The following are iterations 61 - 100 of the Deep Dream generated pictures based on the text prompt regarding the beast of the book of revelation: Revelation 13:1-3 "The dragon stood on the shore of the sea. And I saw a beast coming out of the sea. It had ten horns and seven heads, with ten crowns on its horns, and on each head a blasphemous name. The beast I saw resembled a leopard, but had feet like those of a bear and a mouth like that of a lion. The dragon gave the beast his power and his throne and great authority. One of the heads of the beast seemed to have had a fatal wound, but the fatal wound had been healed."
Following the images, a brief evaluation of the experiment.
This experiment with Deep Dream seemed to stay closer to the text than the other previous attempts.
"The dragon stood on the shore of the sea. And I saw a beast coming out of the sea."
This was not clear enough for the AI to understand that the dragon and the beast were separate entities. Few of the images produced show evidence that this was understood.
"It had ten horns and seven heads, with ten crowns on its horns, and on each head a blasphemous name."
Horns were nearly always present, but the count varied and was rarely close to ten. The crowns representation is debatable. Many depict what could be considered a crown, but not in the sense royalty.
"The beast I saw resembled a leopard, but had feet like those of a bear and a mouth like that of a lion."
Every image appeared to represent a leopard in some way. The bear and lion depictions were inconsistently mixed in. Most commonly, a leopard mixed with a dragon would be the best way to describe the beast produced.
"The dragon gave the beast his power and his throne and great authority. One of the heads of the beast seemed to have had a fatal wound, but the fatal wound had been healed."
I had trouble recognizing anything in any of the images that would illustrate a fatal wound, or a scar from healing from a fatal wound. If the AI did use this information, I was unable to identify it.
The images produced were interesting to see. Each is detailed well and full of curious details. I think iteration 77 may be the one I favor the most in the sense of the art appeal.
The main thing I learned about Deep Dream from this experiment was the confirmation that mixing different animal references causes difficulties for the AI. It seems to usually focus on one for imagery and omit the others. This is not wholly odd as the AI likely has few examples to draw on of mixing visuals of creatures that do exist. I am curious what the result would be if I gave it another 10 iterations with the text altered to steer closer to what I believe the goal would be.
Comments