Core Concepts
The author explores the process of creating a high-quality art print of their dog using Stable Diffusion and Dreambooth, highlighting the techniques and challenges faced in the AI art generation process.
Abstract
The author delves into utilizing Dreambooth, an AI fine-tuning technique for text-to-image models, to create a custom portrait of their dog. They detail the steps involved in training the model, building an image training set, running the Dreambooth training on Replicate, generating images with various prompts, fixing cropping issues with outpainting, enhancing details with inpainting, upscaling the image for printing, and finally preparing and printing the artwork. The project showcases the evolving landscape of AI art generation tools and techniques.
Stats
"40 photos of Queso in various poses"
"4000 steps for Dreambooth training"
"640x1280px final image size"
"2888x3835px upscaled portrait size"
"13x19” @ 300dpi full resolution print"
Quotes
"I love how the cartoony pieces of the original image were replaced with more fur-like elements; while keeping a consistent coloring scheme."
"The pace at which the community is developing new tools and techniques is astounding."
"The actual detail in print is better than I would have guessed possible."