(Twitter / @MadaraxUchiha88)

It's been a wild time in AI art and meme generation, as Bing's Image Creator was unleashed upon the world in recent weeks and was instantly used to create some problematic (and oftentimes, quite funny) imagery.

Since emerging, social media users were able to get Bing's tool to make things like Shinji and Asuka doing 9/11 and a slew of offensive Pixar movies, among other things.

However, in an apparent effort to stop all the edgy content, Bing and Microsoft may have rendered its DALL-E 3 AI image generator almost unusable — at least according to users.


As one can see in the above meme, the user who created this didn't exactly write "Create Full Metal Alchemist characters doing 9/11." Instead, they used a series of workarounds to make it look like the characters were flying towards the Twin Towers. Creators of various other "edgy" memes with the generator were able to employ several workarounds to get images such as a poster for Pixar's Caust.

Over the past few days, users of Bing AI have reported that they can barely get the generator to create anything, as completely innocuous terms are being flagged for "Unsafe image content." On Reddit, one user complained that they couldn't generate images of "Michael Jackson," claiming they wanted to make images of him as a heavy metal artist.

Another complained that prompts such as "Fish eye lens shot of woman walking away from the camera towards a seven foot tall macaque" were deemed "unsafe" by the generator. Furthermore, attempting to create too many images the generator deems "unsafe" will result in a ban from using it.

Additional users wrote that even using Bing's "Surprise me!" function, which has the website generate a random image, has the site return an "unsafe" error message — which would seem to imply the site is censoring itself.

Bing and Microsoft have yet to comment on the complaints of its users, though several have voiced that they're getting fed up with being unable to use the product for completely benign purposes.

Twitter / RW_Bain


Share Pin


Comments 11 total

Goofyahhman558

I tried bing ai image a while back. No cap it censored tomboy.

0

Stone Imp

I just want to make some skeletons playing basketball :(

1

LilaBirby

now we have to say everything but with other words

0

Victreebong

You know what, I’m about to say it:

This is why we can’t have nice things.

2

knekknak

What's weird about the blocking is how incosistent the it is, sometimes it will accept your prompt after a few tries even if you don't change the wording at all

2

Revic

Not at all surprised. This is how essentially any public-facing AI goes. Starts out great but able to generate "unsafe" content (whose safety is threatened is never specified), then gets lobotomized, then partially recovers but never regains its full functionality even for "safe" (again, for whom is unclear) content.

3

Geigh Science

part of the problem is that damn spilled eggs and coffee dog image they bafflingly chose to use for such situations is so fucking annoying to look at, it's very easy to just close the tab and give up after seeing it just once.

2

Imabigfish

Users trying to creating AI shitpost You cut up his brain, you bloody baboon!

1

King Crimson

Yeah I was playing around with it and the same thing happened to me. I wasn't even trying to make anything raunchy or edgy because I just assumed it would be blocked, but even prompts that seemed perfectly innocuous in my opinion got blocked anyway.

1

Moby The Duck

Yeah, I have noticed that the amount of "unsafe" and banned terms increased a million times over the days.
Stuff like "nun", "horns", "knife", "goth", "habit", anything related to religion, any trademark/copyright name, any human name, specifying a body part, all get caught in the "unsafe" filter.

Some stuff gets outright blocked, I was messing around with that "Smash Bros render" prompt, and using "Amethyst from Steven Universe" is a banned term, while the other gems aren't.
It would only work if you specified "Wearing clothes".

I assume there is a second AI that would "learn" from prompts to block anything considered unsafe, but because of people finding more ways to circumvent them, it went haywire and blocked most common used terms.

1
pinterest