All language subtitles for 7. AI Influencer in different environment, clothes, lighting and perspectives

In this video, we will use Prompt Engineering to make our AI influencer in different environments and

it will be cool.

So just follow me on this one.

The first thing is we search for a picture that we like.

I created some pictures here.

These are all random pictures with nearly the same prompt.

I think this right here is relatively good.

So we simply press this and then we get our prompt.

We click generate.

But before we click generate this right here is the seat I want to work with this seat because this

seat obviously works.

So I copy this seat.

Then I press generate.

We will use the prompt that we have.

We always have our negative prompt.

We have our model, we have our Laura's and so on.

But this right here is important.

We need to use the seat.

If we take minus one we have a random seat.

But I want to have this seat.

And now we start to tweak our prompts.

So instead of a nightclub we will do this.

Maybe underwater digital eyes are okay, but we don't use neon lights for this one.

Also, the cinematic lightning is not needed if we take underwater stuff, we don't need this word at

all.

She won't wear a custom and knit dress because it's obviously underwater.

The custom are maybe a bikini and the rest I think is completely fine.

So we simply send this prompt and then we will get our AI influencer with nearly the same face.

Everything should be relatively similar, but it is underwater and she is in a bikini.

Let's just look what we get.

So she is in a bikini.

It's not exactly underwater, but I think the pictures are nice.

These are pictures that will really, really work.

Now we do the same stuff.

We press this right here, we copy our seat.

Don't forget to copy the seat because this is an important step.

Then we press generate.

Again.

We use our fixed seat and then we take her in a different environment.

And don't worry if there are some typos right here.

Stable diffusion doesn't see the exact words.

Stable diffusion always see word tokens.

So here is some messy stuff from the beginning prompt.

But that is totally fine.

So one beautiful girl.

Maybe right now she is in a car.

The custom right now is not a bikini, but maybe a short dress.

And basically that's it.

Especially the first picture.

This looks like really, really nice.

And then we do the same stuff.

We copy the seat, we change the environment.

I hope you get it.

I think she has a bit of a long neck sometimes, so we will.

So we try to make this better.

The negative prompt we start here with long neck.

I think this will get us better outputs.

Now she is on the beach.

The custom is a bikini again.

That is totally fine for me and we send it in a night club professional neon light and our seat is also

set and we press generate.

Okay, right now she is wearing a bikini in the nightclub.

Maybe that is not optimal.

So we can of course also change this.

Maybe a bikini is not the optimal outfit for our nightclub.

I must admit I didn't show it.

So generate again our seat and now she is not wearing a bikini.

She is wearing underwear.

And then we do basically the same stuff.

But again of course with the same seat.

And then she is wearing a dress.

Wearing a nice dress for a nightclub.

Yeah, that's that's basically better.

That stuff with the bikini is also okay.

The underwear stuff is I think it's not that nice.

But this right here, this is also perfect, I think.

And sometimes stable diffusion meshes a little bit up.

You can see right here the hand is maybe not that perfect.

This right here.

Yeah, that's that's like really nice.

But maybe something on the ear.

I think I will run this prompt again, but maybe with a different seat and just check what we get.

If I like your prompt a lot, we can also run this prompt sometimes with different seats.

Sometimes we will get really, really good output.

Sometimes the outputs are not that good and then we take the pictures that are the best.

Of course, you can also see this works like relatively fast in a few seconds.

We always have our generations and if the generations are not perfect like here in this picture, if

there are some messy green stuff on the ear, we can of course make this better because we have also

inpainting in stable diffusion.

So we can simply change this a little bit.

I think this pictures, this are like really, really nice.

They are nearly perfect.

We are in a good nightclub, perfect lighting, sometimes small stuff is a little messed up, but this

messed up stuff is really easy to fix.

So I think right now we have really good pictures.

She is really, really consistent with the face and this little mess ups we will delete in the next

lecture.

So you can see maybe this here, this hair right here is not perfect.

Full body view, close up of the face looking directly in the camera.

And because the prompt was relatively good, I also took this prompt and I simply changed this right

here.

Full body view I added this simple words so that we have also something like a full body view.

Okay, we don't have the feet in this picture, but that is totally fine.

Full body view with this seat works really fine.

Then full body view with a random seat.

You can see we have really, really nice pictures right here.

I think all of them look really, really cool.

And then I try to create a close up of the face.

And also that looks relatively good.

I think I have to tweak the prompt one last bit, and then we have also a close up of the face.

So close up of the face looking directly in the camera.

Detailed eyes.

We don't use neon light right here.

Maybe she is not in the night club, body facing forward.

We can and need to delete this bright lightning.

Maybe also studio light, professional lightning.

No neon right here facing straight ahead.

Yes, I think this is perfect.

So we will also get something directly in the camera because a close up we will need later for our face

swaps.

And it's really important to have a good close up where she looks directly in the camera and where are

not a lot of distractions.

Nice.

I think this is perfect.

And this prompt I will generate again one last time and then we have our pictures.

I think especially this right here is perfect.

So I simply took some more prompts and tweaked them a little bit.

I really liked this one.

And for that instance, I let it run a few times and then simply edit in full body view, because it's

always nice to have a lot of different pictures because we can search for the best ones.

I think all of them are really, really nice and also the close ups, they get really, really good.

These pictures are all nice to post, but some of them, they are not perfectly suited for our Instagram

because if you look a little bit closer, the faces are not perfect sometimes.

Sometimes they are a little bit different.

Yes, they look nearly exactly like the same person, but not 100%.

And later we will fix this via the face swap.

But first let's see how we can fix a picture if it is not perfect with inpainting in stable diffusion,

because this is really easy.

So in this video we created our influencer in different environments.

We simply used prompt engineering and tweaked our prompts.

We try to work with the same seat and simply tweak the prompts a bit.

You can also tweak the negative prompts you have seen.

I included long neck and I think the neck got a little bit shorter.

It looked a bit nicer, but just remember, change the environment, change the lighting, and also

change what this girl or this man is wearing.

And you will get really, really cool stuff.

Duff.

And in the next video, we will take a look what we need to do if we have little masses in our pictures.