Making Drawing Accessible: Fine Tuning a Stable Diffusion Model to Output Anatomical Drawing References
By: Maria Girgis ’26
Advising Faculty: Timothy Becker
Project/Mission Statement:
One of the biggest roadblocks in a casual artist's journey is difficulty in drawing characters, specifically characters in motion. While there is access to many drawing references online, it is often difficult to find the specific range of motion and type of character one has in mind. I fine tuned an AI image generation model called Stable Diffusion to output anatomical drawing references. Through a drawing workshop with fellow Connecticut College students Sofia Blossom and Brooklyn Welch, a small data set was gathered to train what is called a LoRA, which helps inform the stylistic output of the AI model. Using this trained LoRA, several methods were tested, including the use of OpenPose to direct the anatomy of the figures. The intention is to provide accessibility to more artists to draw without the barriers of complex anatomy knowledge, and drawing techniques like foreshortening. The goal is to also help stray away from generative AI art by breaking down these barriers for any artist to draw more complex poses, using AI as a tool to assist, not as the artist.
Related Fields: Ammerman Center