Common Sense

Through an interactive generative comic strip, where characters April and June respond to user prompts, transforming them into witty nine-panel conversations which explore the creative collaboration between humans and generative algorithms.
Common Sense is a project I created together with Wallis Millar-Blanchaer.
Common Sense integrates AI in two key ways: a visual generation model fine-tuned with FLUX.1-dev using the flux-dev-lora-trainer to create imagery based on a dataset of 100 original graphics, and a fine-tuned language model (Meta’s LLaMA 3B-Instruct) that powers the characters’ dialogue. Together, they enable a responsive, generative narrative experience in which user prompts are interpreted and transformed in real time, highlighting the creative potential of human-AI collaboration.
Created using Stable Diffusion, Meta’s Llama (site was built using Next.js)
Creative coding, LLM, Computer Vision
March 2025
Users can interact with the comic through a responsive interface where they enter text prompts, which seed the ensuing dialogue between April and June. The project embodies multiple forms of interactivity: it's generative (creating unique comics for each prompt), responsive (adapting to user input), participatory (involving users in the creative process), and algorithmic (revealing how AI systems process and transform information). The nine-panel format creates a spatial narrative that unfolds in real-time, producing a hybrid creative experience that blends human intention with machine interpretation.Common Sense evolved through the exploration of generative AI models in ITP and IDM classes taught by Daniel Shiffman, Carla Gannis, and Alexandra Marranccini. Their guidance helped us navigate the technical complexities of fine-tuning AI models, while also encouraging critical thinking about creative computation. Through studio critiques and technical workshops, these mentors challenged us to maintain artistic agency within computational processes rather than viewing AI as a threat to creative practice. The comic format, a medium uniquely positioned at the intersection of visual imagery and narrative text— creates space to explore the nuanced dialogue between human intention and machine interpretation, as it unfolds in real-time. The nine-panel constraint not only makes the narrative compact and accessible, but at the same time reveals the interpretive capabilities, and limitations of the vision and language models. By intentionally creating our own dataset to include geometric animals and fantastical creatures depicted in bold primary colors, we aim to highlight how AI interprets and reimagines creative inputs within a clearly defined visual vocabulary.
Common Sense at the ITP Spring Show