Luyện nói tiếng Anh bằng Shadowing qua video: How to Create Viral AI Storytelling Shorts

People are pulling billions of views with these short storytelling videos, but the second you try to make one yourself, it just gets scrolled past.

And most people think it's all about the algorithm, when in reality they just don't have the right workflow to make these fast and good enough to go viral.

So in this video, I'll walk you through the only workflow you'll ever need step by step, so by the end you can make scroll -stopping storytelling videos yourself

and get ahead of most people who are still trying to

win over the algorithm there's a loaf of bread in a

museum in naples someone put it in the oven 1900 years ago

and never took it out it's 79 ad just after midday the mountain opens for 18 hours ash

and stone rain down on the city roofs collapse thousands run

north then at one in the morning the column above the

volcano falls in on itself what comes down the mountain isn't lava it's gas and ash at 300 degrees, moving a hundred miles an hour.

It crosses the city in minutes.

There's no time to suffocate.

Death is instant.

The heat is so sudden that muscles seize on the spot.

Those twisted poses aren't agony.

They're the body locking in the exact instant it happened.

And then the ash keeps falling, 20 feet of it, and it sets hard as stone.

Everything is sealed exactly as it was.

Bread in the ovens, eggs on the table, the people decayed inside it and left hollow spaces behind, perfect molds of the moment they died.

18 centuries later archaeologists filled them with plaster.

That's what you're looking at.

And a big thanks to OpenArt for sponsoring this video and helping me achieve this result.

And the reason most of these get scrolled past has nothing to do with the visuals, because it gets decided before you ever open a single tool.

Your script decides everything that comes after it, what you generate, how long each shot holds, where people lean in, and where they drop off.

And before I write mine, there are a few crucial rules you have to follow to actually make a story people care about.

Start by opening on an object instead of a stat, because a real thing like a loaf of bread left in an oven makes them picture a person and a story,

when some random number never does that.

Then write it plain and factual and let the pauses do the work, so you break your lines into different parts and give the big moments a bit of silence instead of over -hyping them.

And the biggest one of all, you go after something the viewer thinks they already know, because putting a new meaning on a familiar thing beats brand new information every single time.

Now, I'll actually write mine and the story I'm doing is about Pompeii.

The whole thing is built on breaking the belief that those famous figures in the museum in Pompeii are bodies,

because they're just plaster somebody poured into the hollow spaces the bodies left behind in the ash.

And I'll save that reveal for the very end.

Then I'll open on the object, that loaf of bread sitting in a museum in Naples

that someone put in the oven 1900 years ago and never took back out.

From there, I'll just take it through the eruption and the ash burying the city, right up to the point where the ash sets hard as stone.

And that's the exact line where the whole thing turns from everything getting destroyed to everything getting preserved.

The whole script comes out to about 190 words, and that's on purpose.

Because at the pace this format runs, that lands right around 60 seconds.

So by the end of this part, the whole story is written and locked before I've even opened a generator.

But the script on its own doesn't do anything yet, because the next part is what turns these words into something that actually looks like a real film people want to watch.

So now I'll take that finished script and start building every line into its own frame, and the whole trick is getting 12 separate images to look like they belong in the same film.

From here I'm doing the images and later the videos inside one platform called OpenArt, and I'm using it for one specific reason.

It's got every model I need all in one place, so I'm never jumping between three separate tools or paying for three different subscriptions just to make one video.

The link is in the description, so you can follow along with the same setup.

Once I'm in open art, I'll open the image tool, set the model to GPT image 2, set the quality to high, and set the resolution to 4K.

Then I'll switch the aspect ratio to 1x1 so every frame is square, and I'll leave auto polish off so it can't rewrite my prompt.

And I'm making one image for every shot in the script.

And I'll do all of them before I animate a single thing, because an image that comes out wrong barely costs me anything to redo, too, while a bad video costs a lot more.

Now, there are two things I'll keep the same across all 12, and these two are the whole reason it works.

The first is the style.

Every prompt opens with the same style line, which is cell style 2D animation over a painted background, so 12 separate generations actually look like they're from one film.

The second is the negatives.

Every prompt ends with the same core list blocking anything photoreal, 3D, or anime, because without it, the model drifts toward whatever look it defaults to.

And I'll never reword it, because that repetition is what makes 12 separate generations actually match.

I'll start with the first frame.

My hook, that loaf of bread sitting in a dark museum case under one warm amber spotlight.

And I want a non -realistic painted look here so it reads like a relic instead of just a loaf of bread.

Hit generate and see how they turn out.

And that's exactly the look I wanted.

One warm pool of light on the loaf, the reflection cutting across the glass, and deep shadow everywhere else.

Now I'll paste the other 11 prompts and see how they turn out.

And to me, each of these actually looks very realistic, which is exactly what I wanted.

But out of all of them, the second and the fifth ones definitely stand out, because the alley in the city handles the shadowing really well while capturing the volcano in the background.

The fifth one captures the darkness and the sheer size of the actual volcano, with the massive, thick smoke flowing over it.

One quick note before we continue you.

Every prompt I'll use in this video is linked in the description too, so you can copy and use them whenever you want.

The color is actually the biggest thing tying these images together.

The second frame through the seventh is the whole eruption back in 79 AD, so those start on a warm ochre grade and get darker and colder as the ash takes over.

And the one thing I forced into every single one of them is a negative for lava.

No glowing orange and no flames.

Then everything from the 8th to the 12th is the present day museum, so those all get the cold gray color grade.

The only exception is that opening bread shot, which sits on its own under that one warm spotlight so it feels like an object under study.

By the end of this part, I've got all 12 images ready, and they actually look like the ones pulled from one film, but they're still just images,

and now the real test for the AI comes when I actually turn them into video.

Now I'll take all 12 of those images and animate them one by one and see how the model handles that.

I'll go to the video section on the left, and there's a handful of settings I'll keep the same on each one of those 12 generations.

I'll set the model to Seedence 2 .5 and have it animate straight from my image instead of generating anything new.

Every video stays at 1080p resolution and one by one aspect ratio, so it matches the frames.

Each one runs around 5 to 8 seconds, and I'll set it to one video per prompt, so I'm judging a single take each time.

Then I'll completely turn the audio off, and I'll do it on purpose, because C -Dance generates its own audio for each video,

so 12 videos would leave me with 12 completely unrelated background tracks,

when what I actually want is one continuous voice running under the whole thing that I'll add later.

This next part is where the results fall apart for many people.

The second you let the model animate the actual painting.

The lines wobble and it stops looking painted, which is the one thing you want to avoid from your very first video.

So to keep that from ever happening, you keep the subject almost completely locked in place and give all the movement to the camera instead.

That way the only thing really moving in the frame is stuff like drifting ash or dust in a beam of light.

I'll also open every prompt by telling the model plainly to hold the exact painted style,

because these image -to -video models drift straight toward photorealism the second you let them.

And I'll close every prompt with negatives against any style change, morphing, flickering, or camera cuts.

On top of that, every shot gets its own camera move.

A slow push on one, an arc on another, a tilt up, and a slow pan across the last one.

The exact move for each shot lives in that shot's own video prompt, so the whole camera plan is in the prompt list too.

If every shot moved the same way, the whole thing would feel lifeless.

Now let's give the model the prompts, hit Generate, and see if it animated the images the right way.

Thank you.

And with all 12 animated, I now have about 77 seconds of raw footage to work with.

100

But every one of these is still dead silent, so nobody would actually watch this the way it is right now.

101

So now I'll give it a voice so it becomes much more engaging to watch.

102

I'll move to the audio tool, open create voiceover, and set the model to 11 multilingual v2.

103

For the voice itself, I go with one called Bella.

104

It's professional, bright, and warm.

105

Then I paste my whole script straight in.

106

Normally for a story like this, you'd reach for the big dramatic movie trailer voice, but that's exactly what kills the mood.

107

The single biggest control here is the speed.

108

I'll keep the voice fairly bright, but I'll drop the speed to around 0 .92 seconds, and slowing it down like that reads as considered instead of rushed,

109

and finally gives all those written pauses enough space to play out.

110

I'll push the stability up so it stays consistent the whole way through.

111

Then I'll keep style exaggeration low so it never starts acting.

112

There's a loaf of bread in a museum in Naples.

113

Someone put it in the oven 1900 years ago and never took it out.

114

It's 79 AD.

115

Just after midday, the mountain opens.

116

For 18 hours, ash and stone rain down on the city.

117

Roofs collapse.

118

Thousands run north.

119

Then at one in the morning, the column above the volcano falls in on itself.

120

What comes down the mountain isn't lava, it's gas and ash at 300 degrees." And the voice comes out clean and continuous the whole way through,

121

which makes it much easier to sit and watch.

122

So now I can be sure the narration for the whole video sounds clear and smooth.

123

But even a clean voice on its own still won't match up with a bunch of separate shots.

124

And getting those two to line up is the last thing left to do.

125

Now, I'll put everything together in CapCut into one continuous video that can actually go viral.

126

The move that matters most is laying the voice down on the timeline first and cutting the picture to it, never the other way around.

127

Then I'll drop the 12 videos against it in order.

128

Then there's the square window, because my shots are all square, but the final video itself is vertical.

129

I'll change the ratio to 9 by 16.

130

So now any video I put next shows through one clean square window centered with room for captions around it.

131

And when a video doesn't quite fit, I'll trim a bit off the start or the end, so I'll never have to speed anything up to force it.

132

After that, I'll auto -caption the whole thing, pick a clean font with a soft glow and a basic entrance animation,

133

and I'll keep the captions outside that square window so they never sit on top of the footage.

134

For sound, I'll lay one quiet music track under the entire thing.

135

So now let's watch the finished 60 -second short all the way through.

136

There's a loaf of bread in a museum in Naples.

137

Someone put it in the oven 1 ,900 years ago and never took it out.

138

It's 79 AD.

139

Just after midday, the mountain opens.

140

For 18 hours, ash and stone rain down on the city.

141

Roofs collapse.

142

Thousands run north.

143

Then at 1 in the morning, the column above the volcano falls in on itself.

144

What comes down the mountain isn't lava.

145

It's gas and ash at 300 degrees, moving 100 miles an hour.

146

It crosses the city in minutes.

147

There's no time to suffocate.

148

Death is instant.

149

The heat is so sudden that muscles seize on the spot.

150

Those twisted poses aren't agony.

151

They're the body locking in the exact instant it happened.

152

And then the ash keeps falling, 20 feet of it.

153

And it sets hard as stone.

154

Everything is sealed exactly as it was bread in the ovens, eggs on the table.

155

The people decayed inside it and left hollow spaces behind, perfect molds of the moment they died.

156

18 centuries later, archaeologists filled them with plaster.

157

That's what you're looking at.

158

And the final result came out really good.

159

Every shot feels like part of the same painted film and the music keeps the tension right to the end.

160

I went with the square format because I personally prefer it, but you can run the exact same workflow with 9 to 16 videos too

161

but even with the right workflow you still won't get your storytelling shorts to go viral

162

if you have to jump between a bunch of different platforms

163

just to get one done luckily inside open art you've got every single tool you need in one place

164

so you're not wasting time and credits anywhere else

165

so sign up to open art with my link in the description

166

and start making your own viral storytelling shorts today thanks for watching and I'll see you in the next one.

✨ Video gợi ý

257 câu

Từ vựng và ghi chú luyện nói cho bài học này

Video này có 166 câu và 2534 từ để luyện shadowing. Phần lời nói dài 14:41. Người nói nói với tốc độ tự nhiên, khoảng 173 từ mỗi phút, gần với hội thoại hằng ngày. 83% số từ nằm trong 3.000 từ tiếng Anh thông dụng nhất; phần còn lại nên xem trước khi luyện.

Từ vựng quan trọng trong video

15 từ đáng học trong video, kèm phiên âm và nghĩa:

Từ	Phiên âm	Nghĩa
oven danh từ	/ˈʌ.vn̩/	lò, lò nướng
generate động từ	/ˈd͡ʒɛn.ə.ɹeɪt/	sinh ra, gây ra
volcano danh từ	/vɑlˈkeɪ.noʊ/	núi lửa, hoả diệm sơn
instant danh từ	/ˈɪnstənt/	chốc, chốc lát
lava danh từ	/ˈlɑvə/	dung nham
audio tính từ	/ˈɔ.di.oʊ/	âm thanh, âm tần
ratio danh từ	/ˈɹeɪ.ʃoʊ/	tỉ lệ
column danh từ	/ˈkɒləm/	cột
continuous tính từ	/kənˈtɪn.juː.əs/	liên tục, không ngừng
collapse động từ	/kəˈlæps/	đổ, sập
hollow tính từ	/ˈhɑloʊ/	rỗng
scroll danh từ	/skɹoʊl/	cuộn, cuốn
format danh từ	/ˈfɔːɹ.mæt/	khổ (sách, giấy, bìa)
egg danh từ	/ɛɡ/	trứng
muscle danh từ	/ˈmʌs.əl/	cơ, bắp thịt

Phát âm cần chú ý

Người nói dùng 71 dạng rút gọn, ví dụ I'll, I'm, they're. Hãy nói theo dạng ngắn đúng như bạn nghe.

Âm “th”: algorithm /ˈælɡəɹɪðm̩/, thick /θɪk/, smooth /smuːð/, seventh /ˈsɛvənθ/
Âm “sh” và “zh”: ratio /ˈɹeɪ.ʃoʊ/, caption /ˈkæp.ʃən/, shadow /ˈʃædəʊ̯/, capture /ˈkæp.(t)ʃɚ/, animation /ˌæn.ɪˈmeɪ.ʃən/
Từ dài — đặt trọng âm cho đúng: continuous /kənˈtɪn.juː.əs/, animation /ˌæn.ɪˈmeɪ.ʃən/, automobile /ˈɔː.tə.məˌbil/, stability /stəˈbɪlɪti/, generator /ˈdʒɛnəɹeɪtə(ɹ)/

Cách luyện với video này

Nghe hết video một lần, chưa cần nói, và ghi lại những từ bạn chưa biết.
Bắt đầu ở tốc độ 0,75×, nói đuổi từng câu, rồi quay lại tốc độ bình thường khi đã quen.
Ghi âm giọng mình rồi so với bản gốc, chú ý các từ như oven, generate, volcano.

Ngữ pháp trong video

Những cấu trúc người nói dùng nhiều nhất, kèm đúng cụm từ trong video:

Cấu trúc	Trong video
Câu bị động be + quá khứ phân từ — nhấn vào việc xảy ra, không phải người làm	is sealed · is built · is written
Mệnh đề quan hệ who / which + mệnh đề — thêm thông tin về người hoặc vật	people who are · line, which is · realistic, which is

Phương Pháp Shadowing Là Gì?

Shadowing là kỹ thuật học ngôn ngữ có cơ sở khoa học, ban đầu được phát triển cho chương trình đào tạo phiên dịch viên chuyên nghiệp và được phổ biến rộng rãi bởi nhà đa ngôn ngữ học Dr. Alexander Arguelles. Nguyên lý cốt lõi đơn giản nhưng cực kỳ hiệu quả: bạn nghe tiếng Anh của người bản xứ và lặp lại to ngay lập tức — như một "cái bóng" (shadow) đuổi theo người nói với độ trễ chỉ 1–2 giây. Khác với luyện ngữ pháp hay học từ vựng bị động, Shadowing buộc não bộ và cơ miệng phải đồng thời xử lý và tái tạo ngôn ngữ thực tế. Các nghiên cứu khoa học xác nhận phương pháp này cải thiện đáng kể phát âm, ngữ điệu, nhịp điệu, nối âm, kỹ năng nghe và độ lưu loát khi nói — đặc biệt hiệu quả cho người luyện IELTS Speaking và muốn giao tiếp tiếng Anh tự nhiên như người bản ngữ.

Phương pháp shadowing: đọc hướng dẫn từng bước đầy đủ →

Luyện nói tiếng Anh bằng Shadowing qua video: How to Create Viral AI Storytelling Shorts

Is AI About to Change Movies Forever?

The story behind this viral video

Unfortunately, YouTube Really Is This Simple

Learn UX Storyboarding in 5 Minutes: w/ Steps and Examples | UXtweak

Laziest Ways to Make Money with AI (For Beginners)

The Neuroscience of Addictive Storytelling (And How to Master It)

work-life diaries | my last week at work as a data scientist

Animation Basics in 14 Minutes (6 Big ideas for beginners)

AI Animation Pipeline: How I Make Cartoons 100% With AI

This Is Boring But It Will Make You Scary Intelligent

Fink's Bedtime Story (The Wild Robot, 2024) 4K

The ONLY 5 Lines You Need To Tell Any Story