5 Levels of Prompting to Create ANY AI Video

There's five levels of prompting you

need to master AI video and 99% of

people are stuck at level one or two.

And here's the thing, each new level

unlocks incredible possibilities for the

type of content you can create and gives

you way more control over the AI videos

than what you probably think is possible

right now. So today, I'm breaking down

every level of prompting you need to go

from the complete beginner to a seasoned

pro. This might be the most valuable AI

guide I ever make. The first level of

prompting is simply describing the idea

you have. This is what most people do

when they first start making AI video.

So, for example, I've got some pictures

of myself and let's say I wanted to

generate a video with a huge fluffy dog

behind me coming outside the door. In

the prompt, I would just intuitively

describe what I want to happen. The

closet door behind the man opens and a

giant furry fluffy light blue dog crawls

out. At this level, you're just

describing the raw intent behind what

you want to happen. You don't have any

structure to your prompt. You don't have

any special techniques. You're just

telling the AI model in the most simple

way exactly what you want to happen.

Just because these prompts are super

simple and basic doesn't mean that the

quality of the video is going to be any

lower. In fact, the actual quality of

the video itself will roughly be the

same regardless of how complicated your

prompt is. Here are some examples from a

new sea dance model where these videos

were all generated using simple idea

prompts that are only one or two

sentences long. A massive cracking

attacks a pirate ship. The captain

slices it with his sword. Hyperrealistic

cinematic movie scene. And that's really

all you need to generate a super

highquality looking video. Here's a

video created with an even simpler one-s

sentence prompt. A nature documentary

about an otter flying an airplane.

This is the incredible story of the

pilot otter. So, if we can make a super

highquality video using just a super

simple prompt, what's the point of this

tutorial? Well, it's not just about

creating the high quality visuals inside

the video. It's about controlling what's

happening exactly the way you want. Even

though at level one of prompting you can

make some really good-looking stuff, the

main issue you're going to run into is

inconsistency between different video

generations where you feel like you need

to generate over and over and over again

until you finally get a video clip

that's worth using. Level two is

structured prompting. Instead of just

intuitively writing down the raw idea

for your video, you can instead use a

structured prompt formula and fit your

idea into that structure. Now, there's a

ton of different variations and guides

like this on the internet. These are

just structured formulas for writing

prompts that have been proven to work.

The key components of these prompts stay

the same. First is the subject, the

environment, and the action that they're

taking. These are all different

variations of subjects, environments,

and actions. The next key component of

these cinematic prompts is the camera

shot and camera movement. Camera shot

defines how the subject is framed, like

a close-up for emotion or a wide shot

for environment. Camera motion defines

how the camera moves, like a tracking

shot or a slow pushin. Together, they

turn a basic idea into something that

feels cinematic.

The third component of the cinematic

prompt structure, the visual style.

Visual style defines the overall look of

the scene, such as realistic anime or 3D

Pixar, shaping how the final video

feels. So, here's an example prompt with

this cinematic style structure. A 1980s

cinema grainy film, which is the visual

style, a medium shot, which is the

camera framing of a tired office worker

in Japan standing on an empty subway

platform, loosening his tie as a train

approaches in the distance, flickering

tunnel lights, and an analog

advertisement board glows fatally green.

Another common structured prompt that

you're going to run into is the JSON

prompt. So, a JSON file is just a simple

text format that's easy for code bases

to read and write. So, here's an example

of what a JSON prompt would look like. I

have my keywords, which are the subject,

the action, environment, camera, etc.,

and also a value assigned to each of the

keywords. And so in this case, the

subject is a lost hiker. The action is

struggling to walk through deep snow and

the environment is a blizzard in a

frozen mountain range. But the most

useful thing about these JSON prompts is

that it's super easy to organize and

swap out different keywords, especially

if you're working with multiple people

with all the keywords neatly laid out.

the results that you get when using a

JSON prompt are the same as using a

regular prompt as long as the

information that you include in both of

them are the same. So, here's a few

different comparisons of videos I have

where I generated them using a JSON

prompt and one where I didn't use a JSON

prompt format, but I included all the

same keywords and information inside the

prompt and the results are pretty much

the same. And I found this to be true

across tons of different tests. So, the

moral of the story is the JSON prompt a

great way to organize and structure your

prompts. It makes it super easy to work

with huge databases of prompts, but it's

not a magic pill that's going to

suddenly create amazing AI videos.

Another super important structure prompt

you need to know is the multi-shot

prompt, which is a single prompt that

defines multiple sequential shots, each

with its own camera angle, action, and

timing to create a full continuous

scene. Here's an example of a really

impressive cinematic sequence generated

using a multi-shot prompt. Inside the

prompt for this multi-shot sequence in

the different cuts are clearly defined

with a different camera shot and

movement and also exactly what's

happening inside of each cut.

When you write prompts that have these

structured templates, you'll find

yourself having way more control over

exactly what's inside the videos. The

third level of prompting is reference

control. So far in the first two levels,

we've been describing with our words

what we want to happen inside the AI

video. And using that, you can get

really far. But at level three, instead

of just describing with your words,

you're directly showing the AI video

with reference material. exactly how you

want your videos to be created. So,

previously we've seen a basic version of

this where I used an image of myself and

then prompted for a huge fluffy blue dog

behind me. In this case, the reference

is controlling what my appearance is

going to be. But we can use way more

than just a single image to guide the AI

video. In this example, two reference

images of a character and scene were

used. The first one with the main

character holding a dagger. In the

second image, it shows her and all the

defeated asalants on the ground. And

these two images were combined as part

of the prompt. The AI video was told to

generate a sequence of the woman

fighting the assassins with different

shot composition and perspectives thrown

in there. And the result is an extremely

dynamic fight scene all while preserving

the consistency of the appearance of the

woman. Beyond just image references, we

can now also use video references, audio

references, and of course, text

descriptions to guide the video

generation process. Here's an example

for an exciting AI generated fight scene

between two characters. And to create

this video, a bunch of different