Stable Diffusion 2 was recently released, and at first had the full custom controls available, you could set the steps (more == better quality), the guidance scale (more == better follows prompts but possibly less good images) and select your own seed. Obviously it immediately got hammered into the floor so they've now taken the custom controls off and all you can enter is the prompt, presumably with the steps locked to the default of 25 and guidance scale at 7.5. But I managed to generate multiple series of images while the full controls were available, and while some of the results went horrendously into the realms of nightmare fuel, there were some good results too.

I'm also thinking that if posting to this group we should probably create new threads with descriptive titles each rather than all piling onto the first two, esp if using different AIs.

TBH I'm not that worried about being replaced with AI yet, while I think it certainly does have something to add, as yet at least it can't produce a narative series, and the results tend to be rather too random - you can't (yet) have it show a model in a specific outfit walking down a flight of stone steps into a bath of liquid mud, with the mud rising further up her clothes as she takes each step down, in a fully realistic manner. I know that is coming, but until it can do full-motion video with perfect control and reactions, it's not going to replace human models and producers.

Here's one of the series I've generated, the prompt was:

"a happy woman, eyes open, smiling, wearing a lightweight blue cotton boilersuit and hunter boots, sitting getting wet in a pool of liquid mud with her legs stretched out in front of her"

But you are thinking about it, you think one day I'll type in what I want to see and be able to just have it generated?

Great creations above, almost like rainwear.

I can see that when a) it can do realistic people with the correct number of arms, legs, and fingers, genuinely human looking faces, and mouths that don't belong in a horror movie, and b) you can actually have a narative series of images showing say someone walking down steps into a pool, or sliding down into a mud pit, or two people filling each other's clothes with mess, and c) do all of the above in 4k full-motion video on demand, then producers like me will be redundant, and in fact all fetish, erotica, and pornography, will instead be generated this way. There may still be a tiny amount of very high end, very expensive human-made content, for very rich people who really want real people in their scenes. But that'll be less than 10%, possibly 1%, of the market.

But we're not anywhere like there, yet. I think it'll happen fairly quickly, the tech is moving at a furious pace and major investment is going in. But there's at least a decade of development to go yet.


Thanks! I was asking for boilersuits (as in fabric ones) but often it seemed to default to plastic materials instead, "rainwear jumpsuits" effectively.

Aah an AI after my own heart! Lol