Tl;dr: I hope you enjoy at least some of these images - please give them a like or comment if you want to see more!

Long version: Firstly just to again thank renfield9 and others who've inspired me to look into AI generated images - I hope we can all share ideas and experiences to improve the quality of the images over time. It's been a fascinating and, quite often, frustrating past few days trying to produce images that I think are worth sharing. I'll just be concentrating on wetlook for the forseeable future as that's my main WAM interest and others have clearly been making significant in-roads into the messy angle and their efforts are pretty awesome!

I appreciate that my wetlook tastes are pretty niche and the fact that I'm generating AI images makes it even more so, as plenty of people will understandably only feel able to enjoy the real thing. However, it's precisely because there's an almsot negligible amount of what I like the most out there that's encouraging me to do this and there's only really any point me doing this if I'm producing what I most like to see. Indeed if I content myself with producing combinations of clothes and settings that would otherwise be almost non-existent in wetlook producers' real-life material, I'm happy that by posting AI-generated images I'm not risking taking anything away from what they do.

Like renfield9, I've been using the Stable Diffusion 2.0 via SeaArt.ai. This is mainly because it's free (although they state that's only temporary) and also because it's quite easy to use once you get the hang of it, with various configuration options readily available. There's nothing scientific as such in what I've done so far (although I've done a bit of research into prompt wording and syntax), it's really been about trial and error, and a lot of images are produced and discarded for one reason or another, with many falling by the wayside for reasons I didn't expect at the outset. Examples of this are too few/many limbs/fingers/toes (sometimes multiple heads!) and clothes having similarly Frankensteinian tendencies. I've also found very unattractive and weird-looking eyes to be a problem as well.

Just to say in pasing that wet hair is a key element of welook images for me, almost as much as the clothing. It seems a bit disingenous to me when real-life models are willing to get their clothes wet but strive to keep their hair dry whilst doing so. It has to be all or nothing for me to fully enjoy it. As far as AI wetlook is concerned though that introduces another problem - trying to generate images that have realistic looking wet hair and, on the majority of occasions, it doesn't work. Along with the appearance of the wet clothing etc, the realism of the wet hair is something that I'd like to do better with as time goes on.

As a consequence of the last couple of paragraphs, loads of images will have something that doesn't look quite right in them. I discard many for that reason, but do overlook blemishes if I think the overall image looks good. Nevertheless, every image production run (SeeArt creates 4 unique images for every attempt) provides a tangible sense of anticipation and excitement at what's going to be generated.

Trying to find realistic-looking AI models who can fulfil my criteria has also been a challenge. Some models don't work with wet hair at all, whilst I'm sure which model you choose can also affect how wet the clothes look, for reasons I don't really understand. As renfield9 spent time finding the excellent model "Am I Real", I've been trying to find other ones who also look good, rather than just effectively pinching his work (at least that's how I see it). There will be some "Am I Real" images in my profile, but the vast majority here will be from other models. Please note that many models are called "mixes", so won't always feature the same AI woman.

Just one final point before I start sharing some images. I have been, and remain, distinctly unhappy with the amount of blur (I gather it's called "bokeh") in the images I produce. I want a sharper, deeper focus and have tried various things to achieve this (e.g. prompts cotaining "deep focus", "depth of field", "prime lens" etc), but all to no avail so far. If anyone has any suggestions in this regard, they'll be very gratefully received!

Anyway enough waffle from me - on to the images! Hopefully there will be enough people out there who enjoy them sufficiently to make it worthwile for me to spend time sharing them (both directly in the forum and also others if you go into my profile). There have only been a handful of responses so far to the images I've posted elsewhere in this forum (many thanks to those who have), so please give the ones I'll be adding here after this post a like or comment if you want me to keep sharing more!

The first batch of images is from the model "KKW Ideal"

Edit: There should be about 45 images in all, but some have somehow gone missing during the uploading process, so I'll try to work out which ones they are and upload them later.

Edit 2: I've now added the 10 images that weren't uploaded originally.

This next batch features a model called "GrapeLikeDreamFruit". These names aren't mine, but assigned by the creators of the models!

All the models I'm using are available for anyone to add into their image creation process.

The final model that I've currently generated plenty of images for is called "SDVN Real" (note to self: get her to wear some dresses!). Almost all of the images featuring denim illustrate a problem that I've experienced on all but a few lucky occasions - the denim doesn't really look wet. Denim's not my own personal thing as such, but it would be nice to be able to reliably generate images featuring some denim to be able to have more variety.

Here's a selection for SDVN Real:

Thanks for sharing! There are some where the pose its so good that little imperfections do not bother me.

One of the images brought back a memory I had. Not related to AI, story time: On a hot summer day in the city, it rained. Pouring rain. I was sitting on the top level of a double deck train car waiting to depart for my commute home. A women enters just before the doors close. She is soaking wet, her hair was wet, and she was panting. She must have run in the rain to catch the train. It was a quit car where no one is allowed to speak, so you can hear her panting. She was wearing a very revealing white top. Her chest was wet and shiny from either the rain or sweat. She had her head tilted back, wet blond hair, moist face, eyes closed, chest out, and she was breathing deeply. She was the only one on the train who was wet because all other commuters were regulars who brought umbrellas. She really stood out. It seemed like she didn't even care, and was just relieved that she caught the train. I couldn't stop taking fleeting glances at her for my 30 minute commute. I really wish I could recreate the image of her.

I really like the idea that these images could inspire someone. Maybe it is a photographer, or maybe it is just a lurker trying to find their passion. It could be an angle, an expression, the outfit, or any other concept. There has certainly been photography on umd that has inspired me. It certainly inspired me to try learn how to train models. So maybe forget your umbrella purpose one day and enjoy the rain.

These are all quite good. Thank you for sharing.


Yes please!

I'm happy to oblige, my friend!

If only Stable Diffusion understood all the nuances of wetlook and water as well as it does the background scenery then I'd be an extremely happy bunny!

If any producers are looking this thread. Please get this guy on board as a director! These are incredible mate. You've achieved what I started out looking for, at least with the posing and clothes. I just wish bathrooms didn't end up so freaky, but outdoor shots like these are beautiful.

You have great taste in clothing for wetlook, and I would genuinely pay lots of money for a photoshoot in these styles

You're way too kind, but thank you all the same!

I'm just pursuing trying to recreate my own wetlook preferences as realistically as I can. As you'll have found in your own experiments, the AI engine spends most of its time toying with me cruelly, but acceptable images are generated sufficiently often to keep my interest and enthusiasm going.

My personal favourites are very niche as I've said, but it's really a lot of fun trying to pander to them like this without being dependent on others. I'd much prefer the real thing, but can no longer afford to pay for customs for various reasons so I'll have to do the best I can with this, albeit limited by the time I have available to devote to it.

If I get chance to upload the images, there's an interesting experiment I've been delving into on the way later and I'll be very curious to know what people think about it!

It feels like forever that I've been generating AI images for in some respects (must be all those discarded failures!), but it's actually only been a week. In that time, the idea's never occurred to me of having more than one person in a generated image. Apart from glitches in images I've produced where the AI includes the same woman in an image twice, every single AI image that I've seen so far has only featured one person - until today....

I came across what coincidentally was a wetlook image of 2 female lovers almost kissing. I wondered how difficult it would be to produce images like that myself and I soon discovered that images featuring 2 people of the same sex aren't yet supported as such by Stable Diffusion, at least not in a way that allows you to specify individual criteria in the prompts for each person. Therefore, users are basically at the mercy of the AI for any differences at all between the 2 people, whether it be their appearance or their clothes.

As I wasn't interested in having a man and a woman in the images, I decided to see what I could produce featuring 2 women and thought it would be a nice extra touch (because why not?!) to make them lovers.

Many images (far far too many) have to be rejected because the person in the image has inexplicable physical deformities, quite often very bizarre deformities, and having 2 people in the image is basically doubling the chances of that. Add in that fact that it's down to the whims of the AI engine how different it makes the people from each other, as its natural inclination is to duplicate the same person. By a lot of trial and error repetition, I've come up with some images that I think are worth sharing. The usual warnings about not to examine the images too closely for perfect accuracy still apply.

As I was experimenting with images featuring mediaeval royal dresses at the time I first became aware of the possibility of having 2 women in the images, I'm calling this sub-group of images "Ladies in Waiting", partly because I imagine them as waiting for what comes next (whatever that might be) and also because it would be more understandable for real ladies in waiting to have matching clothes!

Please forgive the similarity of appearance and clothing in them, I've explained why it is above and there's nothing I can really do about it. The engine also doesn't seem to have had too many images to learn from, as the poses are frequently pretty similar. It doesn't have any of 2 women kissing!

I hope some of you reading this will enjoy the images, warts and all! would be interested to gauge what level of interest there is in this sub-group to help me decide if it's worth the extra effort of continuing to produce them?

(For info, the model I've used for this on SeaArt is "StablyDiffused's Wild mix". It just happened to be the model I was trying out when I first started experimenting with 2 women in the images)

It is quite nice how you were able to get different clothing and hair styles. It is really hard for me as well to get different appearances, especially in their faces.

You also seem to getting a lot of good expressions on their faces. Some are quite connivingly real expressions. I feel like facial expressions are pretty important for realism. Most of the images I generate are either idle faces or smiling, like a stock photo.

I've been continuing to try out different models on SeaArt.ai and have found another one that works well for wetlook which has the catchy name of cineMaErosPG_V4. Here are a selection of images I've generated using that model.

I've been trying to vary the actions/locations a bit, but it doesn't take much to freak the AI engine out. It can't handle swimming or even lying down very well and anything more like complicated like trying to have a woman waterskiing or surfing for example confuses it! I'll try to think of more things to try to see if I can get some greater variety.

Can You generated high heels in water images?

I'm not sure why, but I haven't had much joy with getting shoes to appear in my AI images but, now that you've expressed a particular interest in that, I'll renew my efforts. Please keep an eye out for future image selections that I post.

Good wetlook= wetlook with high heels on

These really are excellent x