For the past few weeks I've been trying something different - putting my Poser renders through Wombo Dream... and acually getting something that more or less matches my render in composition but looks as if somebody with far more artistic talent than I has created it (although admittedly that somebody doesn't always seem to know how things fit together, or how many fingers and toes, or even arms and legs, people have) - - - - - - - I usually now leave the input image influence at normal with which, provided I select v3 styles, Wombo seems to do a very good job of semantic segmentation*. I've found that using low influence usually destroys my composition, while using high influence results in an output that doesn't improve my render much, if at all, although there have been notable exceptions.r Since Wombo Dream requires at least a single character in the prompt before enabling the 'Create' button I usually start with just a single word identifying the main 'thing' in my render. Sometimes this is sufficient; at other times I need to be a bit more precise. But my aim is to keep the prompt as short as possible. As far as style goes I tend to use the recent v3 ones as v2 and earlier, with influence set to normal, tend to produce results that bear little or no relation to my input image (I discovered quite quickly that most of the v3 styles seem to be far, far better than previous ones at recognizing the constituent elements of the input image*). I usually try several different v3 styles, hitting the once-apocryphal 'Make Art' button a few times - firstly because the first attempt after changing style often seems to ignore my input image, a secondly to check that I get consistently reasonable approximations to the composition in my render. This is the point at which I sometimes find that I need to be a tad more specific in the prompt. - - - - - Enough of the generics, on to the specifics of this picture. The first image here is one of my Poser renders of a work in progress that I doubt will ever progress any further, already uploaded to the Renderosity gallery as Unfinished (2016-04-11) - an attempt to recreate Edward John Poynter's painting "On The Temple Steps" in Poser. I started with just the single word prompt "girl" but ended up with the slightly longer "beautiful young woman wearing a loose transparent cotton robe sitting on a wall". *if you're interested have a look at this half hour YouTube video How Stable Diffusion Works (AI Image Generation), specifically around the 5-7 minute mark where he talks about semantic segmentation and biomedical imaging. Another good Youtube video on how text to image works, this time under 15 minutes, is this Vox one AI art, explained My reference(s): 2016-04-11 - OTTS .....(4F/OTTS-4F-MAIN.jpg)