And in another test the faces are super shitty.
Dall e also gives you 4 pictures per credit and dream 1.
So good to have more options I think. Two different products feeling different.
But I also played around with sd.
I still think my original comment is valid.
DALL-E is very good at conceptually representing complex prompt. Something like "a bear with a diving mask surfing in the ocean, a pelican is sitting on its shoulder", DALL-E will immediately produce coherent results, while SD requires lot of prompt tuning, and sometimes it's even impossible to get it to represent some concepts (I haven't tested this particular prompt tho)
SD is good for producing "artistic" images if that makes any sense
edit: ok I tried the "surfing bear" prompt with DALL-E 2 and SD and the results are consistent with my point, I put the raw prompt without tuning, and cherry picked the best image out of 4 with both models, here is what I got :
DALLE-2: https://labs.openai.com/s/Q9824QOfXln4r9FLFNM3v9v1
SD: https://imgur.com/a/czcMgiC
For SD, even by tuning the prompt I wasn't able to get the diving mask or the bird on the shoulder
Your criticism is in my opinion not valid.
Do they need to react to the market? Perhaps depends on what there goal even is.
Is dall-e 2 fun to use and cost wise totally fine? For me yes.
But I also have people running SD with a hacky webui on some good GPUs for free. How many people actually have access to it.
Is there also a good benchmark on which tool is inherent better? Because it is also totally fine to have multiple offerings.
I really don't sure if you ever seen product development for yourself.
Dall-e clearly took the potential misuse risk much further than others.