Quicksand Fans

Most of you are all familiar with Perchance, which I just stumbled across last weekend and spent way too much time goofing around with. Indeed, it's very nice to simply time "woman sinking in quicksand" into a text box and get a dozen images of varying levels of interest spat back out into your web browser in less than a minute, but attempting to mold that output into something more relevant to an individual's particular preferences can be challenging, frustrating, and rewarding.

I found it was easy to customize your model(s), the quicksand type and texture, the scenery, and to some extent the camera angle. Much harder were the pose and sinking depth. Poses I kinda gave up on; if I clicked the button enough times eventually I'd get something close to what I was looking for; "beach quicksand" in particular tended to have models far more interested in sunbathing then sinking.

Sinking depth on the other hand I had to really play with. After a TON of trial and error, "standing up" with "legs" in the negative column actually gave the best results for knee- or thigh-deep. Switching "standing up" with "submerged chest" got me to torso depth. "Submerged face" was pretty good for shoulder depth (though I had to add "underwater" in the negative field), and "submerged nose" would occasionally get neck deep. It helped a lot to remove anything that wouldn't be visible in the final shot, because the AI actually tries to draw everything you mention. No point mentioning that the model is wearing jeans if you're not going to see them! Anything below neck deep seems borderline impossible, but not for a lack of trying!

Still, as amusing as it is to generate a seemingly infinite number of pictures of twins (or triplets) doing their best imitation of a Pete Boggs drawing, you quickly reach a point where you feel like you've seen in all before, and you're just clicking the generate button like a slot machine waiting for the visual jackpot to hit: a one-off masterpiece that you can't seem to recreate no matter how many prompts you try. Then you're excited all over again.

Anyway, the images in this post were all created with this generator with the photorealistic style filter.

These first two are really just an example of what I was creating for my own personal enjoyment; I have plenty like this and won't bore you with the rest.


The particular generator I used was not great with facial expressions; most attempts at getting anything along the lines of "scared" or "panicked" ended up with the model being cross-eyed and/or with some kind of birth defect. Still, a few managed to surprise me.

A few more that I found significant simply because I wasn't able to replicate them, so they really stood out:

Nice job! Definitely can relate to that slot machine feeling, also the incredible effort it takes to create something unique once you can create "almost anything". It is fun to see the directions that different folks head in, which is maybe made clearest when looking at pages filled with images on deviantart accounts. Thanks for sharing your pics!

Welcome, that's a very good start. Perchance seems to have gained some options since I used it about a year ago. It is a good service though and one of the best of the free ones, in my opinion. You've given a pretty good assessment of perchance and I certainly agree with the "challenging, frustrating and rewarding" bit. For framing the images, certain phrases seem to work in most cases, "extreme close-up shot", "close-up shot", "long distance shot", etc and "profile view", "partial profile view", "front view" (you don't need to use this, most of the images you get will be front view anyway), "overhead view", "birds-eye view", "worms-eye view" (you may get some images of a worm's eye with this but it often works quite well) etc. For poses, yep, it mostly ignores what you tell it and does what it wants, it'll give what you want sometimes to tease you along! For depth you're pretty much spot on, aspect ratio can also help, landscape format and close-up shot gets head and shoulder type shots, portrait format up - to about thigh/hip deep and square format somewhere in between.

The slot machine description is spot on, it is like that. But prompts can be a lot more complex than it would seem from the small box they give you. I started working on prompts in notepad and copied/pasted them across to Perchance between "generates". This has two benefits, you can have something ready for the next generate and you end up with a list of prompts that have worked well in the past for further development.

Anyway, well done on these, have fun and keep posting.

nice start! i'm drawn to triplets, and the villainous smirker in particular, but several good images!

So I started playing around with a different generator that features a "realistic human" style filter. It actually produced some very lifelike results, especially close up. The issue relevant to our interests is that the generator seemed almost exclusively biased towards faces in full frame. That's fine neck- and shoulder-depth sinking, but a full-body shot took more trial and effort than I was expecting. Thankfully cerberus's guide to camera control was very helpful in getting the results I wanted.

But shortly after that, something kinda magical happened. I started playing around with props, something that Perchance in general doesn't seem to be great with, and got a somewhat unexpected result, which let me down another rabbit hole where I must have generated thousands of related images. It's when I realized that the true beauty of AI generation is not in trying to imitate what's been done, but in trying to do something that can't be done without a Hollywood-level special effects budget. Hell, at this point I'm inspired enough to consider subscribing to some more powerful paid services just to see how far I can go....

The government sold the construction of the old women's prison to the lowest builder, and as such it was never built on the most solid foundation. No one should have been surprised when after a week straight of unusually heavy rain, the entire complex began to sink precipitously into the soggy ground. As a result, the prison is now filled with a meter-thick layer of soft mud filling every corner of nearly every building. Yet, due to overcrowding in the prison system, there's no other place for these buxom inmates to serve out the remainder of their sentences. So the prison continues to operate under the nickname Cell Block Mud. Will these girls earn their freedom before the sinking prison claims them in it's muddy depths?

Glad you found the camera tips useful.

This looks like a lot of fun. I've done some closed room settings but never tried anything like this, well done.

the 3 prisoner cells are working nicely!

The next few posts are going to feature a tour of the facilities. The complex is made up of two connected buildings, creatively called The Good One and The Bad One. The Good One is slightly newer and features semi-private cells. I say semi-private because the cell doors are no longer in use; there's no point when there's so much mud in the way. Besides, it's not like anyone's going to be running in the hall...not unless they want to end up with a face full of mud, or worse!