16-01-2026, 09:30 AM
Experiments at generating Voynich pages based on an example (f9v).
The results are obviously far from perfect in all respects, in particular, both the plant illustrations and text layout are too “regular” and not much Voynich-like.
I think it’s interesting what they do with the text. Gemini starts by copying the model, but diverges after a while. ChatGPT appears to be more creative: at the end of line 4, it writes kaiin, which does not occur in You are not allowed to view links. Register or Login to view. but has several occurrences elsewhere; at line 5, it writes chkaiir, a plausible but unattested variant of chkaiin that occurs twice in f9v. daody, second line from the bottom, does not occur in the Voynich manuscript but looks like a plausible word; if one reads it as dchody, there seems to be an occurrence of f95v1. Maybe we are not far from some decent fake Voynichese (not sure if anybody needs that).
The results are obviously far from perfect in all respects, in particular, both the plant illustrations and text layout are too “regular” and not much Voynich-like.
I think it’s interesting what they do with the text. Gemini starts by copying the model, but diverges after a while. ChatGPT appears to be more creative: at the end of line 4, it writes kaiin, which does not occur in You are not allowed to view links. Register or Login to view. but has several occurrences elsewhere; at line 5, it writes chkaiir, a plausible but unattested variant of chkaiin that occurs twice in f9v. daody, second line from the bottom, does not occur in the Voynich manuscript but looks like a plausible word; if one reads it as dchody, there seems to be an occurrence of f95v1. Maybe we are not far from some decent fake Voynichese (not sure if anybody needs that).


![[+]](https://www.voynich.ninja/images/collapse_collapsed.png)