A New day, a new beginning

[cerberus]

I'm being a bit lazy today and am posting some more from the Red/Green set, solos this time. I have been experimenting with SD today, but it has been on ControlNet (using pictures to constrain the AI) so lots of very similar results as I try the different option (and there are lots of them).

Nicely desolate, resigned to her fate.

00239-2862582691.png

She wasn't expecting this when she stepped backwards.

00315-496834679.png

I could see people buying this outfit if it were on sale.

00544-2907904503.png

Probing with her feet. Just how deep is this?

00556-3302320964.png

At peace, awaiting developments.

00599-164259466.png

[Duncan Edwards]

Interesting design work on the wardrobe. Curious about how much of that is a result of your input and how much of it is just AI doing it's thing?

[cerberus]

Interesting design work on the wardrobe. Curious about how much of that is a result of your input and how much of it is just AI doing it's thing?

Mostly it is SD, my input was as follows:

"two attractive European women 25 one wearing red green latex and high boots wade into very deep treacherous quicksand mud, in heavy dense fog muted colour grading ..."

As you can see instructions were fairly simple, there was quite a bit more regarding desired quality etc and the negative prompt. Instruction on clothing was loose and so was SD's interpretation. I don't think I altered the prompt much during the various generation runs, I was happy with what SD was doing.

For today's set I've gone back to the missed opportunities pile and King Solomon's Mines, this had a quicksand scene but not the one we all wanted! Below are a few of my take on how the quicksand scene should have looked. It is based on a picture from the movie promotions but somehow doesn't transfer as well as expected. The pics are nice but, to me, don't look that much like the desired subject, though I'm not sure why.

00137-4246163132.png

00175-4086222467.png

00246-656631232.png

00311-3309662794.png

00372-819613571.png

[cerberus]

I've messing about with ControlNet again, trying to us reference image to pose the generated ones and see how close I can get to the original.

The originals are in this thread if you want to check them:

viewtopic.php?f=15&t=71

Not that close but OK in its own right.

00049-400983336.png

Pose is spot on, clothing different, there was one that was closer to the original but I liked this one better.

00057-400983336.png

This is pretty much spot on, I was surprised at how close SD got with this.

00117-400983336.png

The pose is good, but I couldn't get SD to put in the footprints that are in the original.

00221-2006313350.png

Not quite there but nice anyway

00255-3744626685.png

[Viridian]

I had a go using your photomanip. Is this what you were going for?

[cerberus]

I had a go using your photomanip. Is this what you were going for?

Well, I was trying to see how far I could get towards replicating the look and feel of the photomaip, but I didn't have any expectations of getting as close you have succeeded in doing. I'd been trying to do these using Txt2Img and ControlNet in SD, obviously that isn't the way to get really near the original. So I've had to open a new can of worms; this morning I've tried Img2Img with ControlNet and made some progress (see below), I'm guessing this is similar to how you did yours. These things are a bit like trying to drive a car with 15 gearboxes and 12 steering wheels aren't they!? Anyway, thank you for pushing me in this new direction, It's been and interesting morning so far.

00094-2509209076.png

[MadMax359]

"not quite there"? looks just fine to me!

[cerberus]

"not quite there"? looks just fine to me!

No, she reckons about another 10 minutes should do it!

[dlodoski]

"not quite there"? looks just fine to me!

No, she reckons about another 10 minutes should do it!

Very nice work.

[cerberus]

Very nice work.

Thank you. I am pleased you find it so.

Today's set is a collection of experiment results. I was experimenting with Img2Img (the SD facility for basing an AI generated image on a source image). As up can see from above you can get extremely close to the original image but if you ease the control and let SD do more you can get some interesting results.

This came from a cutout that never made it into a photomanip. It's a girl on a white background, but I gradually eased the constraint. The first result looked like she was sinking in snow, later SD started putting in realistic backgrounds and changing the appearance of the girl, so in this one she is quite different from the original and the background is believable.

00153-3280337152.png

The next two were done similarly but were based on a photomanip, the end results are quite different from the source.

00194-1213016933.png

00197-1980104404.png

This is from another of my photomanips, the pose is similar but the setting and clothing have changed.

00228-3754794057.png

Again, this is from a photomanip but pretty much everything has changed apart from the basic idea.

00329-743404663.png

It is quite intriguing what you can end up with using this process!