I participated in a community event DreamBooth Hackathon organized by Hugging Face between Dec. 2022 - Jan 2023. The goal is fine-tuning a Stable Diffusion model with a technique called DreamBooth.
My choice of a subject was Rex the Toy Dinosaur from the Toy Story. The dataset (sooolee/rexthetoy) was obtained from the web. Samples from the dataset are presented below.
I chose version 1.5 of the Stable Diffusion model (runwayml/stable-diffusion-v1-5). The model was trained with a learning rate of 1e-06 and 1000 train steps.
The trained model is uploaded to the Hugging Face hub. Please visit my repo: sooolee/rex-the-toy-dinosaur
Here are some examples of inferences (the good ones! 😃) made with a guidance scale between 8 - 11 and inference steps of 300 - 400.
instance_prompt examples include: 'A photo of rxty dinosaur in Grand Canyon', 'A photo of rxty dinosaur in Paris with Eiffel tower as a background', 'A photo of rxty dinosaur in the Acropolis'.
The prompt examples include: '... in the style of Johannes Vermeer', '...in the sytle of Vincent Van Gogh', '...in the style of Michelangelo', '...in the style of Leonardo da Vinci'.
The prompt examples include: '...reading a book in a cafe', '...playing baseball', '...riding a bike in a city', '...surfing in the ocean', 'swimming in a pond'.
