Naaaa, thanks to yoU!!!

are doing weighting adjustment within the prompt with bracket? Using too much brackets and even worse: nesting them, you can mess up a basically working prompt.

Another thing are dimensions, like being to "unusual" with protortions of the image or too far away of what the model has enough proper data been trained on.

some Loras work only properly on certain base models or derived models - mixing up loras can have also unexpected and often more negative effects.

But as the OP said: Without seeing a full generation information, the possible reasons or combinations are too countless to point you on the right way. It could be as unobvious but simple as an accidently toggled flag that messed up a single step, to change your results from acceptable to - what the hell is this. Like a control net instance you accidently toggled to the wrong processor (that did cost me quite some hours to discover).

A general first tip, if you are affraid of sharing generation parameters: Are u using a CFG of 10 or higher and your query is complex and maybe also using a lot of weighting like mentioned before? Lower the CFG to test to a lower value, change to 5.5 and look how the results change. Get better? That basically means: You asked the AI to be focussing the prompt too much, limiting its creative freedom, while beeing too specific, maybe too nested on your prompt, or even a typo somewherre could make a huge difference. It starts to look weirder, but in a different way? Increase the generation steps. Using a CFG Value below 10, means, it wont do the full number of steps you enter, lowering your CFG that much, might require you to adjust your steps number - the same applies the other way around.

Do you use upscaling as part of the poblematic generation? Disable it, to exclude it. If its better, you might be using wrong upscaler for the task, trying to scale up too much with it or not the right number of steps on upscaling.

You can see, there are so many potential sources. Even more complex ones like a messed up or wrong optimizer used.

About the watermarks... the only way to avoid them a 100%, so you dont need a remove step for sure, if it's a model being trained on handcurated and adjusted training data. That begins at the base model. Alternative? Create an image file containing a mask with the watermarkl positions that appear, send your txt2image output together with the mask you prepare to inpaint - in most cases it has no issues making the watermark disppear in the background.

So , post your generation details and we can most likely help a lot more ;)

All I get is close-up.

I got fisted near every day 😝😈

@pyros_sd_models should I take a Photo? 🤪 Look @ my Pornhub Profile

how about throwing other people's photos in the PNG info tab, send to txt2img, choose same checkpoint? at least you can generate same photo

Try (female, female, woman, vagina, female, girl, woman, girl, nude female:1.3) in your negative prompt.

Have you added add_detail and more_detail Lora’s? They work wonders after tweaking

lots of creative potential in this model indeed

@pyros_sd_models  Dang, yeah you guy have brothels and whatnot you can go to and enjoy attractive willing women. For us in "freedom land" we would be risking a felony and jail time, so it's just pictures/videos for those of us who can't date such women.

@pyros_sd_models Sorry -- just noticed this. Welp, in that case, I herby declare you to be an honorary American, on humanitarian grounds! Please remember to show this comment at customs for your free beer and complimentary assult rifle! 😄

blender + 3d models posed as you like + depth map export from blender + controlnet + this model = WINNING

@tvcomputer Have any pics you’ve made with this workflow? Curious to see the results!

made some guided penetration lora, for this model, might do a wider sexual position lora soon

@pyros_sd_models what??oh man, we are not ready for it

Working fine here in Forge

What do you mean, what lora should you use? to what effect?

@kunde2 yes lora

Ditch Foocus, use Forge instead. Foocus is very limited, and dont like nsfw that much.

@madPsyko can i run forge using google colab?

@syeterian981 literally one of the first google links:
How to run SD Forge WebUI on Google Colab - Stable Diffusion Art