Hi everyone! I am doing a PhD in collaboration between humans ad generative AI. I want to share the most exiciting ideas from this research with you.
Ultimately, I want to help more people stay up to date with the latest research developments on this field.
I am starting a series called 1 in 5. One big idea in less than five minutes. So here we go!
Idea # 1
The first idea I want to share with you is the concept of seeds!
I want to start sharing with you some of my research and things I find. I am starting a series called 1 in 3. One big idea, in less than three minutes.
The first idea I want to share with you is the concept of seeds in image generation AI. Let's dive into it!
**The problem**: it is hard to generate consistent characters across image generations. Every generation is new!
**The solution**: enter seeds. Seeds allow you somehow control image generation a bit more.
One of the main challenges of text-to-image models is being able to refine the same image, or generating consistent characters within images. Each new image is a new person.
I am going to show you how you can mantain somewhat character consistency using seeds.
Seeds are an important concept in machine learing and generative AI.
Generative Ai has a degree of randomness, that I can explain in another video. Each time you prompt it, there is a degree of uncertainty about what you are going to get. Defining a seed, which is literally just a number, allows you to generate the same thing for a given prompt.