Few messy game show slime round animations. Add some captions!

It's a dirty job, but someone has to do it...

Few more animated slimings. Happy Friday!

I tried to make a whipped cream on faces Flux LORA by clipping screenshots of online videos. It's blurry as heck, as you'd expect, but the results seem ok given a large enough sample size. I wish I had higher quality training data

Another round, mostly wet, a few messy. I've been working with ideogram this week, so quite a few of these originate from there, though there are a few from other sources. I really like the wetlook capabilities of ideogram, but I seem to have mixed success getting the results I want. Some inputs work great, others, not so much. I'm still not great at mess with ideogram, more often than not it looks too artificial for my liking.

They're not bad at all. Flux does a really good job even with 512x512 images. The key is trying to get a frame of video without blurring. You may have an easier time downloading the video and using some video software that makes it easier to run frame by frame so you can pick the clearest.
I would also say, it's much better if your image set is as varied as possible. 20 frames from the same scene isn't going to give as good as result as 20 from different scenes. Different poses, woman, clothing etc is going to net you better results.

Once you've got you screen grabs, you need to normalise the sizing. you really want to crop the screenshots to 512 or 1024 depending on the resolution of the image. If resizing to 1024 gives a blurry image, drop it down to 512. It's also a good opportunity to re-frame too.

With captioning, Flux is very flexibly and people are still trying to understand the best way of captioning. I've had great results with highly description natural language. Anything between 100-200 tokens to fully describe the scene. That said, I've also had great results from 3 word captions too.

I'm seriously impressed with fastflux.ai Seems to actually follow instructions and understand most things. It's not as heavily customisable post+hoc as your midjourneys etc, but it's much easier to use and pretty consistent (and uncensored AND blazingly quick!)

These are pretty good! Honestly, I'd say that even the not so great ideogram results are still much better and more realistic than the best one from Bing. Lately I've been also playing around with ideogram a lot and will upload some results, although they usually don't tend to get much attention round her.

Another batch of various pie pictures I liked recently .

Here's a few more from my custard splat set. Using Flux and Krea, with a few on a training lora, but still getting use to doing the training bit. It's all very trail and error, but getting there.

Just a few silly pics I made with a slightly tweaked LORA. I'm working on a new one now trained at 512x512 and hoping that increases the quality. I tried to upscale my training images with krea.ai but they looked so different I didn't like it. Anyway, bring these ladies to life and add some captions!

Some wet ones from ideogram.

Nancy seems to end up with all the mucky jobs. I don't think she minds, though. And she always stays cheerful and professional when dealing with dissatisfied clients.

Ok, I took my earlier "WAM girl" concept of a submissive with her mistress and made a party of it. The neighborhood wives are invited to witness this WAM girl get pied and slimed for their viewing pleasure.

It contains two animated gif slimings (the last two images), and some full-facial coverage pieings (not animated). I'm trying to figure out how to mix them more, but still working on the prompting and training data. Still no joy on animating actual pieings - they end up looking grotesque and way uncanny valley, but maybe my prompting is lacking.

Anyway, add some captions!

More of my custard pics. I'm trying to get it sorted the person is pouring it over themselves, or holding a pie.

Which website or app do you use to generate these images? Is it paid or free?

I use Flux - https://blackforestlabs.ai/#get-flux - to generate the images, which is open source/free but requires a powerful GPU to run locally. I use Kling.ai to make the animated slimings, which is a paid account.

There are a number of sites that you can generate images on too which use flux. If you want decent Wam images, you are going to want to use something like Civitai or another that has custom loras. These are smaller additions for specific subjects. The basic Flux model doesn't do Wam very well at all. I train my own models which I'm uploading to CivitAI. 0b1000101 is also training his own so don't expect to get similar results without using these models.

Yeah, very good call - I should have set expectations from the start. For my slime photos I use Cumzilla - https://civitai.com/models/712265 (you'll need to be signed into Civitai to see it since it's NSFW) - which is a cumshot LORA but it seems to have learned how to do slimings if you tell it that cum is green/pink/etc and tells it that cum is pouring from a bowl. e.g. one of my prompts looks like:

a high quality photograph in the style of cumface of a sexy Asian woman naked with a black leather collar that says "UMD.NET". Her face and forehead are fully coated in tons of thick, dripping, pink cum. Another woman is standing in a sports outfit and is behind her dumping tons of cum from a full bowl on her face. Her whole head and body is coated in the cum dripping from above her. There is a ton of pink cum coating the naked Asian woman's entire body and face in an excessive, sticky, messy layer. The naked Asian woman has a neutral expression while holding a white poster saying "I lost the round! Strip & Slime & Pie Time! Enjoy the show!" with a laughing emoji and she's looking at the viewer. Around her are women in sports uniforms laughing at her holding paper plates full of white fluffy whipped cream.

For the pie stuff I've been training my own LORAs. If I ever make anything good I might add it to Civitai, but right now I'm pretty "meh" about my pie LORA. I had to make a TON of images and sift through them for anything decent. But maybe some day! "Realistic" looking pies seem especially hard for AI to do imho. It loves to coat the faces in a white-face-paint like manner too often.

Hope this helps!

Come on in, the water's...slime?

Some messy ones from my collection

Some cake stomping here

Using bing

Some messy Midjourneys

More custard pics, a few random objects in some of these. That was a bit annoying as I did like the picture, shame about the random floating container.

Some animated slimings. Some lost bets, some lost rounds, some are streamers hitting their slime time goals. Various hues and quantities of slime at play.

Some new experiments, using the POV from the pie itself or trying to achieve maximum or side coverage.

Great results, do you mind sharing what you used to create these? Do you have an example prompt you could share?

I use Flux.1-dev - https://civitai.com/models/618692/flux - which takes up a lot of memory, like ~23 gigs of VRAM, so if you're gonna run it locally you need an RTX 3090/4090. You can also run it in the cloud using services like RunPod. There may be some cloud-based Flux offerings but I only know how to do it locally or on a cloud server by running forge - https://github.com/lllyasviel/stable-diffusion-webui-forge - to generate the actual images at 896x1152 pixels.

The secret sauce is the Cumzilla LORA - https://civitai.com/models/712265 (you'll need to be signed into Civitai to see it since it's NSFW) - which is a cumshot LORA but it seems to have learned how to do slimings if you tell it that cum is green/pink/etc and tells it that cum is pouring from a bowl. e.g. one of my prompts looks like:

a high quality photographic picture of a sexy skinny pretty smiling Asian woman with tattoos looking at the viewer with her eyes open and a black leather collar that says "WAM GIRL" holding a white poster in her hands saying "I LOST THE ROUND! Slime & Pie Time! Enjoy!" with a laughing emoji with a thick stream of liquid, gooey green cum pouring from a bowl above her all over her upturned face, coating the woman's face, forehead and hair in a thick layer of cum. The lighting is white and natural. It is a full body shot showing her flat stomach and white string bikini. The background is a gameshow set and there are women sitting in sports outfits laughing at the woman holding paper plates full of white, fluffy whipped cream.

From there I use klingai.com to generate the animations using img2vid, of which the prompts are pretty simple, e.g. "a woman having a bowl of green slime poured on her face"

From there I use ffmpeg to convert the mp4 to an animated gif:

ffmpeg -ss 0 -t 5 -i 0.mp4 -vf "fps=10,scale=448:-1:flags=lanczos,split[s0][s1];[s0]palettegen[p];[s1][p]paletteuse" -loop 0 output-0.gif

That's the whole process in a nutshell. No secrets here!

Ideogram ist just good at wetlook...