Concept
For this project, I created a series of self-portraits based on my photographs. All the self-portraits are united by a single style: black-and-white graphic work, a sense of nonchalance, simple line drawing, and realism. Yet, each portrait is unique in its pose, angle, emotion, clothing, and hairstyle.
Self-portraits that were used to train the model.
Self-portraits used to train the model.
I was curious whether I could generate similar self-portraits in a painted style using a trained model, where I and my facial features would still be recognizable, and whether the neural network could recognize my emotions and generate similar ones.
Working with Code
The code work was performed based on the source code provided in the assignment.
Installing External Libraries
Importing the dataset. In my case, this is 20 self-portraits, 1000×1000 pixels in PNG format.
Creating image captions and grouping pairs of images with their captions.
Creating a unique prefix.
Model characterization and model training.
Stable Diffusion Setup
Image Generation.
My Generations
When generating images, I aimed to understand the model’s capabilities—specifically, what emotions it could render on my face. As a result, I ended up with several variations for each emotion.
Joy
«painting of a DORA person in DORAS style, pure white background, smiling face»
Relaxed face
«painting of a DORA person in DORAS style, pure white background, resting face»
Sadness
painting of a DORA person in DORAS style, pure white background, sad face
Shock/Surprise
