Upper right: “The part of St. Peter will be played by Peter Sellers.”
There’s a newcomer! Dream Studio Lite!
I did Mystery Science Theater 3000:
It generated some weird things that look more competent than the dall but it sometimes ignored some of my words. Then I ran out of credits I guess because it stopped working. Sometimes it would blur out some of the options it generated. I guess it thought that Garfield the cat was too scary for public consumption!
Right-click and copy on Chrome made it so I didn’t have to use download credits.
Here’s “Poster for an Ed Wood film”
DALL•E 2 has some interesting new editing capabilities. You can start with an existing photo, and then extend outwards from it.
Dreamstudio.ai is just too much fun!
Prompt: Zap Rowsdower and Troy go on a Canadian adventure!
They were close! They got Mike Pipper!
Played with the settings with the same prompt and got-
Dezgo.com has no idea what Crow T. Robot looks like
“A topiary bush cut in the shape of Crow T. Robot from Mystery Science Theater 3000”
Sorry, Dezgo; that’s clearly cambot.
I wonder if the upstart dall-e clones are training theirs with things people have been searching up from the start instead of whatever general nonsense dall-e mini was doing. This one seems to know the person pretty good:
Dall-E (mini) was hardly the first text-to-image generator. Plus it would require all these projects to be sharing their prompt histories with each other. Plus it would then mean finding source images to match the prompts to train the networks. There are easier ways to go about this:
Duckduckgo has a pretty good idea too; this may not be coincidence:
It apparently only knows who Felicia Day is sometimes. Or a leg.
“Godzilla dancing with Felicia Day”