46:08experiment. What will happen? While
46:10I’m experimenting and learning, I
46:12will spend a lot of credits.
46:13>> I will struggle and make mistakes once
46:15and twice, why? Because the video on
46:17YouTube explains everything but doesn't
46:18explain things for my specific case,
46:20and it won't tell me about the
46:21workarounds that come with actual work.
46:22It will explain general things. I will
46:24get stuck on things along the way. So,
46:26I will spend a lot of money while I am
46:27learning.
46:28>> Yes.
46:28>> That's the first path. And it's a very
46:30good path. If you are prepared to spend
46:32money, or if you are a person who knows
46:33how to learn alone and is good at that
46:35area. Or the second path is that you
46:36sit with a mentor or take a course.
46:38This course or mentor is a person
46:39working in the field. They have spent a
46:41lot and done many projects. And they
46:43show you the workarounds, the tips and
46:44tricks, the subscriptions, and what
46:46they did step-by-step. And show you how
46:48every video was made. What will that do
46:50then? In the course, you paid let's say
46:525,000 EGP. While you are learning on
46:54your own, you will spend more than
46:5530,000 EGP.
46:56>> Right.
46:56>> Here, you would have spent 5,000 EGP on
46:58the course and 5,000 EGP on AI
47:00subscriptions.
47:01>> While I am experimenting and applying,
47:03the total is 10,000 EGP and I learned.
47:05There, you spent 30,000 EGP and learned
47:07the same, or maybe even less.
47:08>> So don't calculate it like that,
47:09calculate it because AI is expensive, I
47:11should learn correctly to spend less
47:12and create value.
47:13>> So it's done once and I get to the goal
47:14immediately.
47:15>> Exactly, these are the two paths I can
47:16learn through, and you choose what
47:17makes you comfortable.
47:18>> It's very normal, you get what I mean.
47:20>> Very good.
47:23>> Okay, I want to wrap up and understand
47:25something specific. All the things you
47:27said are technical, scientific, and an
47:29intensive course, that's great. But
47:31what about a guy like me who has no
47:33experience?
47:33>> I want to make money with AI.
47:35>> I want to make money, yes, I want this
47:37stuff.
47:37>> In my work, in my area.
47:39>> I learned it, how do I make money now?
47:41>> Where will the money come from?
47:42>> This is nice, but are you selling us
47:44your course? Should I go and buy a
47:46course and so on?
47:46>> You want to make money? And you went
47:48and learned by yourself on YouTube, not
47:50from me. What will you do? You are now
47:52offering value. I am offering many
47:53things now that I can do better than
47:55others. First thing, make a personal
47:57what? What are these personal projects?
48:00Say, for example, two or three projects
48:02in your field. In images, videos,
48:03websites, and all that stuff.
48:05>> Like a portfolio, you mean?
48:06>> That's what it's called, a portfolio. I
48:08am showing work that I can do something
48:10similar to. I won't go to a client and
48:11tell him I can do this and that, then
48:12he asks me to show him something I've
48:14done, and I say I have nothing.
48:15>> So I give myself a job.
48:16>> Fake, pick a fake brand, someone from
48:18my relatives, do work for them for free
48:21.
48:21>> Nice.
48:21>> Well said, yes it will cost me money
48:23and AI, but I am doing it for myself, I
48:24am investing in the future.
48:25>> Nice.
48:26>> Well said. You made the portfolio, make
48:28it clean and nice. You have the skills
48:30now, you know how to make a website,
48:31right? Make your portfolio as a website
48:33and display your work on it. So the
48:34website itself is professional. And
48:36then I send you: Here is my website
48:37with my work. It's much better than
48:38sending a PDF or Google Drive link. I'm
48:40a professional. This alone increases
48:42your money and boosts your closing rate
48:44.
48:44>> Great.
48:44>> That's number one. Number two, where do
48:46I even find the people to send this
48:48stuff to?
48:48>> You have two approaches. An approach in
48:51reality, in real life. And an approach
48:53on social media. There are other
48:54approaches, but I don't recommend them.
48:55And I'll tell you why. What is this "in
48:57real life" approach? I've seen friends
48:59of mine do these things. He's sitting
49:00in a restaurant, for example, and feels
49:02that the restaurant's social media is
49:03weak.
49:03>> They aren't posting good posts,
49:05carousels, and such. So he eats there,
49:07and asks to meet the manager. To talk
49:08to him and so on. He meets the manager,
49:10tells him I really liked the food. And
49:11your restaurant is very nice. But I see
49:13potential on social media. You're not
49:15doing carousels, not doing videos.
49:16These things will bring you views and
49:18get you more business. So he tells him,
49:19"Yes, I've always wanted to do that.""
49:20But agencies are ripping me off, and I
49:22can't find anyone," and so on. He says
49:23to him, "My friend, I'll make you a
49:25good proposal.""I'll give you my
49:26service in exchange for you giving me
49:27money." But be clever, be a good
49:29salesman. While you're explaining to
49:30someone, give them a value proposition.
49:32>> I'll give you value, so I'll say, "Sir,
49:34what are agencies charging you?" He
49:35won't tell me a price, I'll give him an
49:37estimate. He'll say they take 50,000.
49:39I'll give you the same thing the agency
49:41does for 30,000, not 50. It's 20,000
49:42cheaper, and I'll give it to you better
49:44and faster, and I'll give you a
49:45discount on the first month.
49:47>> I'll give it to you for 20 instead of
49:4830, and if you like the work, continue
49:49with me. And I'll produce a sample
49:51beforehand. This sample photo won't
49:52cost you anything; photos are cheap.
49:54Show a sample so he sees that you can
49:56provide custom quality for him, not
49:57just a generic photo. Then, 90%of the
49:59time, he will close with you. This, of
50:01course, depends on your way of speaking
50:03, your presentation, your look, and
50:05your appearance.
50:06>> Your work, your performance, and all
50:07that; these are things you must invest
50:09in yourself. I can't help you with that
50:10; I taught you, now you do it. This is
50:12the "in real life" approach. The social
50:14media one is the exact same principle.
50:16But on social media, I'll start posting
50:17content, explaining the things I've
50:19done, and showcasing my work. And start
50:20searching. For brands and people who
50:22need something like this. And go send
50:24them messages. I sent 100 every day. If
50:26at the end. Of the month, I sent to 100
50:28, that's 3,000. If three respond, you
50:30have three clients. You'll say, "I'll
50:31go and send to 100 every day?"
50:33>> What do you have to do? This year is
50:34provocative, go send.
50:36>> No, no, I'm with you. I'm very lazy,
50:38I'm the laziest person. I won't send to
50:40100 a day, but I use AI.
50:41>> Great.
50:42>> Come here, Mr. ChatGPT. You have my
50:44portfolio, and it's connected to my
50:46social media. I want you to research
50:48perfume brands, for example. They have
50:50less than 100,000 followers. And their
50:52photos are weak. And put them for me in
50:54an Excel sheet, with their Instagram
50:55and their emails. And I leave it
50:57running for two hours. I come back with
50:58the Excel sheet of the 100, my friend.
51:00It did automation for me. I understand
51:02AI now, and I’ve gone in-depth and
51:03all that. It goes and sends them all a
51:05custom message, based on the service
51:06they want. But these things work
51:08automatically. I get clients. I have a
51:09sales team, without actually having a
51:11sales team.
51:11>> You’re encouraging people to be lazy
51:13like that, I swear.
51:14>> It’s not laziness; making money is
51:15about seizing opportunities.
51:16>> Yeah.
51:17>> An opportunity is in front of you right
51:18now. 99%of people don’t take
51:20advantage of it. You watched this video
51:22and understood it. You can take
51:23advantage of it now while the slope is
51:25going up. Don’t wait until the slope
51:26goes down, it’ll be over. Take
51:27advantage of this window now. I’ve
51:29told you already. Start right now.
51:31Start and do this. This stuff is real.
51:33And I know people who are doing it. I
51:34do it myself. I mean, I’m making
51:36money from this. Otherwise, I would be
51:37sitting here with you right now. You
51:38know what I mean? So I’m telling you
51:40this is real. I taught my friends about
51:41this. Before I taught them, they used
51:43to tell me, "That doesn't exist.""
51:44Nobody makes those numbers.""Nobody
51:46makes that kind of money.""Nobody is
51:47going to pay for those things.""That's
51:49not real, it's impossible." They
51:50started working with me. And none of
51:52them want to work with anyone but me.
51:53They are convinced that this is the
51:55future.
51:55>> Right.
51:56>> I'm telling you, seize the opportunity
51:57and the decision is up to you—will
51:59you execute it? This video, for example
52:00, was seen by a million people; do you
52:02know how many will actually do what I
52:03said?
52:03>> Three.
52:04>> I swear they won't even reach 10, I
52:05swear they won't reach 10.
52:06>> So, a million people saw it, and only
52:0810 might start, and start what? Start
52:10for a week.
52:11>> Honestly Wael, you gave me a
52:12masterclass. I wasn't planning on
52:14making such an intensive video, but
52:15it's your luck that it became a
52:17masterclass. And of course, don't
52:19forget, if anyone wants to join Wael's
52:20course, we will put the link below.
52:22Thank you very much, Wael.
52:23>> My dear friend, God bless you.

About this transcript

This page contains the full transcript of إزاي تسبق ٩٩٪ من الناس في الذكاء الاصطناعي وتعمل فلوس مع وائل by Wael El-Selouky, generated from the public captions YouTube serves with the video. The transcript has 12,509 words across 1,908 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.