The ONLY 7 Prompts You Need to Create Any AI Video

I used to spend hours trying to write the perfect prompts for my AI videos only to get results that look like this.

But then I realized something. The best

AI videos aren't made using complicated prompts. They're generated using simple,

prompts. They're generated using simple, clear, intentional prompts that the AI models actually respond to. So today,

I'm breaking down the only seven prompt styles you'll ever need. The ones that work for any AI video.

>> Take cover. The first style prompts are cinematic prompts. These focus on the

cinematic prompts. These focus on the camera work in the scene. It's not just about what you're seeing, but how you're observing it. For example, I can have

observing it. For example, I can have the static shot of a soldier on a battlefield. The camera stays perfectly

battlefield. The camera stays perfectly still, highlighting a moment of peace and contemplation. And here's a second

and contemplation. And here's a second shot of that same exact soldier, except this time we're zooming in and rotating around to focus in on his face. It's a

much more personal shot.

Other camera movements include panning to follow our subject as it rolls into the battlefield. We can use a shaky

the battlefield. We can use a shaky handout camera that highlights the chaotic environment of the soldiers in the trench getting hit with artillery shells. Or take that same exact scene

shells. Or take that same exact scene except this time the camera flies above to a bird's eye view. It's a less personal and more cold and detached view of what's going on. Use prompts like the

camera rotates or the camera orbits to rotate around a subject in a circular fashion.

There's tons of different camera movements from camera zooming to camera pullback, from tilting up to tilting down, panning left to right, rotations, and you can also combine different

camera motions together as well. By the

way, the AI videos I'm showing you in this guide were made using the imagetovideo technique. First, we give

imagetovideo technique. First, we give the AI video generator a reference image. And then we can enter a prompt

image. And then we can enter a prompt describing how we want the scene to play out. Here's a image frame of a soldier

out. Here's a image frame of a soldier in a trench I want to animate. And then

adding a prompt that describes what's happening inside the scene. So for this particular shot, I didn't add in any camera motions. Instead, I just want to

camera motions. Instead, I just want to focus on the subject himself of him pulling a small wooden cross from underneath his jacket. So with cinematic prompts, we can control how the camera

is moving throughout the scene. But what

if we want to divide our AI video into multiple different segments, each with their own different camera movements and action happening? That's where the

action happening? That's where the second prompting method comes in, which is timestamp prompted. Here's an

8-second video of an astronaut. In the

first 3 seconds, the camera zooms in towards him, and then it tilts down to show him looking at this recording device. And finally, the camera tilts

device. And finally, the camera tilts back up again to see him looking up towards the sky. we can actually tell the AI video generator exactly what motions to happen along each segment of

the AI video. This will give you a lot more control over the sequence of events that are happening and exactly how you want them to happen. Here's another

example of timestamp prompting where first I told the camera to zooming to a over- the-sh shoulder shot of the astronaut and then end the scene by zooming in closely on the creature

space. Time stamp prompting is the best

space. Time stamp prompting is the best way to organize multiple actions happening inside a scene. So, we can divide our AI video into different segments. But to take things a step

segments. But to take things a step further, we can actually generate multiple camera angles using just a single AI video. This video shows a full body of the astronaut turning and

walking towards a spaceship. But then it cuts to a close-up shot of his footsteps along the ground. To create this effect, I use a third prompt style, which is the cutscene prompting. For this particular

cutscene prompting. For this particular video, I used the prompt, "The astronaut turns and walks towards a spaceship."

Cut to the boots of the astronaut walking along the terrain inside a cinematic film, which let me divide this video into two separate segments. We can

combine this technique with timestamp prompt to get a huge amount of control over the direction of the videos. Here's

a shot divided into three segments of an astronaut on an alien world. We cut to a close-up shot of his expression of wonder and awe on his face. And it ends with a final shot of him exploring the

alien planet. To create this effect,

alien planet. To create this effect, divide the prompt into different timestamps. And in each timestamp,

timestamps. And in each timestamp, define a different camera angle or a different emotion that you want to happen. This turns AI video generator

happen. This turns AI video generator into a storyboard tool or video editor.

Time stamp prompting gives you control over the sequence of events and cutscene prompting lets you add in different scenes to tell a full story, turning you into an AI director. A word of caution

with cutscene prompting. Sometimes if

you prompt for different cuts that are completely different from the original video or requires AI video generator to imagine a lot of new things. So here I wanted a different cut to inside the

mouth of the plant. it does sometimes have a hard time maintaining consistency in the video style. So, it went from a more realistic style to kind of like a 3D animated scene. So, if you're going

to use cutscene prompting, you want to make sure that the scene that you're cutting to isn't too overly different from what you had in your original shot.

By the way, the prompt techniques I'm showing work for Google Veo 3, but they can also be applied to different AI video models like Sora 2 or Seance or

Cling. Prompt style number four are GPT

Cling. Prompt style number four are GPT prompts. Since we're already creating AI

prompts. Since we're already creating AI videos, why not use an AI language model to help you write the prompts themselves? I learned this from the AI

themselves? I learned this from the AI video school channel. So, thanks for the tip. Different AI video models usually

tip. Different AI video models usually come with their own written guides for how to prompt inside them. Here's one

for the Google Veil 3 model. And the

documentation includes information like the best formula for writing effective prompts and also some essential prompting techniques. So let's see what

prompting techniques. So let's see what happens if we try teaching an AI using this information to write prompts for us. So what I did is I took that prompt

us. So what I did is I took that prompt manual and copied over all the relevant important information like the formula for writing prompts, uh prompting

techniques, a few examples as well. And

inside chat GPT in the explore tab, uh, I'm going to go ahead and create my own GPT. And inside here, what I did is I

GPT. And inside here, what I did is I gave it a little bit of a job description. I want this GPT to be a

description. I want this GPT to be a prompt helper that arrives prompts to generate AI videos for Google BO 3.1.

Exactly. Um, so I said the user can enter a description of the scene and you should generate a prompt for that scene.

And what you'll need to do is actually upload a PDF file that includes a documentation of how to write prompts into the knowledge section. And once

that's done, we have our own GPT prompt helper. So here's an example where I

helper. So here's an example where I uploaded a picture of this woman uh with this stag. And I told her to write a

this stag. And I told her to write a prompt for a sad scene where the female elf is saying goodbye to her old stag friend. And this is what the GPT

friend. And this is what the GPT produces for us. So it'll give you a much longer detailed prompt that includes all the different elements of

the cinematography, but also to the sound effects, the style, the lighting, and all the action that's happening in the scene. And the video that it

the scene. And the video that it generates more or less captures what I requested, which was a sad goodbye scene. It can help reduce the amount of

scene. It can help reduce the amount of time you spend typing out the prompts and instead have this AI helper. Now,

the problem with this is that the AI is sometimes confused. Here's this image,

sometimes confused. Here's this image, and I told it to write a prompt for a reticent scene as he gets ready to leave and go on an adventure. Looking at the prompt GBT gives us, it starts with a

wide morning shot, which doesn't really make any sense. I think it meant to say a wide angle shot in the morning. Now,

problems like this are fairly minor.

It's easy to go in and correct them manually. But the biggest problem with

manually. But the biggest problem with using this GPT prompt method is in the AI documentation rarely rarely tells you

what the AI video generator is bad at.

And the difference between being good at prompting and being bad at prompting is actually understanding the limitations of the AI video, what it can't do. So,

if you train a GPT using information that's incomplete, that doesn't include what the AI video generator can't do well, you're going to have problems that

you're not really sure how to solve.

Here's one example where I chatb to write a prompt for seeing where the elven woman is in trouble. The town

spoke have discovered who she is and they're angry. So, let's just pretend

they're angry. So, let's just pretend she's like, you know, a secret spy or something. And if we go into the prop,

something. And if we go into the prop, some of the elements that the GPT asks for include the crowd of angry towns folks surround her on both sides shouting and pointing. So something AI

videos can't do well is animating crowds of people. If you have a lot of people

of people. If you have a lot of people in the scene and you want them to perform some actions, it's going to look really weird. Either they're all going

really weird. Either they're all going to do the same thing at the same time, which happens in Beo 3, or they're just going to turn to like blobs and blur together. And of course, the GBT prompt

together. And of course, the GBT prompt helper has no way of knowing this since this is not mentioned in the official prompt guide from the Google Veil 3 model. And so in the videos, you're

model. And so in the videos, you're going to get some weird behavior that doesn't look realistic. Instead, to get the best results from a scene like this where there's so many different people,

a better approach that can convey the same emotion of uneasiness and fear is to have her looking around nervously while these people stare at her in

silence in kind of like a weird eerie way. This will give the same idea across

way. This will give the same idea across that she's in hostile territory, but a prompt like this is much easier for the AI video to animate. So, I think a GPT prompt helper like this is very useful

for more advanced users that actually understand the nuances. But if you're more of a beginner, I'd really recommend sticking to simpler, shorter, more intentional props. So, that way, it's

intentional props. So, that way, it's much easier to go in and figure things out when they don't work exactly how you expect it. AI does weird stuff. Things

expect it. AI does weird stuff. Things

that seem super obvious to you are not obvious to the AI at all. For this shot, I just simply prompted for my character to be smiling and show a happy expression. And the AI completely

expression. And the AI completely transforms the way he looks, not knowing that the ash and red embers covering his face are actually an important part of

the video. To handle situations like

the video. To handle situations like this, you need prompt style number five, which is the anchor prompt. We need to use anchors in the prompts to remind the

AI of what's important to us. So, if I add in the prompt that there should be red embers and ash on him, this time it's actually able to create the same

happy facial expression, but it keeps the same visual style instead of transforming him. So, for this prompt, I

transforming him. So, for this prompt, I anchored in his physical appearance by saying that he has red embers and ash on him. Here's another example of a scene

him. Here's another example of a scene that starts with my orc character riding on top of the wolf. But as we progress the scene and the fight continues, you'll see that he slides off the wolf

in kind of a weird way. And so to fix that, instead I modified the prompt to adding the anchor, reminding the AI in the prompt that the orc should be on the back riding the direwolf. And this helps

the AI maintain that relationship between the two subjects. Sometimes you

have to use anchors to help the AI video generators understand parts of the scene that it can't see yet to understand how to generate parts of the scene later on

that it can't see right now. So I have this character of an orc and you'll see on one of his shoulders, his left shoulders, there's this armor pad and on

his right shoulder there is no armor.

Instead there's this like blue tribal tattoo. Now, when I try to generate a

tattoo. Now, when I try to generate a video of this specific character from a styled profile shot that only shows the left side of his body that has the armor

covering his shoulder, if we watch it, you'll notice that on his right shoulder, the armor pad also appears.

And that's because the AI has no way of knowing based on the initial image frame of what his other shoulder should actually look like. And so what I need to do is adding an anchor to say the orc

has no armor on the right shoulder. Dark

blue geometric tribal tattoos on the orc's shoulder. Using this anchor, it'll

orc's shoulder. Using this anchor, it'll give the AI a better understanding of how I want my complete character to look. So it's not going to be exactly

look. So it's not going to be exactly consistent to the original shot of the orc because the AI is starting from nothing. But by anchoring what his full

nothing. But by anchoring what his full appearance looks like, we can create a video where his shoulder looks a lot more like the original character. To

maximize AI videos capabilities, you got to use image prompting. To get the most beautiful shots, like these serialistic scenes, it's really only possible to do

this consistently if you're using image prompts. To get this dreamy seralistic

prompts. To get this dreamy seralistic scene of the koi fish flying above a canal, first I generated an image using the propped serialism photo, a giant

koiish in the canals of Venice City. And

then I used a clean 2.5 video generator to animate it into a video. Be direct

and clear in exactly how you want the scene to unfold. The fish floats forwards through the air. It's simple,

it's clear, and it's easy for the AI to understand what I want it to do. A city

floating in the clouds. The lights blink on and off. Serialism. The giant giraffe waves in the ocean and eats leaves from the tree. She blinks. The paper covering

the tree. She blinks. The paper covering falls off her face to reveal mechanical gears inside. Image prompting is also

gears inside. Image prompting is also the best way to get multiple shots of a consistent character inside the scene.

It's the best way to use AI to generate multiple camera angles. First, I can start with a front view of my character sitting inside a plant store. And I can use an AI image model like Nano Banana

to create a side profile of her studying at a desk inside of her plant store. Or

I can create a photo of her watering the plants inside the store. If she gets tired, she can take a lunch break and look outside the window to see the people walking by. With these character reference photos, we can then use image

prompting to generate multiple shots of her throughout her day. Whether it's

working, watering her plants, studying, or taking a lunch break, there's always incredible new AI image tools being developed. Here's one I found called uh

developed. Here's one I found called uh Q1 image edit camera angle control where we can actually upload an image reference of a character and use these

camera controls to generate different camera angles of the same exact scene.

So, let's first rotate it by 45°. And by

the way, this is done completely in real time. As I'm talking, this is happening.

time. As I'm talking, this is happening.

And uh there you see it rotated, I think, to her left. Um let's go in the opposite direction. There it is. There's

opposite direction. There it is. There's

other options like zooming in closer to her. You can also tilt the camera up and

her. You can also tilt the camera up and down. So bird is above and worm is

down. So bird is above and worm is below. So let's try tilting the camera

below. So let's try tilting the camera down below her. And there you have it.

If we zoom in on here a little bit, you'll see that the textures on her face are kind of smooth. So, you'll need to use some kind of an image upscaler to get higher resolution or like sharper, more realistic details, but still, this

is a pretty incredible AI image tool for controlling the camera angles of your scene. With image prompting, not only

scene. With image prompting, not only can you just use them as references, you can actually join them together to control the beginning and the end of a full shot. This is called start and end

full shot. This is called start and end frame prompting where we can actually use an image for the first frame of the video but also use an image to control what's going on in the last frame of the

video. And then when you combine these

video. And then when you combine these things together, the AI video can actually render a complete scene of your character between the two frames. The

seventh prompting style you need to know is called negative prompting. Sometimes

to get what you want from the AI, it's actually easier to just tell it what you don't want to happen. So, I have this vlog scene of my astronaut on a lunar

colony. Let's say I don't like the way

colony. Let's say I don't like the way the moon looks outside the window. Maybe

the perspective just looks a little bit off and I don't want any windows on the wall. The easiest way to handle this is

wall. The easiest way to handle this is actually just to directly tell the prompt no windows. Instead of trying to describe exactly what the wall should look like and by using this negative

prompt and telling the AI what I don't want to see inside the scene, it can generate that same shot except this time there is no windows on the wall. Here's

a trickier example where I created this scene of a soldier and you can hear a bunch of different gunshots. It's easy

to understand why the AI would use those sound effects. He is aiming his weapon,

sound effects. He is aiming his weapon, but the result looks a little weird.

There's even like this burst of smoke coming out from his gas mask. And what I wanted to create was actually a more quiet scene of him cautiously observing his environment. And so I went pretty

his environment. And so I went pretty heavy with a negative prompt in this case where I told it to be completely silent and quiet and no gunshots, no clicking, no trigger. It sort of worked.

This time you can just hear the flames in the background without him firing his weapon. So, you know the different

weapon. So, you know the different prompt styles you'll need for AI video.

But there's still so many different cool AI video tools out there. From lip sync tools to motion capture. I made a guide that covers the five levels of AI video generation that will take you from a

complete beginner to an advanced pro. Go

take a look.