All language subtitles for 6. Prompt Engineering for Stable Diffusion

Let's talk about prompt engineering.

As you can see right here, I already messed up it with the prompts.

So we have our girl right now in different environments.

So so she is in the city.

She is in the garden.

She's also on a mountain right here.

So you can see the mountains.

And she is in something like a nightclub.

But before we mess with this prompts, I want to show you how to write this prompts we need to understand

prompt engineering.

Midjourney does a really, really good job to explain prompt engineering.

In its simple terms.

Midjourney is another diffusion model.

So with Midjourney you can also create pictures.

But we will use stable diffusion because it's free.

Think about what details matter.

This is the stuff that you need to include in your prompts.

So what's the subject?

A person, an animal, a character, a location are object or something else?

In our case, it's an Asian girl.

What's the medium?

It's a photo painting, illustration or it's something else.

Most of the time it's a photo for us.

What's the environment?

It's it is indoors or outdoors, on the moon, in an arena, under water.

Or is it maybe in the mountains, in the club or whatever you like?

What's the lighting?

Soft light, ambient neon, studio lights and so on.

What's the color?

Vibrant, bright, colorful black and white, pastel or whatever you like.

And what's the mood that our person is in?

Is she calm?

Is she is she sad?

Is she angry?

Whatever you like.

And of course, last but not least, the composition.

Is it a headshot?

Is it a close up?

Is it a bird's eye view or something else?

And then we have also something that I like to call magic words.

So for example, high resolution 8-K and so on.

Let's take a closer look at our prompt.

So what's our prompt.

If we simply press this right here we can look at everything here.

We start with something like a magic word.

So masterpiece.

But then we use our character.

So it's a beautiful girl.

And where she is in this example in a nightclub, then we have other words that describe exactly what

we want to see.

We want to have detailed eyes, puffy eyes, neon lights.

So you see the lightning top quality, high resolution, realistic, cinematic.

She is Japanese.

It's an Asian beauty, super beautiful, beautiful skin.

She is slender, body facing forward Hyperrealistic High resolution 8-K and so on in Stablediffusion.

You nearly can't overdo it with this prompt engineering.

Most of the pictures that I see that are really, really good.

They have a lot of words in it.

So we simply describe exactly what we want to see.

If you want to write your prompt completely on your own, that's totally fine.

If you copy the prompt just like we did right here, that's also fine.

You just need to know at what you need to think as soon as you start to tweaking this prompts.

So for example, we can use this prompt and then we simply tweak just the environment.

Instead of a nightclub we use a mountain or we use underwater or we use whatever.

We can simply change these things that really, really matter.

So we can use our prompt and we can change the subject and the medium you want change, but we will

change the environment.

We can change the lighting, the color, the mood and the composition.

This stuff right here, this is what matters for us.

The rest of the prompt is completely okay, how it is.

And then we change the environment, the lighting, the color, the mood and the composition.

And so we can create different stuff but really, really specific.

So our person ideally should be the same.

The medium also should be the same.

But the other stuff we can and will change.

And maybe you have seen also this right here negative prompts in stable diffusion.

It's a bit different.

Stable diffusion is the only diffusion model where you also have negative prompts.

That means simply that you put in what you want.

See.

So we want.

See acne.

We want see bad hands.

We want see text.

We want see errors.

We want see messy fingers, extra digits, fewer digits, extra limps, and so on.

We simply include in the negative prompt what we will not see.

That's the basic concept of the prompt engineering.

So in the prompt we will type in what we want to see.

And ideally this stuff right here in the negative prompt we will include what we will not see in our

picture.

And yes this stuff works.

Extra arms, extra legs, extra foot and so on.

This diffusion models are not that great at creating fingers and limps, and for that reason we always

include extra arms, or extra legs, or extra fingers or messy hands and so on.

Like I said, you can simply copy the prompts and then tweak it a little.

And if you want to write your own prompts, just think about what matters.

And in the negative prompt, think about.

Also think about what matters.

And I would strongly recommend to include stuff like extra arms, extra legs, extra foot, messy hands,

missing arms, missing fingers, and so on.

Too many fingers, long neck.

This is all stuff that we can and should include.

Like you can see it also right here.

Maybe the neck is a bit too long of our influencer.

We can and should include in the negative prompt, maybe also long neck.

And that's the concept of prompt engineering.

Think about what matters and type it in your prompt.

And then use your negative prompt and also type in what you want.

See especially messy fingers and that stuff.

And in the next video, we will tweak our prompts to create our AI influencer in different environments

with different lights and so on.