AI-generated "Voynich manuscript pages"

[MarcoP]

I recently found You are not allowed to view links. Register or Login to view., which mixes two images taking style from the first and content from the second. It clearly works better when input and output are photographs, but in many ways it is quite impressive. All parameters were left at default values, with the exception of "guidance scale 12" which (if I understand correctly) makes the output closer to the content image.

After 14 months, I was curious to see progress in AI image generation. I uploaded the Voynich Sagittarius asking to reinterpret it as a renaissance painting of a crossbowman with a white hat. 

• Top-right is the 2024 image by Instant Style.
  
• Bottom left: Gemini Nano Banana. It keeps the dwarf-like proportions, manages to fix the right hand. It cannot fix the left hand and misinterprets the tail of the hat and some details of the crossbow, e.g. the loop at the tip).
  
• Bottom right: ChatGPT. This looks more XVII Century than Renaissance, but it could almost pass for an actual painting. It gets the beard of the figure and renders the tail of the hat in the correct way. Gemini gets closer to interpreting the crossbow's trigger thought.
  

Not perfect yet, but I would say that the progress is significant.