media · Picked by us
A network that invented a script nobody can read
Gene Kogan trained a DCGAN on about a million handwritten Chinese characters so it could write glyphs that never existed — then mapped the space between the real ones.
The artist’s site
Official project page with the full gallery and notes
What it does
A 2015 study of machine imagination: new characters generated from a million examples, with interpolations that glide between real glyphs while keeping their structure.
Who it helps
Generative-art and typography people, and anyone curious what a model sees in writing.
How it was made
- Models
- DCGAN · Radford et al. 2015 — creator-reported
- Tools
- DCGAN · Training and generation
- Kogan trained a DCGAN on a labeled subset of about one million handwritten simplified Chinese characters.
- He then explored the latent space: single characters, interpolations between pairs, and visual arithmetic between concepts.
- The radicals that structure real characters survive the transitions even though the model never learned meaning.
What the person did
Gene Kogan chose the material, trained the model and curated the study — including the interpolation experiments that give it meaning.
What we know
Creator-reported: A first-person project page with method notes and a matching GitHub profile.
Access checked — 2026-09-26 · Opened the project page and the creator’s GitHub profile on 2026-09-26.
Can you use it?
The work stays on the artist’s site; we link only and host none of its images — the page states no media reuse licence. The cover is an original Made With Models illustration, not a frame from the work.
About this project
Picked by Made With Models. Added 2026-09-26. Updated 2026-09-26.