Halftime shoutout: Props to @l4rz@AydaoAI@pbaylies@Norod78 for ideas and starting notebooks. Go check them out if you find this stuff interesting (signal boost activated).
Here's the crazy bit, these were optimized using *only* the text, the photo for is for reference.
CLIP + StyleGAN FFHQ (noise vecs) is amazing! Same seed for each image (9):
"it was one of those rare smiles with a quality of eternal reassurance in it that you may come across four or five times in life"
vs
"Grey eyes that grow sadder than sunset or rain" #AI#art#poetry
I think the problem vs other methods (CC CLIP gang @advadnoun@erocdrahs@JanelleCShane) is an unstructured image+network is too complicated to optimize -- starting with a limited latent space (humans? cats? churches?) limits the space for CLIP to shine. Credit to @l4z for idea
More CLIP + FFHQ(noise) optimization (seed 27):
"A person with big anime eyes"
vs
"her white face turned upward the hands thrown out and clutched full of grass the clothing deranged the long dark hair in tangles and full of clotted blood" (Ambrose Bierce)
This is big ... semantic search is here. Let me show you with poetry. Yes poetry! I coupled @OpenAI's new CLIP model with 25K landscape photos from @unsplash. With it, you get a glimpse into how the AI *visualizes* poetry. A thread 🧵:
"Where Alph, the sacred river, ran through caverns measureless to man Down to a sunless sea." (Kubla Khan)
And on the pedestal, these words appear: My name is Ozymandias, King of Kings (Ozymandias)