"Happy young women enjoying a mudbath" and tried various clothing options, "Business Suits", "Maxi Dress", Triathlon Suit". The last few either didn't have an outfit specified or were just "clothed".

And finally "overalls and gloves".

And some rainwear...

I managed some very realistic fabrics with SD, and it would put people in a muddy scene, but I couldn't persuade it to make them sit in it.

Getting it do put people in at depth is the tricky bit, it's easier to get standing deep in mud (or at least muddy water) than sitting in it, and most of the time if it does do sitting it'll provide the model with a nice high dry bit to perch on - arrghhh!

Standing deep is a bit easier, as per my various businesswomen enjoying mud baths in their suits in the first post.

I just tried "sitting in liquid mud, enjoying a mudbath, pretty young woman, business suit, boots" and eventually it gave me this one (below), it did manage several other sitting ones but they tended to have too many legs, or odd extra limbs that were arm-legs, and random extra boots not connected to anything.

This is what I got from SD, amongst others.

I used "emerging from" and that seemed to work.

If you use the suggestions page, you can read what others used to get too results.

The one I'm struggling with it pouring gunge over, or pied in the face, sitting in cake. It doesn't seem to get it yet.

Some of those are really, really good. Back views AI does nicely.


Hadn't looked at the suggestions page, will do.

It doesn't really understand complex concepts. Think of it a bit like the liquid metal terminator from Terminator 2, when someone asks if it could become a gun, Arnie explains "no, guns have complex mechanisms, moving parts, it can only become simple stabbing weapons."

If you ask the AIs to draw things like railways, they tend to get them horrendously wrong, tracks that aren't parallel, wheels not on the rails, impossible curves. It sort of knows what a railway "looks like", but doesn't understand conceptually what one actually "does".

These are ace, amazing. What wording, phrases, and settings do you use? Do you doddle with the advanced slide?

I tried some non-wam prompts and got very mixed results. For example, typing 'baroque football stadium' doesn't produce a stadium with elaborate stonework and sculptures - it just shows rather wobbly aerial views of generic sports venues. 'Colourful football kits' produced some interesting designs but nothing wackier than already exist. 'Baroque electric trike' was a bit better and came up with some interesting concept vehicles. My understanding is that AI can only recombine things that have already been modelled and not really invent new concepts even when it recognises the prompts. There was a forum post where an AI had suggested how to arrange a chord for orchestra, and whilst its answer was perfectly 'correct' it was also very generic and functional and had no imagination. I suppose this is rather good news for artists.

I usually put in something like 'women plastic boilersuit hood rear view'. It struggles a bit with hoods, often drawing one up with another one down too, or a hood half smudged into hair, or hair coming down through the hood. Getting it to print images/text/numbers on the back is tricky too. I found it was quicker to add these manually in an image editor:

You got any more of that maxi dress one?

So this is the best i;ve managed so far.

I;m still struggling on Linux to do textual inversion to train the thing up, but I'm getting there.

I've gotten some really, really good "covered in mud" ones on my phone that'll post here. "Swimming in dress" just gets you what looks like a low quality version of a stock photo of someone swimming underwater. Also getting slimed works really well.