Water Flows
Through My


As of Jan 31, 2022, this project is being showcased at Meta Meta, an online exhibition at TingDAO

The images you see at the beginning of this page are generated by the CLIP + VQGAN network. The title of each image is the prompt that was used to generate the accompanying image.

When you click on each image, a link will bring you to the detail pages, which includes the image in its full size, the prompt input into CLIP + VQGAN, and the paragraph from which the prompt was derived.

What is this project?

Water Flows Through My Closet is an Ai--generative project that explores the subject of bathroom experience using the most advanced neural networks* thus far in existence.
* Neural Network refers to a series of computational nodes that are used to compute the pattern of probability of any given dataset. It is a widely deployed structure for training higher level Artificial Intelligence. Such a process is also referred to as Machine Learning.

The two neural networks involved in this project are GPT-3 and CLIP + VQGAN, both owned by OpenAI and accessible to the public. GPT-3 is a Natural Language Processing neural network (meaning it processes direct inputs of human language, and generates human-like responses). And CLIP + VQGAN is a text-to-image generative neural network.

Both neural networks can generate textual or image responses based on prompts that have been input in human language. That provides the basis of this project.

I generate initial prompts, such as, “List five of the most awkward moments in a public bathroom :” I input the prompt into GPT-3. The GPT-3 then responds by generating paragraphs of what it thinks would follow such prompts. I then excerpt those paragraphs and input the excerpts as new prompts into the CLIP + VQGAN neural network, for which the neural network generates images of what it thinks the prompts are describing.

Virtual Environment

This is the bathroom model that I created for the purpose of displaying my project in a virtual space. When the model is shown at TingDAO, each bathroom stall would feature one Ai-generated image and the textural prompt fro which the image is born.

Exhibition Documentation


I would like to express my special thanks to Nan Nan and Echo Dark Art Space. This project was conceived during my participation in the Blind Project in the winter of 2021, for which Nan invited a handful of artists (myself included) and a brilliant tech specialist to explore artificial intelligence as an art medium. I also would like to thank Tim Zhang for his technical expertise and patience during this process.

I would like to thank Professor Bess Williamson for the inspiration and advice throughout my research on toilet design at the turn of the 20th century in the United States.

Finally, I would like to thank Jiaqi Zhang, Nicky Ni, Yuzhu Chai, Paul Cheng (OF Course, Shanghai), and TingDAO for their tireless work in hosting the Meta Meta exhibition and including me as part of the show.