I've been experimenting with various free AI imaging sites for a while and the results have been a mixture of disappointing and downright weird, but it seems that we're finally turning a corner where AI understands what gunge is and how it works. Here is a series of still images I managed to coax (with a fair bit of gnashing teeth) from Google's Gemini system using the Nano Banana model.
"Create an HD cinematic hyper-realistic image: Photograph of a woman sitting on a stool in a tall and narrow glass booth, her head tilted upwards as a torrent of marbled-green gunge pours over her. She is wearing black opaque tights, a very short skirt suit, high heels. The lower interior walls of the booth are splattered in the gunge as it falls. Standing next to and outside the booth is a tall cute woman in black opaque tights and very short skirt suit, leaning on the exterior of the booth and laughing. All of this is taking part in a television studio with an audience who are shocked and laughing hysterically."
This produced an okay image, but not one I was particularly happy with. The good thing about Gemini is that you can ask it to edit the photo for you (and it's free), so I was able to incrementally change the image to make it the way I wanted it (image 3) and then I created the other images with that one as a template - Image 1 is basically image 3 again but with all the gunge removed and an added description of the woman when she is clean.
It's all a case of trial and error and expecting the AI to get things wrong... a LOT.