Why it’s interesting
UNDEFINED is not a general image generator. It is a digital exhibition that keeps assembling itself on the visitor's device. A small language model proposes themes and wall text, a diffusion model creates abstract images, and a speech model reads the writing aloud. The prompts and outputs stay local. The waiting, movement through the rooms, narration, and changing pictures are all part of the artwork, making the system feel like a transparent automatic curator rather than a form with a Generate button.
How to try it
Choose ENTER WITHOUT DOWNLOADS to inspect the gallery shell first. For the complete experience, choose DOWNLOAD & ENTER. The page lists about 90MB for narration, 2.4GB for painting, and 350MB for theme writing, so use a modern desktop browser, stable broadband, and preferably WebGPU hardware. Once the models are cached, move through the rooms and approach works to read or hear their labels. We verified the public page, both entry paths, model notes, and no-download space, but did not fetch the large model bundle.
Requirements and sources
The work is free and needs no account, but the full mode has a substantial first download. The official page credits Shahab Nedaei and names SD-Turbo or SDXS, Qwen 2.5 0.5B, and Kokoro TTS. Its metadata gives only the year 2026, so this record stores January 1 as a year-precision placeholder and does not claim that exact launch day. The cover is the official 1200 by 630 Open Graph image used only to identify the project.
[Enter the gallery](https://undefined.variable.gallery/) · [Artist project page](https://variable.gallery/undefined/)