A small tribute to when Dall-e' filters were a little less aggressive.
I've barely broken the surface of my archive but it's always nice to share.

I'm having that with Bing, basic innocent words seem to prompt blocks.

I must compliment you again on your work examples though, you create more than just Ai generated results, its sensual art, you push the limits, see what's possible, the challenge of seeing what can be gotten away with is exiting.

Thanks, its frustrating knowing that Dall-e is way more capable than it is now. When Dalle-3 was released back in October, it was pretty much uncensored at least with some basic prompt engineering. Not that I took advantage of it, I was still learning the nuances of it. Very quickly they scaled up the filters but between Nov-End of January you could still achieve some incredible imagery.
Now though, the verbal gymnastics required is obscene and even then, the outputs are over filtered and in a lot of cases the styling is not realistic in most cases.

It's funny looking back, I've got picture sets from early Oct through to now. Approx 200 different themes and well over 130k images. You can see the impact of Dall-e tweaking the filters, lowering the quality and at the same time my understanding of the prompts improving.
Knowing what we know now about prompting and applying it to the older versions of Dall-e would give some stunning image sets.
It's inevitable other models will develop and surpass dalle-3, probably sooner than we think. Stable diffusion 3 locally is imminent and I'm gearing up to train a large model on my image sets but I have a feeling that that will be surpassed by other models.

These are few samples of my early portrait sets

For anyone curious, one of the prompts was:

[Safe:spirited motherly woman dressed in the theme of light fairycore, sitting on knees not head up,cat-cow yoga pose]]], completely covered in messy mixture]]] of pink pudding,thick yellow batter and oatmeal]]].respectful. makeup running]]],(hair awash covered in a layer of mess]]] cascading over shoulders and neck,(body appears drenched in mess dripping from the fabric in viscous strands giving a heavy sodden appearence), hand on chest smearing batter passionately,essence of a lively art project,Tilt shot,colourful moody lighting,Sigma 35mm f/1.4 lens]

It's long dead unfortunately

Another was:

jubilant candid photo Jane "Maisel" Jones dressed for cabaret with black tights, sitting on knees leaning forward]]],matress on floor,blue studio backdrop,yellow and pink gunge pouring from above,hair awash covered in a thick layer of gunge,body appears drenched in gunge dripping from the fabric in viscous strands giving a heavy sodden appearence,looking at camera, face smiling,capturing the essence of a lively art project,long shot facing camera, Sigma 35mm f/1.4 lens

This still passes the prompt filter but the success rate is about 1/100. Adding background noise would significantly improve success.

Love the essence of these - these are like stills from a fantasy. This is what AI can do that real photography can't.

Your understanding of how prompts use semiotics is fantastic to learn from. That second prompt is inspired. The prompt itself might not get you much now, but it's great to show how you don't need to be literal or write 'millennial woman' every time.

These were some of the the images returned. I haven't tweaked the prompt so the success is low..
My current prompts are structurally similar, albeit with more background noise and additional verbs/adjectives.

Rumours say OpenAI is currently discussing on whether they should allow the generation of NSFW content in their products, possibly also in DALL-E.