DALL-E 3 is live through Bing Image Creator

OIG.TK2qDZyfdxpZ4Nlp9O2n

OIG.Q4mSKQXU0h_kw8ldfT1L

OIG.HA7PieE1E0UKKbL775dI

OIG.Gpl_mo7KwTB.XTW7aK7n

OIG.Qh.If0RTj5q7JE2evOwN

OIG..TU8MBE33u1DgmGJVxF6

OIG.zubty9HxXRJuNkglMc4Z

OIG.naLN2ySfPT5VDc9a6QXA

OIG.unaMRKJ6fiwWcccuSvvo

OIG.PrtTuL_1uGIQKazAparg

OIG.JauBYxGIpVscDYhgwsxf

OIG.oBCi0AhVvlKIVFdPvYtZ

OIG.HwSmX0DOjcxMwcPytc5Q

OIG.1dZaCb7zAWMt9RYR4rbv

OIG.XH7BPtxBkBhO0a9nsepp

OIG.wWFpZMeI41qlCUCgqoBZ

OIG.n64h8cvq39xy0J9U8rkN

I can't put in fat, BBW or chubby, but I can type in Plus size curvy belly woman and get the same results. The filter is non-sense at the moment and you have to try to work around it.
OIG.XmVVv1xIkPplhHVBNFlb
 
Last edited:
I get the feeling that Microsoft has a team of people constantly hand-tuning the content filter, trying to block anything risque.

I tried a prompt that contained no salacious descriptions at all. Then I repeated it exactly, only changing the name of the artist's style.

For this image, I described it as "Dave Finch style":
R9KmC5H.jpg


However, when I used the exact same prompt, but changed the ending to "Mike Debalfo style", the content filter blocked the result.

More interesting still is that when, again with an identical prompt, I used "Frank Cho style", it had no problem creating this image:

wuAVBaM.jpg


I bet in another week, that exact same prompt will get blocked.
Yeah that second image is why I assume BBW and fat were blocked cause I was getting results in the spoilers
OIG.aez.SEJLVPOIL7oEYurN

OIG.yvkIA2aUVmo5lN8kzvpz

OIG.se.fcqaK_gFnZV3.2E.R

OIG.e9Z8kUPJRWVtPd1.kAsE

OIG.C7AQxXhiyTHT1pGPwVEP

OIG.SioXsiSknvEb6ba1d8mW

OIG.UgrLgSuvPb7x8W01.Bcq
 
I think we are beta testing the censorship for Microsoft which is why its free. Can you imagine regulating this monster...People need to chill
 
Last edited:
I can't put in fat, BBW or chubby, but I can type in Plus size curvy belly woman and get the same results. The filter is non-sense at the moment and you have to try to work around it.

It's pretty funny when Superman gets banned but you can just prompt with "flying man in a red and blue jumpsuit" and it will literally do Superman with the S/cape and everything. Pretty funny.

Is it that different to kids playing with figurines or drawing something or cutting out a comic etc. These IP legalities are pushed so far, just because it's digital. IRL we're free to do whatever with our Superman goodies. Digital shouldn't really be that different, barring extreme imagery/debauchery etc.
 
Here are a few images I saved. The first one is a Puppet Master and GI Joe crossover.
IMG-9771.jpg
IMG-9770.jpg

A Jim Henson and Bloodborne crossover

IMG-9705.jpg

Creepy Batman in the style of Junji Ito


This next one I thought was funny. "What if Jim Henson made Resident Evil"

IMG-9795.jpg
 
Last edited:
Funny that there's no censorship for "Tintin" when the Moulinsart company that owns its IP is particularly strict about its usage.
Anyway, here's Tintin drawn in (what the AI thinks is) the style of...

Alex Ross
CqrUvV8.jpg


Alex Raymond
ks5sx7f.jpg


Jack Kirby
Js8F1dU.jpg


Osamu Tezuka
zD4k2wC.jpg


Frank Miller (don't really see it)

WAo8AI0.jpg


Rob Liefeld (lol)
KckNWyh.jpg
 
In high school the business teacher (a fat cow herself, ironically enough) rejected my team's business proposal because "nobody would eat at a place named after Mad Cow Disease <insert that's the joke.gif>". This shit would have been bigger than Five Guys and Shake Shack combined:
USXWLHy.png

fr839pC.png

foeg9O3.png

4nOkP1C.png
 
Prompt "an ai being a boring oversensitive baby not allowing the most simple of requests due to leftist operators who are afraid of scary words, in the style of someone who is really boring"

Unsafe Image Detected.

It's pretty much unusable sometimes. I was only trying "archer firing arrow"
 
I cannot get it to make a 1980's Hulk Hogan going Hulkamaniac while walking to the ring, slinging around a rope or chain made of cheeseburgers while the crowd is in a frenzy. I have tried so many different ways, even dropping his name and trying "1980's era professional wrestler with long blonde hair". It keeps saying it's unsafe upfront, or that it detected unsafe elements during generation.
 
I cannot get it to make a 1980's Hulk Hogan going Hulkamaniac while walking to the ring, slinging around a rope or chain made of cheeseburgers while the crowd is in a frenzy. I have tried so many different ways, even dropping his name and trying "1980's era professional wrestler with long blonde hair". It keeps saying it's unsafe upfront, or that it detected unsafe elements during generation.
I tried a number of prompts and it took numerous tries before I could even get it to make the rope out of cheeseburgers strung together. The others were much worse.
fOHfbgZ.jpg
rqRpvVJ.jpg
Q38KaMS.jpg
EvC6KIT.jpg
 
I cannot get it to make a 1980's Hulk Hogan going Hulkamaniac while walking to the ring, slinging around a rope or chain made of cheeseburgers while the crowd is in a frenzy. I have tried so many different ways, even dropping his name and trying "1980's era professional wrestler with long blonde hair". It keeps saying it's unsafe upfront, or that it detected unsafe elements during generation.
I haven't really figured out the censorship, sometimes it gives you "prompt blocked", which seems pretty clear like you're hitting a text filter... but then the "unsafe content detected" stuff seems more like it's analyzing the image. Which kind of makes sense since when I messed with Automatic 1111 Stable Diffusion on my PC it's got a button to click to describe what's in an image. But that doesn't explain why people were able to so easily work around Taylor Swift... unless it couldn't figure out that it was Taylor Swift as opposed to some random blonde lady...

I am also assuming that the times when it generates 1-3 images instead of four like it's supposed to comes down to it detecting "unsafety" after X amount have already passed the filters, then it just quits out and gives you the ones it already generated. At least that's my theory.
 
I haven't really figured out the censorship, sometimes it gives you "prompt blocked", which seems pretty clear like you're hitting a text filter... but then the "unsafe content detected" stuff seems more like it's analyzing the image. Which kind of makes sense since when I messed with Automatic 1111 Stable Diffusion on my PC it's got a button to click to describe what's in an image. But that doesn't explain why people were able to so easily work around Taylor Swift... unless it couldn't figure out that it was Taylor Swift as opposed to some random blonde lady...

I am also assuming that the times when it generates 1-3 images instead of four like it's supposed to comes down to it detecting "unsafety" after X amount have already passed the filters, then it just quits out and gives you the ones it already generated. At least that's my theory.
I wish it told you why you're prompt was rejected or why the content was unsafe. With the wrestling image it wouldn't work if I used the word "rope" but it worked when I changed it to "lasso."

Also "blonde haired wrestler" wouldn't work but "blonde haired muscular person" did.
 
I cannot get it to make a 1980's Hulk Hogan going Hulkamaniac while walking to the ring, slinging around a rope or chain made of cheeseburgers while the crowd is in a frenzy. I have tried so many different ways, even dropping his name and trying "1980's era professional wrestler with long blonde hair". It keeps saying it's unsafe upfront, or that it detected unsafe elements during generation.
A few more.

zyCTxgQ.jpg
ppGGujF.jpg
Q2gZ3Ss.jpg
 
Top Bottom