Why My AI Videos look Ultra Realistic - Higgsfield AI
By Dan Kieft
Summary
Topics Covered
- Prioritize Images Over Videos
- Master Cinematic Lighting First
- Train Characters for Consistency
- Reverse Engineer Prompts
- Emotion Drives Cinematic Scenes
Full Transcript
Most AI videos look impressive, but they still feel like they're AI generated. To
me, cinematic means reaching a point where people stop asking, is this real or is it generated? And if you want to reach that point, then you got to learn
how to make realistic AI videos. And
you've come to the right place because in this video, I will break down my entire workflow step by step.
Part one, to make cinematic videos, you need cinematic images. The biggest
mistake I see people making is putting all their effort in generating the videos. Most of your effort should be in
videos. Most of your effort should be in generating your image because that will be your reference, basically the ingredient for your cinematic video.
When I make these kind of videos, I spend 99% of my time perfecting my images. It is all about concepting,
images. It is all about concepting, storyboarding, getting the right shot and feeding that image into a video generator. When it comes to generating
generator. When it comes to generating cinematic images, there are three main tools that I'm using. We have Nano Banana Pro that is the original model that I have been using for a while. And
now we have Nano Banana 2 that just came out. Besides that, I'm also using a tool
out. Besides that, I'm also using a tool called Hicksfield Saw. I kind of reverse engineered that tool and I will show you in this video what I mean by that. Now,
if you want to try this out yourself, I will leave a link to Hicksfield in the description down below, because there you can access all of the tools that I'm about to mention, including Cinema Studio, which is insanely useful for
making cinematic shots. Before we go even more in depth, let me first explain what makes an image look cinematic.
First, we have the lighting. Take a look at this image. The lighting is flat and it doesn't look that appealing. If we
now prompt it and apply some cinematic lighting that has motivation, direction, it creates shadow, contrast, highlight, and even some depth, then we get a shot
like this. Lighting is a insanely
like this. Lighting is a insanely important aspect. On film sets, you have
important aspect. On film sets, you have entire lighting crews trying to light that perfect shot. Next up, we have composition. This might get complex
composition. This might get complex really quick, but let me just show you this image. You see, your eyes have no
this image. You see, your eyes have no idea where to look. You see the people, you see this guy on the bench, and then you have this guy at the forefront. But
still though, like your eyes are drawn somewhat to the center. If you have no idea where to start with your composition, let's apply the rule of thirds. Basically, what it means is you
thirds. Basically, what it means is you turn your frame into a 3x3 grid. You
want to make sure that you place important subjects on these lines or if you have key focus points, you want to keep those at that intersection. Now,
the question is, how can you actually apply this in your images? You simply
prompt for having it slightly off centered. Or if you want to be more
centered. Or if you want to be more specific, you can prompt it the subject is positioned left third of the frame.
You can also use leading lines as a composition guide. For example, right
composition guide. For example, right here we can see that it draws you into the bridge. Next up, we have depth. Like
the bridge. Next up, we have depth. Like
I said, with lighting, you don't want to have a flat looking image. Like an image like this is not that appealing. But if
we want to create depth, then we want to add things to the foreground uh to the background and also to the midground.
Lastly, we have color. And this is a powerful one because your generation might be really good, but you might not like the colors. You can simply prompt it to be changed with those colors. And
now we have a cinematic looking shot like this one. All of that might overwhelm you, especially if you're just getting started. Luckily, there are
getting started. Luckily, there are tools for this. So, in this video, we will be using Hicksfield Soul and Nano Nano 2 for this to make our images look way more cinematic. Keep in mind, you
want to master images first before you move on to videos. So, how do I generate my cinematic looking images? Basically,
what Soul 2 is, it's a way more cinematic looking custom model of Nano Banana. I don't know what they've done
Banana. I don't know what they've done to it. I tried reverse engineering it,
to it. I tried reverse engineering it, and I will show what I learned in this video, but it's actually pretty cool.
Now with Soul 2, if you selected this here, then you can have a few custom things. So first up, we can use
things. So first up, we can use different styles. So this is pretty
different styles. So this is pretty cool. We can use these different styles
cool. We can use these different styles like a 2000 brands, subto flash, street photography, Asian nostalgia, all kind of cool things which will get you totally different results. If you
combine that with a bit of Nano Banana and a bit more prompt engineering, then you get cool shots like these ones. The
important thing to know here is it's not just the aesthetic, it's also the perfect character consistency and that's what we try to achieve with Soul 2. Now
to get that what I did is over here at character you can make your very own character. So if we take a look at mine
character. So if we take a look at mine then we can see that I have trained this on like 70 different images of me.
Literally, I asked my friends to just take a bunch of photos of me and have different expressions, different angles, side angles, all of that. So, I can throw that inside of Hixit Soul so it
understands what I look like, how I look. But if you don't have that or you
look. But if you don't have that or you simply don't want to generate a character of yourself, then you can also do that of any other type of character.
Let me introduce you to Veronica. Let's
make a character of her. Right now, we only have one image of her. And in order to train or soul ID, it's recommended that we have over 20 different images.
Now, to get to that point without having a whole photo shoot in place, what you can do is you can use my document right here, which will give you 20 different images that you can put into so ID. So,
we have a front view, we have close-ups, we have full body shots, we have full body to the left, full body to the right. Essentially, we have all the
right. Essentially, we have all the shots you need in order to train this sole idea to get the most consistent looking results. All you need to do is
looking results. All you need to do is you need to copy and paste each prompt individually, put it inside of Nano Banana 2, and then prompt it and upload all of those images inside of Soul ID.
If done correctly, you will get something like this. And keep in mind, you can do this with yourself, too. Now
that you have this character, you can now start using Soul to make cinematic looking images. Inside of Soul, you can
looking images. Inside of Soul, you can now select your character. So I'm
selecting Coronica. Then I can change the style to any of these 22 different styles that we have. Or I can even create my own style. So if I have
certain references, I can build my mood board in here and I have my own unique style. So for example, if I now choose
style. So for example, if I now choose street photography, then I can just generate this and I get this result. Now
with this template, I managed to create all of these different looking styles just from my character. And the texture on it is amazing. And here's the result with prompting. And this prompt will
with prompting. And this prompt will look complicated to you, but I will explain this later because this is actually a reverse engineered process of Soul 2. To give you an idea what's
Soul 2. To give you an idea what's possible, here are some more different results. Here we have Y2K Street. Then
results. Here we have Y2K Street. Then
we have Siren. And lastly, Mystic City.
All of these look insane. So now you understand all the tools that I'm using, but let me now explain to you what my exact workflow is. First, and maybe by far the most important step before you
start generating anything is starting your own mood board. This is an example of my mood board that I made for this intro. So, I have collected all of these
intro. So, I have collected all of these from a number of different locations.
Some are from Pinterest, some are from shop deck, some are from social media like YouTube and Instagram. Even this
combined led me to my vision. This is
taking inspiration from others. This is
not stealing. The first step is using Soulu for image referencing. Let me
explain what I mean by that. So, I got this reference image that I want to use.
I'm dropping it into soul.
Unfortunately, you no longer have a prom box, Hicksfield. I don't know why. Like,
box, Hicksfield. I don't know why. Like,
please give us a prom box. But either
way, we're using my character now. And
all I have to do is I just have to hit generate. Now, it's using my character
generate. Now, it's using my character and replacing the reference, but putting it in the reference image. As you can see, even though I'm far in the distance, it somewhat looks like me.
This already looks pretty good. But now,
let's bring this a step further. So now
what I do is I download this shot and then I'm using this as a reference in Nano Banana 2. Then I'm adding in a image of my site face so it understands
what I look like from the side because we have this bicycle and the AI doesn't understand what I look like as a side profile. And then lastly, I just give it
profile. And then lastly, I just give it a prompt. I just say change the man
a prompt. I just say change the man outfit to a tan canvas jacket layered over a dark sweater and a light blue denim shirt while preserving his exact identity. facial features, hairstyle,
identity. facial features, hairstyle, pose, bicep position, camera angle, composition. Literally, I'm telling it
composition. Literally, I'm telling it to keep everything else the same. And
then we hit generate. Now we have the exact outfit that I wanted. And now we have this completed image. Another
method I use is reverse engineering Hicksfield Soul because I like the look of Hickfield Soul so much that I want to copy what they're doing. And I don't know if this is supposed to be happening. And I don't know if this will
happening. And I don't know if this will get me in trouble, but if you're generating anything using Hixfield Soul, it leaves behind the prompt that it is using. So we can use this in our own
using. So we can use this in our own images. And let me show you how I do
images. And let me show you how I do that. So let's try and recreate this
that. So let's try and recreate this image using this method. I found this shot on Pinterest and I find it an interesting shot. So I took a screenshot
interesting shot. So I took a screenshot of it. This will be our reference. I'm
of it. This will be our reference. I'm
putting it into Soul 2. Then I'm going to add in my character, which will be Veronica. And then all I'm doing is just
Veronica. And then all I'm doing is just generating this. Hicksfield just gave me
generating this. Hicksfield just gave me the exact prompt that they used to get this result. So we're going to copy
this result. So we're going to copy this, steal it. I hope I don't get in trouble. But now we can use this. And
trouble. But now we can use this. And
because this result is not quite what I had in mind, let's go back to the drawing board and start over from generating our image. So we delete the reference image. We paste in the prompt
reference image. We paste in the prompt that we just copied. Then we still use Veronica. And now let's change this up
Veronica. And now let's change this up and try out some different things. So
we're going to use a different style.
For this one, I'm going with digital camera. Now we hit generate. And here's
camera. Now we hit generate. And here's
our result. Now for the final touchup.
And I know it might be getting complicated here. We're going to use
complicated here. We're going to use this again as a reference. And then
we're going to remove everything. We're
just using this as a reference. Then
we're going to use Nano Banana 2 for this. And now we're going to paste in
this. And now we're going to paste in our prompt. And that is half the woman
our prompt. And that is half the woman wear a ruined once beautiful expensive white wedding dress and she's crying with tears and mascaras running down her face. Basically we add in the little
face. Basically we add in the little details like that is how you keep reiterating your images. You keep adding stuff or removing stuff making it better with each generation. So now we hit
generating and this is the result. So
this is how we made this cinematic looking shot of this girl using this method. Here's another method of how I
method. Here's another method of how I use this in my own shots. So for
example, for this shot, I used this reference that I found on YouTube. Now
the way I did this is not your traditional way. What I did is I took
traditional way. What I did is I took that screenshot, then I uploaded it into Hixit Soul, then I selected me, then I just hit generate. It gave us this result. As you can see, it doesn't look
result. As you can see, it doesn't look at all like me. But we do have this prompt which we can now like copy and then we can use that to regenerate this again with me as a reference and no
image. We copy the prompt. We put it
image. We copy the prompt. We put it back into soul. And now we just start changing up the prompt to the details that we want to have in there. So for
example, I don't want to have a red mug with the Strand Book East 1927 New York City, but I want to have a mug with best creator ever. Essentially, what I did is
creator ever. Essentially, what I did is I changed up the prompt to match my style. So that is like describing that
style. So that is like describing that I'm a man, that I want to have a blue mug in there, and then I keep everything else the same. So then we hit generate and that gets me this result. If we
compare this to the intro, then you can see that there's one more trick that I apply to make this look even more dreamy and cinematic. I will tell you what it
and cinematic. I will tell you what it is in a minute. Let's now move over to making cinematic videos. Now, most of the factors we already covered in the previous section, but for scene
generation, the secret ingredient to making it look cinematic and making it look real is emotion. Adding emotions in our scene is what makes it relatable to the viewer. Before generating a scene,
the viewer. Before generating a scene, always think to yourself, what emotion should my viewer feel? And if you know how to successfully achieve that, then you've done a good job. Now, there are
several things that you can play around with to achieve that. First, we have the color. For example, the color red can be
color. For example, the color red can be done for scenes that are emotionally charged, like angry or really sad or intense situations. Then, we have blue
intense situations. Then, we have blue for calmness and for isolating scenes.
Then we have camera directions you can play with. For example, zoom in can add
play with. For example, zoom in can add in intensity while zoom outs can give you emotional distance. The last thing we have is speed of some scenes especially in AI and good looking AI is
all about using the right sense of speed. So there are some editing that
speed. So there are some editing that comes into play at manipulating that speed or specific tools like a tool like Cinema Studio. Now Cinema Studio is in
Cinema Studio. Now Cinema Studio is in my opinion something like Clink 3.0 No, but with way easier controls. As you can see, we have this director panel and it can do all kind of crazy things with
helping you prompt and get the exact results that you want to have from your scenes. You can add in your references
scenes. You can add in your references or even your characters in here, any movement that you want to have in terms of your camera. You can select the speed ramp, and this is pretty cool. So, you
can customize it. Do you want to have the camera speed ramp fast or slow? You
can also choose if you want to have multi-shot like for example like this.
We can describe different scenes and we can add all kind of different scenes in there and then we can customize it for each scene. So you're literally becoming
each scene. So you're literally becoming a director where you're analyzing and choosing and thinking about each scene before you shoot it. Let's try and recreate my intro shot. And let me
explain what my workflow was. Now how I made this shot is as follows. The first
thing I did is I uploaded my start frame. and that is this shot that we
frame. and that is this shot that we made previously. So this is the start
made previously. So this is the start frame. Then what I did is I selected my
frame. Then what I did is I selected my character and that is me. So I'm just going to use me. I'm going to position me at the end of my prompt. So don't
mind this. Then I'm going to enter in my prompt which will be a slow steady zoom in. And essentially the structure that
in. And essentially the structure that I'm using here is camera plus subject plus action environment lighting. And
you can also add in some texture and what you want to have happen on the audio. Now, for this, I'm going to keep
audio. Now, for this, I'm going to keep the speed ramp at auto because I want to have it figure out on its own what to do with this. For the duration, I'm going
with this. For the duration, I'm going to do 3 seconds. And now, we can move over to the second shot. So, here what I'm going to do is same thing. I'm just
going to prompt it. This time, we have a closeup of the mug. So, this close-up will be just of the mug. That's why I'm also keeping the speed ramp at medium.
So, we're not doing anything with that speed ramp or anything with the camera.
Then, we add in one more shot. And we
click on that. And now, we can add in our last detail. For the last one, we have my prompt again where we have me basically taking a sip of the mug and then we are going to hit generate. The
only downside about this is Higfield Cinema Studio 2 takes a bit long in my opinion. After waiting for a few
opinion. After waiting for a few minutes, this is the result that I have.
As you can see, this dirt shot is not looking pretty. And that is why what you
looking pretty. And that is why what you want to do is just generate it again. So
now I generated this one more time and I got this shot. Now, this looks way more like me. So, I can use this shot now and
like me. So, I can use this shot now and I just have to make a cut in my timeline. Now, I did the same thing for
timeline. Now, I did the same thing for a bunch of other shots. And after doing all of that, then you get all of your different results. The final part is
different results. The final part is putting it all together in your editing program. The first step I do is I import
program. The first step I do is I import all of my clips and I arrange them in the exact position that I want to have them to have. Then, next up, I'm going to make some quick cuts. So, anything
that's way too long, any mistakes in the generation that I don't want to have in my scene, I just cut them off. And then
I have my trimmed down version. Then I'm
adding in some overlays. And this is super underrated. Now, what this will do
super underrated. Now, what this will do is if you have this shot like this, if I add in this overlay like that, then it looks way more cinematic. To achieve
this look, you want to add in a luma key or a gauzing blur effect. I use this on Premiere Pro. I'm not sure if other
Premiere Pro. I'm not sure if other programs will have it. If you do some research, I'm sure you will find it. As
you see, it requires quite a bit of work to make these cinematic looking videos.
Now, if you're doing all of this yourself, it's a lot of fun to get into.
Like, this hobby to me is where I spend most my time in it. If you want to try this out yourself, I would highly recommend using a tool like Hicksfield.
So, you can play around with these cinematic tools and actually make cinematic looking videos. Click the
video on the screen right now if you want to learn more about how to use AI for cinematic filmmaking.
Loading video analysis...