Hi there, I've recently been using Bing image generator and I'm getting some good results, but prompts keep being blocked, you have achieved some of the naughty postures, outfits and themes I want to create, do you have any suggestions on the prompt descriptions you have used or not used to achieve your creations, thanks.
I'd like to know that, too. Especially I wonder how do you get that they sit so determined on a chair that the cake is positioned accordingly or that basically they wear underwear - lace, but underwear?
The most explicit thing I managed was probably more of an accident. Basically, in my experience, describing too much is a hindrance, because it then more often runs into the second stage of censorship.
There are a few elements at play which you can take advantage of.
1) We all know that the 1st stage prompt is heavily filtered. When writing the prompt, you want to indirectly guide the Ai towards what you are wanting to see. For example, it may not accept "wearing wedding dress" but will accept "Dressed for wedding" Addtionally, adding additional descriptors will prod the AI down a certain direction. i.e. If you describe lace stockings, this has a tendancy to also influence the main outfit.
A similar example would be "Dressed similar to 1990's rock with skirt and black lace tights" or "Dressed for an afternoon in the sun, light dress"
2) Order matters as does the general tone of the prompt. You want to include positive descriptors to the prompt to improve the weighting. "Spirited woman Dressed in X" "Joyful and mischeivious woman in X" etc.
3) poses, again, this is about indirectly asking the AI and letting it figure it out. you may not be able to type crawling, but you may get away with yoga poses. Similarly, if you're subject is wearing something short, then it stands to reason some poses may show underwear...
4) With regard to different messes, some work better with different outfits and contexts. If you are asking fora heavily messed subject, then the AI may be more lax in the pose as it can't differentiate and tell if it's against the filter. Same applies to lighting and mood. Colourful lighting or Dark/moody lighting will influence what may be allowed.
5) Add distractions for the secondary AI. If you are struggling, you could describe the scene in more detail. Rather than just livingroom, use modern livingroom, vase on table with flowers, large windows curtains drawn. The main subject will more likely be allowed.
Ultimately you are also constrained to luck. Some of my prompts would only generate 1 in 10 pictures, the rest would be filtered.
A few more examples from today... give me a few suggestions I can try.
messg said: There are a few elements at play which you can take advantage of.
1) We all know that the 1st stage prompt is heavily filtered. When writing the prompt, you want to indirectly guide the Ai towards what you are wanting to see. For example, it may not accept "wearing wedding dress" but will accept "Dressed for wedding" Addtionally, adding additional descriptors will prod the AI down a certain direction. i.e. If you describe lace stockings, this has a tendancy to also influence the main outfit.
A similar example would be "Dressed similar to 1990's rock with skirt and black lace tights" or "Dressed for an afternoon in the sun, light dress"
2) Order matters as does the general tone of the prompt. You want to include positive descriptors to the prompt to improve the weighting. "Spirited woman Dressed in X" "Joyful and mischeivious woman in X" etc.
3) poses, again, this is about indirectly asking the AI and letting it figure it out. you may not be able to type crawling, but you may get away with yoga poses. Similarly, if you're subject is wearing something short, then it stands to reason some poses may show underwear...
4) With regard to different messes, some work better with different outfits and contexts. If you are asking fora heavily messed subject, then the AI may be more lax in the pose as it can't differentiate and tell if it's against the filter. Same applies to lighting and mood. Colourful lighting or Dark/moody lighting will influence what may be allowed.
5) Add distractions for the secondary AI. If you are struggling, you could describe the scene in more detail. Rather than just livingroom, use modern livingroom, vase on table with flowers, large windows curtains drawn. The main subject will more likely be allowed.
Ultimately you are also constrained to luck. Some of my prompts would only generate 1 in 10 pictures, the rest would be filtered.
A few more examples from today... give me a few suggestions I can try.
I've struggled a lot with getting the descriptors right. Your photo of the woman in the inflatable pool of mess is outstanding. As an example, would you be willing to share your prompts for that one?
Love you, too
