All language subtitles for 4. Seeds, the magic Number for consistent character in Stable Diffusion

In this video we will talk about seeds.

The seed is simply the first starting noise of the diffusion model.

You have learned that all these models make pictures out of the noise, and if we give the diffusion

model a seed, the seed is a random number.

The diffusion model will start from a specific start point.

If we give the number ten, for example, the diffusion model will look at the 10th picture where it

was trained on, and it will generate a similar picture.

If we take a random seed, we will get more random pictures.

That's the basic of the seed.

Now come with me in card and take a little look.

So we have here the prompt.

We have also lauras and so on.

We will take a closer look in the next video.

But first I want to show you simple what the seed does.

We have our prompt and we get the same picture twice, exactly the same picture.

And if I scroll up a bit, we have the same prompt, but a different picture appears.

That's the basic of the seed.

So if we press this right here, you will get this.

In the next lecture we will take a closer look.

If I press generate here we have the seed.

If the seed is minus one we will take a random seed.

If we take the seed for example at 20 we will use the 20th picture to start our generation.

If I press send now, we will get a picture that is different than this picture, but it's the same

prompt and all.

If I do this again, so simply generate and I use the same seed.

So 20 we will get the same picture twice.

And then for proof I will do it one last time I press generate, but this time I will use another seed.

We can use a random seed or even just one picture lower and we will get a different output.

So now I will send this and you will see.

So this is the first picture.

The second should be exactly the same pictures.

Yes you can see this.

So we have the same prompt, the same seed, the same lauras and so on.

And we get exactly the same pictures.

And then we take a seed that is a little bit lower and we will get different pictures.

Yes.

You can see we have nearly the same girl because we use Laura's, we use checkpoints and so on.

And we use of course the same prompt, but simply the seed does here the magic.

We can create exactly the same picture twice.

If we use the same seed, the same Laura's, the same checkpoints and the same prompt, same prompt,

same Laura, same checkpoint, but different seed will generate slightly different variations of our

picture.

If we take same Laura, same seed, same checkpoints, but a different prompt, we can create the most

consistent characters.

I would argue that also this works like really really fine.

This are also nearly the same girls, but you can see the seed is here.

The missing link that some people don't get.

In this video you have learned what the seed is and what the seed does.

The seed is simply the first starting noise of the diffusion model.

Same seed with same prompt will generate exactly the same picture, same prompt, but different seed

will generate slightly different variations of this character.

Same seed, and a little bit of a different prompt will take our character, and we can make our character

in different environment agents.

But you have always to check that you use the right, Laura's the right checkpoints and so on.

And in the next lecture we will create some pictures in card because this was just the illustration

of the seed.

So in the next video we will take a closer look how card works and then how we can generate perfect

prompts for card.