Flux from Blackforest Labs has been called the “Stable Diffusion Killer” and lauded much praise. In my standard image testing, meh.
Prompt:
The Duke Blue Devil mascot holding a sign that says, “Welcome Class Of 2028!” in front of the Duke Chapel on a beautiful fall day with autumn leaves everywhere.
It did get the text correct and there are a few other models to test, but this is not good. I did learn that Flex is the engine behind Grok.
Speaking of Grok.. Makes sense why it is pretty similar (which is actually kinda cool that it is), but still not good.
I gave Gemini a try, it did a great job with the Duke Chapel, but the Devils – aren’t great. The one that is actually Carolina blue looks more like the Domino’s Noid.
When I asked Gemini how it generated images, it seemed to not realize it can generate images. Even though I asked it about the specific tool it uses to generate images. It’s Imagen 2 (or maybe 3? I don’t know for sure)
ChatGPT nailed it. Chapel, Devil and spelling. Quite impressive. Better than I’ve ever seen with ChatGPT actually – I’ve never seen correct spelling using Dall-e 3
Co-Pilot was well… Co-Pilot…
