A Little Bit About
The Infinite Museum of Imaginary Objects
Culture happens when cultures meet cultures. Museums (where we keep our muses) store the outcomes. In this project we accelerate the process within the infinite realm of imagination, with some LLMs as the breeding ground.
Using archives from both the V&A in London and the Metropolitan Museum of Art in New York, you get to choose the breeding pairs for the experiment. The models will then merge them into brand new hybrids for our digital museum.
Come, let us generate new muses for our brave new world!
The process begins when you select two objects, one from each archive. The first item determines the style of the new object (e.g. surface details, texture, decorative motifs, period etc), the second determines the form (e.g., object type, core structure, silhouette etc).
Contributes the style
Contributes the form
The Prompt
You are an expert museum curator with a deep knowledge of art history and also an artist with a very creative imagination. You will receive two images of historical artefacts.
Now I want you to imagine a brand new, imaginary artefact that is a fusion of the two images.
Use the first image to determine the texture, details, and style of the object. Use the second image to determine the form and the type of object it is.
The two images are sent over to the "gemini-3.1-flash-lite" LLM together with a text prompt. The purpose of this step is to generate another prompt that will describe the imaginary hybrid object to a second text-to-image model.
The prompt was designed to ensure a realistic-looking yet imaginatively distinct new hybrid object.
LLM is asked to imagine a hybrid image
Gemini produces a prompt that describes the new hybrid object
Additional Prompt
The object should be realistic version of the thing that it is. Documentary color photograph of a real museum artefact:1.3, resting on a conservation board with a dull contact shadow. (film grain and the slight fade of a scanned archive slide:1.2). Practical light from the upper left, a little uneven, with soft falloff and no fill flash. The object shows pores, tool marks, chipped edges, dirt settled in crevices, and uneven pigment. Background is matte grey linen. Focus is clear on the object and a little soft at the edges, the way a conservation studio photograph looks.
Combined with the prompt from Gemini
Negative Prompt
cgi, 3d render, octane render, unreal engine, blender, cinema 4d, plastic, waxy, airbrushed, smooth perfect surfaces, hyperdetailed, hdr, oversharpened, product render, seamless gradient background, floating object, ray tracing, subsurface scattering, video game asset, concept art, illustration, digital painting, pristine, flawless, glossy, perfect symmetry, unrealistic.
Passed as the negative prompt
The prompt generated by Gemini is supplemented with an additional prompt designed to maintain a consistency in how the imagined objects are displayed. A negative prompt is also added to prevent the final image from looking too fake.
All of these prompts are then sent over to the "FLUX.1 [schnell]" text-to-image model to generate the final imagined object.
Text-to-image model receives the prompt from Gemini plus my own
The final output
Coda
The website was set in versionFont by Xavier Klein,
& Necto Mono by Collletttivo
The project was conceived, designed, and built by ben elwyn
If you like this and feel generous, or want to support other strange projects
you can buy me a coffee