Using AI to illustrate nonsense poetry (DALL-E 2 vs Stable Diffusion vs Midjourney)
What happens when you feed gibberish to text-to-image AI models?
Comparing the output of DALL-E2, Midjourney, and Stable Diffusion text-to-image GAN models when asked to illustrate the text of the poem Jabberwocky.
This one’s a little different than my usual! I wrote up a whole script around this idea weeks ago, then never had time to record it, so I decided to upload the result anyway.
This was based on only providing the exact text of the poem, with no style guidance, then subjectively choosing the “best” of the first 1-4 images provided by each model.
I realize Jabberwocky isn’t strictly gibberish since most of the models have probably been trained on some human-crafted images already labeled with these words. I think the more interesting part is to pass in the sentences without any style guidance and see how much of a consistent default “point of view” each model has. Could you easily guess which is which given three more random images?
#dalle2 #midjourney #stablediffusion