//=time() ?>
I'm experimenting with a dumb idea for a loss function.
Which one of these do you like better? Same seed, same prompt:
'a professional #pixelart illustration of' +
['tree', 'car', 'house', 'dog', 'pc-98 portrait', 'skateboard', 'jumanji', 'donkeykong', 'lofi hiphop']
Here's what this prompt looked like when I *didn't* fuzz it, much closer to the target concept.
I quite like these!
Last tweet was 4cutn, this is 48cutn.
Not too much difference, but there's something there
Some more results for this prompt on the current version of my Quick CLIP Diffusion notebook.
Outputs each result in 60 seconds on a P100 https://t.co/AT5WK0ub0x
@dontnormally @GlennIsZen @PaulFishwick If you happened to go check it out immediately, please refresh, I made a couple more changes.
Here's the first result I got after running the latest changes on Colab (happened to be a P100)