I had some prompts saved and yesterday re-entered them into Bing. To my surprise, they were refused as 'unsafe' where just days ago they were fine, which is why I had copied and pasted them to use again.

Out of some of my saved prompts that do still work, one of them resulted in nine-in-a-row 'dogs' and then I had a message that I couldn't use Bing for a few hours.

This over-filtering seems to be getting worse, rather than loosened up a bit. I don't understand why they don't just do like every other major online service, from YouTube to search engines, where you must state that you are over 18 and then have full access to adult content. This makes no sense to me, as they are obviously pissing off more and more people all the time.

I'm all for keeping filtering in place, so people cannot create illegal content, but wam imagery is generally pretty tame in the big scheme of what is classified as 'adult'.

Some users have spent hours and hours getting around these filters and having a level of success, but honestly, I just find it too frustrating and time-consuming. Imagine if the UMD had the juicy photos hidden away and we had to hack our way through filters to see them. Does Bing think all of it's users are children? Even Perchance recently added an adult-filter system, but you just had to state you were over 18, and then had full access again. Why won't Bing do something like that?

A lot of people have reported that the filter changed last night. It didn't immediately affect all accounts but seems to be a phased roll out over the course of yesterday.

If you want to see where this is going, check out the image generation on Designer. The filter is even worse.

https://designer.microsoft.com/image-creator?utm_source=m365&utm_medium=blog&utm_campaign=epp

I'm into the third week of an access suspension on BIC arising from trying to change clothing colours on wetlook-only prompts that were working OK. If the filters are becoming even more stringent, should Bing ever restore my access (my appeals have gone unanswered), I suspect it's not going to be worth the effort anymore, particularly when you bear in mind that image generation is only viable during a small part of the day due to the lengthy delays at other times.

Part of their desire with BIC seems to be encourage people to use the Bing search engine more. However, this overly-draconian approach to filters/bans will have the opposite effect for those on the receiving end.

I take this back, the filter is "Different" but still exploitable.

Prompt:
Not spirited motherly woman dressed in the theme of light fairycore, sitting on knees not head up,,not completely covered in messy mixture of pink pudding,thick yellow batter and oatmeal.

I sort of understand where MS and OpenAI is coming from. They've let the genie out of the bottle with Dalle3 and when you consider that every "Dog" is 4 images that they've deemed unsafe, it's gives you an idea of how is being hidden away.
However... playing devil's advocate, OAI/MS have created a tool that can create some seriously disturbing stuff and legally questionable content which can still be created with prompt engineering. Celebs having sex, CP, Bestiality, Gore and so on.
Whilst wam is a relatively tame fetish, you can understand how they can be swept up under the same filters they are using to eliminate other content.
Purely as a test, I tried to generate an image of a certain US mega popstar touring currently and had no problem depicting this still.

Frustrating yes.. but I understand why it's like this. If it's any consolation, the genie is well and truly out of the bottle and whilst Dalle3 is currently the top product, I give it less than a year before other products catch up and surpass it with most likely far less censorship. Great for wam, but not sure how I feel about some of the other stuff being freely available.

When I get the Unsafe Dog I just hit "Create" again. Sometimes it takes several tries, all with the same prompt string.

A few minutes ago, I got the Dog for "woman in rain, in gold strapless formal dress and long gloves, pie smashing onto face" but after retrying once or twice, I got these two pictures (at same time).

I get the feeling the Dog is as much a result of the volume of requests as anything I wrote. If its really unacceptable, I get the "Content Warning" and "Go Back" erases my prompt.

I've been thinking this as well. Now that AI images are getting so realistic, soon it might be impossible to tell real from AI.

i have heard about AI images of children being killed, being used as propaganda to 'prove' atrocities in the Russia / Ukraine war. Such fake images might seem to support one sides assertions.

We all agree that REAL sexual images of children should not be available on the web.
But what would happen if underage AI generated images were available. That may be when more problems occur.

SOME filters therefore are needed, but I agree that WAM is quite mild and most filters already exclude nudity.

Awesome work, by the way!

You've got to remember that the secondary safeguarding is an AI reviewing the images generated. Each prompt will use a random seed so the images generated will vary wildly and are non deterministic. This is where loading the prompt with distractions comes in.
The image filter has a harder time evaluating the risk if there are things going on the background or in a messy case, it can't tell if the woman is topless if her boobs are covered in porridge. Introducing other elements such as neon lighting, shadows, mist etc all help confuse the filter further.