How to start making AI Videos in 2026

Learn Film Photography
93 followers
10 Video Views·Jan 17, 2026
AI images and videos are everywhere. Whether it's cinematic AI shots or full-blown AI commercials,
they're all over the internet. And in 2026, knowing how to use AI will be the most important
skill to learn. Because AI is becoming so good that the people that understand how to use it
are making serious money with AI-generated content. And this already is crazy, but here's
the problem. Getting started with generative AI is quite overwhelming, because there are so many
different choices you have to make. You need to choose which tool you're gonna use. You need to
understand how to prompt. And lastly, you need to make sure that you pick the right model when it
comes to generating your images or your videos. Because picking the wrong one, yeah, that's gonna
set you up for failure. Now, I've been in the AI space for over three years. I've generated countless
of different images and videos. I've spent, honestly, like over thousands of dollars in AI credits.
And I've grown this channel to over 170,000 followers just by teaching people how to use AI.
The most received question that I keep getting is, how do I get started? So today, I will walk
you through four simple steps that will teach you how to get started with generative AI.
Now, the first step that you gotta go through is choosing which platform you're gonna use to make
your AI generations. Currently, there are over 15 different AI video generation models and over 10
different image generation models. So that is quite hard and overwhelming if you gotta start
out with just picking one subscription. So you could go for a tool like Google DeepMind, which
has Nano Banana Pro and Google VO 3.1 and all the updated Google models in there. The advantage of
this is it is a bit cheaper if that is what you're looking for. But the downside is that you're
limited to just using Google tools. That's why my preference goes out to choosing an all-in-one
platform. That means that you combine all of the best AI video generator models together with all
of the best AI image generator models underneath one subscription, which allows you to easily
switch between different models. Let's say, for example, right now, Nano Banana Pro is currently
one of the best AI image generator models. But let's say next week a new model launches, then
that doesn't require you to switch your subscription. You can just use that same all-in-one platform and
will also have that newest tool in there. Same goes for video. Some models are better at one
thing while worse at the other thing. Having the ability to switch in between them allows for
massive flexibility and that will in return make you better generations. Okay, so the question is,
which all-in-one platform are you choosing? Now, I've done a comparison on this. I've done
multiple different videos testing all of these tools out. And in my opinion, if you are a beginner,
OpenArt might be the best choice here. And again, all of them work. You could use any
of them for this tutorial. Just make sure you're using one that you like. Step two,
learning how to prompt. Now, prompting is the most important part of every image or video generation
because your output is as good as your input. So to help you with prompting, I have two different
methods that you could use. Starting off with method number one, the prompting formula. For
both image and video, these are quite similar. But let me break them down. Starting off, we have
these subjects. Who or what is in the image? This could be a character, a person, whatever you can
imagine. In this case, we're using a man as example. Then we have the composition. So here,
you want to type in what type of shot. So that could be a close-up. It could also be a wide-angle
shot. Or in this case, we're using a medium shot. Next up, we have the action. What's happening on
the screen? Here, you can add in what you want your character to do. So for example, I want to have my
character hold a sword over his shoulder like that. Then we add in the location. It's basically
some extra context to explain where this is happening. So in this case, we're in the Colosseum
of Rome and we're fighting. Lastly, you could add in the style or the aesthetic of the type of shot
that you want to make. Now, this could be a 3D model. It could be a Van Gogh painting. Or in my case, I
wanted it to be shot on a Kodak disposable camera. So that is the first method of how you can come up
with your prompting. Now, you can switch these things around. It doesn't have to be in this exact
order. But keep in mind, the more context you give it, the better your prompts will be. It is also a
good exercise to give the AI exactly what you want to have seen in the image. For generating AI videos,
you can use this exact formula too. I would just add in the camera shot type. So this could be a
panning shot. It could be camera zooming in, camera moving down. Whatever you want to have
happen with the camera. Now, the second method is using ChatGPT. Within ChatGPT, you can ask it to
come up with more details regarding your image or video prompts. Now, to give you best advice,
I would stay within this context that I already gave you. Or what I would do is, I would go over
and find different prompting guides about, for example, Nano Manana Pro or the exact tool that
you're using. For example, I based my formula on these tips that Google made themselves. So,
what I can do is, if I print this, I can download this as a PDF. Then, I can put this into ChatGPT
by dragging it over. And then, I could say, using this exact PDF, could you help me make an image
about a person walking on the moon? Then, we hit enter. And now, ChatGPT has wrote out the
entire prompt for us. So, we only have to ask it to, can you sum up all of this into one prompt?
And then, we copy this. And then, we go over to OpenArt. Click on images. Here, we go to create
an image. And now, we make sure that we have Nano Manana Pro selected. I will show you which
models are worth it in a second. But here, we paste in our prompt. And then, we can choose our
output as settings. So, we want to do 16 by 9. And then, you can choose your resolution. We can have
4K. And then, we select the number of images we want to create. In this case, I'm going to go with
because the more images I generate, the more images I can choose from. And then, we got an
image like this one. Now, another method of how you can use ChatGPT to prompt is using the explore
page which is the GPTs. Here, if you were to search for Nano Manana Pro, then you can find
custom GPTs that people have trained in order to help you with image generation. For example,
this one already has 50K chats. So, this supposedly should work decently. Always check the ratings on
these. Some of them work better than other ones. You can do the same for this with video. For
example, if I type in VO 3.1, then over here, you will find different prompting bots that will help
you with making good videos. Okay. So, now you understand how to prompt, let's put this to
practice. We're going to start with step number 3 which is generating our images. As of the recording
this video, and this will change constantly, I would recommend the following AI image models.
Now, first of all, I would use Nano Manana Pro. Anytime you need to edit images, this one is
amazing. It is super good. It is super realistic. And honestly, I use this one the most. Next up,
we have Seedream 4.5. Also, one of my favorites. It has super high resolution, good quality,
and also makes some decent generations. Lastly, I would try Flux2 Pro and Flux2 Flex for making
images of people. Honestly, I would just use all of them and cycle between the three of them. That's
mainly how I do it. Some of them might be a bit cheaper to use than the other ones, but in general,
there's not much of a difference besides how many images you can put in as an omni reference. But
more on that later. Let me just now start with using a prompt formula to generate our first
image. So, what we're going to try out right here is we're going to give Chachipiti the following
prompt. A 25 year old female archer holding a mystical bow and arrow at tension. She's standing
in the jungle wearing a mix of futuristic and normal clothes. Now, we can copy this prompt
right here and then we can paste this into Nanomana Pro. And simultaneously, I will also
generate this using Seedream 4.5 and also Flux2 Pro. Make sure to check out if you have the
resolution right. This is always a step that I tend to forget. And other than that, you don't
want to touch a lot more things. I wouldn't recommend playing around too much with the
advanced settings if you're starting out. Okay, so these bottom four are generated with
Nanomanana Pro. Here you can have a look. Then we have these ones generated with Seedream. And
lastly, we have these ones generated with Flux2 Pro. Okay, let's spice it up a little bit. Now,
we're going to get into image to image generation. So, what I'm going to do right here is I'm going
to start up a new chat. Then I'm going to add in an image of me. Here you can see this is the image
that I'm using as a reference. Now, I'm going to give it the following prompt. Create a detailed
prompt for an AI image generator that transforms me into a movie character floating in a bubble
above the city and looking down at everyone. Make it extremely cinematic, realistic, yet
fantastical. Now, we're going to copy this prompt. It is a bit on the longer side for my liking,
but you could always prompt Chachipiti to make it shorter. Then we copied it into here and then
we're going to switch over to using Seedream 4.5. Now, you could also generate this with Nanomanana
while you're waiting. It will give you a bit of a different result. I most of the times test out
both of them to see which one I like the most. So, I generated this a couple of times and this one
I like the most. Now, the other one that I think looks pretty cool is this one, but also this one
is quite sick. Like, honestly, they're all quite good. And this is Seedream 4.5. Now that you got
your image, you can start making your videos. So, part number four is generating your video.
When it comes to generating video, you want to go over to video, then you want to click on text to
video. The best thing about OpenArt is that you have all of the different models built into it.
So, you can choose between text to video, which allows you to use any of these different models
like Sora 2, Seedance, Kling, VEO, VDU, ONE, all of them. Then you have image to video. Basically,
you have the same models. Then you also have elements, which allows you to use Kling 01,
VEO 3 or VDU as well. And lastly, you have video, which is video to video. And this is also quite
interesting. But for now, when you're using text to video, you're basically using that when you
don't really care about consistency. Because each time you're using text to video, you won't be
getting the exact same shot as the next one. And this makes it pretty hard if you're having
a advertisement where your product needs to be consistent or a film where your character needs
to be consistent. Using text to video could be quite hard. But if you're making something simple,
like for example, I made this where I have a close-up of a reverse POV shot of a frightened
soldier. I just used Kling 2.6. This is all complete text to video. The model that is pretty
good for this would be Kling 2.6, VEO 3.1 or Sora 2. Those are the ones I recommend. I would also
play a bit around with LTX, maybe 1 and maybe HALO 2.3. Most of the times though, you want to be
using image to video. Now with image to video, you have a lot more control over the output. Because
here you already have done the first step, which is establishing what your video is going to look
like. And you do that by importing your video. So what you can do here, you can choose from your
history and you can choose the image that you want to use. So let me just double check. I believe it
was this image we wanted to use. So now we have this image as our start frame and you can now
choose what model you want to be using. So for this example, I'll be using Kling 2.6. Now what
you can do is you can type in your prompt or you can leave it blank, but that doesn't give you any
instructions or any control over what you want to have as an output. So basically now you're using
a similar type of motion. So you're saying what type of shot you want to have, what the subject
needs to do, and also what type of actions you want to see within the scene. So you can type in
a prompt like this. Or the second method that you could be using for this is downloading your image,
moving it over to Chachapti. Then you put it inside of Chachapti alongside with a question.
So basically you're saying create a video prompt with this text and image input. Then you add in
the context that you have given it and your image. And now it will draft up a video prompt for you.
Ideally though, if you're getting started, start with writing. Like the more you do with writing,
the better you understand how it responds to your writing. So my main thing is what I'm always doing
is if I'm using Chachapti, is I always go through it and see if I agree. Because the issue is
generating videos is expensive. Like Kling 2.6 is relatively cheap. But if you do this on Sora 2
or if you do this on VO 3.1, it's a bit more expensive. So before you generate, you want to
make sure that you agree with everything that's written here. So right off the bat, this we can skip.
We want to do this. Camera tracking shot of a man floating inside a transparent glowing spherical
spherical spherical. I cannot pronounce that word. Bubble high above the futuristic city. The camera
slowly orbits and tracks alongside the bubble as it drifts forward through the skyline. The bubble
gently glides between towering skyscrapers, subtle fracturing neon city lights and the sunset sky
across its surface. The man remains calm and centered inside the sphere, looking downward at
the city below. Building paths, different depths, create strong parallax, blah blah blah. And then we
ended off with cinematic skyfire realism, smooth motion, realistic reflections and reflections on
the bubble. I quite like this prompt. In case there is something you don't agree with, just change this
prompt. Like you are in full control and also fully responsible for what you're putting out. So now I'm
going to generate this using Kling 2.6. I will also generate this using VO 3.1 and for VO 3.1
you want to make sure you use the highest settings. So I'm using VO 3.1, then I'm using the 1080p
resolution and for this one I'm using fast because it's a bit cheaper than the normal mode. Now for
the duration you can choose how long you want to have your scene and then you hit create. I would
also love to use this on Sora 2, but the issue with this is once you have a realistic character,
a realistic person in your image, then Sora 2 doesn't allow you to make videos of realistic
people. So this doesn't work in Sora 2, that's unfortunate. So while we're waiting on this,
by the time you might be watching this, there might be a new model that has come out recently
that might have been better than the current models we got this far. The basics of what I'm
explaining in this video still remain the same. You want to use the latest models and you want
to make sure you give it enough context. In the future, like what's happening right now, what I
see happening is that these contexts need simple language and they don't require that much of a
complicated prompt anymore. They're better at simplifying your language because they come to
the realization that most humans are stupid. So essentially, don't worry too much about your
prompting. Try and see what works best for you. So let's have a look at Kling 2.6. Decent. The
only thing I dislike is that it changed my face throughout this video. Let's see what Google VO
has to say about it. So yeah, that might happen. You might have some generations where it didn't
work. So you want to generate this a few times. For example, I tried this with another example
and here I got this. If you want to have even more control over the output of your videos,
then you want to be using start and end frame. Not every video model has this. For example,
as you can see right here, Sora 2 doesn't have it. But if we switch over to VO 3.1,
then we can see start frame and end frame. So that means that here we can upload what our end
frame is supposed to be like. Now for that, we first need to create an end frame. So let me just
go back over to image. And now we're going to create an image based on a previous image. So
I'm going to use this image right here. I'm going to download this. I'm going to remove this. Then
I'm going to upload it right here. And now I'm going to give it a simple prompt. So I'm just
going to say he's staring at the camera while still entirely within the bubble. But now he's
in outer space. And then of course, I messed up my settings here. So we're going to generate this
in widescreen right there. Okay, so now we have this second image, which is going to be our end
frame. So we download this, then we head back over to video. And over here in image to video,
we want to be selecting either VO 3. Or you can also do something like Cling 2.5 with start and
end frame. But let's see what Google VO 3 can make. I will do both. So we have the start frame
already, which is this image. Now for the end frame, we're going to drop in our second image.
And then we're going to give it a prompt. Again, I'm going to keep it simple right here,
because I noticed that that might work better with generating videos. So I'm going to mention
the following. A portal opens up behind the man in the bubble and sucks him up into it,
transporting him into outer space. So let's see what we get with VO. And let's also do this with
Cling 2.5. And why not, let's also do this with one 2.5 now that we're here. Okay, so here we
have the results starting off with Google VO 3.1. And again, I haven't cherry picked any of these,
these are just what I just generated. I mean, it's not bad, but it could definitely be a lot
better. I'm pretty sure that if I were to like regenerate this, I use these settings again,
and I could get a better result. Now let's see what Cling 2.6 has to offer.
Pretty cool. It doesn't quite open up a portal though. Then lastly, we have one 2.5.
Okay, that transition is sick, but it just doesn't end at our end frame. So I'm not sure what went
wrong there. So what we can do is we can use this start frame now with this end frame. And then we
say the camera transitions through a rock hole and pans towards a man floating in the bubble. So now
I'm generating it using VO. But again, you can do the same with one 2.5 or Cling 2.5, whatever
you're using for start and end frame. Here's the result of our last transition. So that's pretty
cool. We can now stitch the two together. And using this method, you've just created your first
transition using start and end frame. So now you know exactly how you can get started with start
and end frame. You have learned how to prompt, how to make images, and how to make videos. All of this
has been done through OpenArt. If you want to try this out yourself, I will leave a link in the
description down below. But if you're not yet sure which out of all of the all-in-one platforms is
the best one, then click the video that's on the screen right now. And there I will explain out of
all the different platforms that I've tried, which one I prefer the most.
