Dataset Preparation
To train the model you will need photos of the desired person, ranging from 10-15 pictures to an infinite. The most reasonable range is 10-50 photos. It's crucial that each photo should contain only one person. The higher the photos quality, the better: this includes both resolution and clarity. It's advisable to gather highly diverse photos: different lighting, environments, poses, and facial expressions. The less diverse your dataset, the more limited your final model may become. However, it's also important that the person's appearance in the photos remains relatively stable. Having significant variations such as weight, age, or facial tattoos in the dataset is not ideal and will lead to results instability.
In a general sense, considering the inherent randomness of the results, you can disregard all the optional rules and simply use all the decent photos you have. Graphic styles like 2D, 3D, anime, Pixar, and others can produce good results with any dataset. However, if you want to create something more realistic and resembling real photos, high-quality photos in the dataset are essential. Without them, generated faces may appear plastic and blurry.
Let's take a look at samples from the datasets used to generate the images from the Examples.
Danila Poperechnij | More
Karina Istomina | More
Lana Del Rey | More
Skryptonite | More
And let's also take a look at our today's main character the ideal model of which we will be training further - Billie Eilish.
Billie Eilish | More
As you could see, in half of these datasets I deviated from some of the described rules, but the results turned out at least decent. We'll find out how it turns out with Billie as we go further.
