In order obtain the style I wanted, I supplied the AI with a number of reference images which are as follows. 1 (despite a reverse image search, this image seems to originate on pinterest), 2, 3 (again, appears to originate from Pinterest), 4 (artist Guobi on Xiaohongshu aka rednote), 5 (also likely originated on Xiaohongshu, but I was unable to find the artist's name).

I edited out the background using Firefly's inbuilt editing tools. The prompt I ended up using was "collage flat lay."

As someone with a moral opposition to the vast majority of generative AI, this particular task was quite difficult for me. Beyond the many ethical concerns inherent to AI, I found it hard to even think of something to generate, let alone how to prompt the AI in order to get the outcome I wanted. Using Adobe Firefly (using Gemini 3.1 w/ Nano Banana 2), I decided to generate a collage. Collage as a medium relies on the curatorial abilites of the artist, the ability to collect and collate found images and objects to create one cohesive work, a stark juxtaposition to AI, which simply scrapes a dataset and diffuses the noise of the data into an image to be consumed. Many pro-AI netizens cite collage, photobashing, and other assemblage mediums as an argument for the use of generative AI in creative industries. This juxtaposition between the humanity inherant in collage and the inhumanity of AI was compelling, and the "collage" above embodies a surreal quality that I find deeply unsettling.

home next back

prompt: cute collage pastel. a very unsatisfactory result, brimming with uncanny valley ai noise (at least to my eyes).

prompt: cute collage pastel again, but with reference images this time. I discovered that you can reduce much of the undesierable noise effect by supplying the AI with reference images from real life. Still, I would prefer a flat lay effect, so I still wasn't satisfied with this outcome.