Alice SchwarzeEveryone loves cute cat pictures so, I tried making some via #ML image generators. A ๐งต of successes and failures
I first tried good ol' craiyon/dalle-mini. I requested "a cat with a farmer's hat" and ... was surprised by just how bad the results were
#stablediffusion has been all the rage lately, so I thought I would try that next. The results look better. In a way, they do. But some of these seem to be drawings or paintings rather than photos and I wasn't sure whether it's just my brain being more forgiving when looking at drawings
So I updated the prompt to "a photo of a cat with a farmers hat" and got what looks like a set of photos. Except that these images are still far from the wholesome content that I wanted to generate. The cats all seem unsettling to me. It might be something about their eyes
Stable diffusion seems to like very discriptive prompts, so I spelled out in more detail what I wanted to have: "a photo of a cute lil cat with a cute lil farmer's hat" (I thought it was important to write "lil") and I think now #stablediffusion finally understand what I want. The cats on the left seem somwhat off, but I think the ones on the right are pretty spot on!
So after some journey, here are the two cutest pictures that #stablediffusion generated for "a photo of a cute lil cat with a cute lil farmer's hat" in full size
I kind of bothered me that one of the cats was missing an ear. So I thought about asking #stablediffusion specifically for "a photo of a cute lil cat with two cute lil ears" but I wasn't sure whether I would get human ears attached to a cat in a weird Frankenstein way. So I asked specifically for "a photo of a cute lil cat with two cute lil cat ears" ... and that was when I realized that I needed to share this story with you
To end this thread (my first thread on mastodon ๐ ), here is the best output that I got when asking #stablediffusion for "a photo of the cutest kitten in the world".
I have to admit, it is very cute, but the eyes are missing color. And of course, the cutest kitty in the world is my own not-computer-generated kitty Emmy โค๏ธ
person@aliceschwarze It's funny how StableDiffusion is.... simultaneously an incredible feat of engineering but also just as infuriating as talking to a robot has always been??? I think I spent 20 minutes today trying to get it to render an "illustration of a pink toothbrush". Apparently StableDiffusion does not know what a toothbrush is...
๐๐๐ ๐ป๐๐ ๐๐๐ 