Can't you use add difference in checkpoint merger to remove pornmaster from it?

@iamddtla you`re welcome.

Agreed, I absolutely love this checkpoint. my only complaint is that it's hard to get it to generate older women. It focuses on younger faces too much, aside from that, easily one of my favorites.

@SomebodyOnceToldMeh just put some weights to your prompts and add "::2" to every age prompt. never had any issue with that

@stapfschuh I may need to try that and see if it helps.

@iamddtla allright

In other words, in IMG2IMG tab (if you use A1111, stability AI or other 'stable diffusion based' app), there is an INPAINT tab. There, you can fix hands, head, add gens, etc. For that, it's better to exchange the 'normal' Pornmaster to 'inpaint' pornmaster model.

装个MultiDiffusion 放大器插件,4G也可以,就是慢

我最近试了试FULL，1066显卡，只比2gb的版本慢两三成左右。

@iamddtla 谢谢回复~

which version you tried?

@saman_a464 Thank you so much <3

@PM786  v3 3.7gb It is enough and the quality is good

we dont need real girls.

@iamddtla please explain in more detail since I'm a noob?

What is your workflow?

@Incognito15647 workflow? you mean for what i use this checkpoint? i'm a prompt nerd and create all kinds of prompts. and therefore i need 100% working checkpoints and nothing half baked.

@themancalledmojo999 Well, do you use ComfyUI?

@Incognito15647 nope. a1111 webui

You got a universal negative prompt that you could show us? I've picked up a mishmash of random ones from other people but considering you're a prompt nerd and are using this model, I might as well ask

@BobTheBurner i might have, but for what should it be? anatomy? physique? quality? i already wrote a bunch of articles for anatomy prompts. if you want to know more, search for "stapfschuh" in articles and try these if you want. but other than those articles, i have an insane amount of other prompts to use for anything.

@themancalledmojo999 I suppose instead of asking for specific, I've been wanting to ask. For making prompts, is there an order of what you jot down, like doing clothes then position then location?

And a second question, besides character loras, what resources did you tack onto this model during your use of it? I saw the author mention like age slider among other things for example.

@BobTheBurner order? me? never! xD

my head is always full of ideas. and i have a huge list where i write those ideas down. there is never any order on this list. but if you had something else in mind about this question, look at this article Weights, but not the way you know them | Civitai

and your second question:

i did not make this checkpoint here. and i never made even any checkpoint or lora.

and if you want to know an age slider prompt, use "grades". like "kiddy-grade", "teenie-grade", "adult-grade" or "senior-grade" on any model.

@themancalledmojo999 Ahh, forgive me for not being clear. I meant that when you downloaded and used the full model, did you download anything else to use with it or you just used the full model alone?

@BobTheBurner the full model alone. to test prompts, it's important to always use any checkpoint "naked".

@themancalledmojo999 I'll trust your judgement on the model, oh. I did have another question. I've seen the Pony model based on sdxl and wanted to know if you've had any experience with it cause I wanted to know how that and this model compares

@BobTheBurner sorry, but i don't use any pony model, because it's a pile of shit when it comes to prompts. it's simply not made for real free experience at this point. xl was the same for many months until i found 2 insanely good working checkpoints that could handle any prompt which made me super happy.

@themancalledmojo999 See, this is why I always ask before burning time experimenting by myself with new shit. And by the way, which two checkpoints are those?

@BobTheBurner JadeEpicPortraitWorldXL_V1.1 - v1.0 | Stable Diffusion Checkpoint | Civitai

AnimatrixXL - v1.0 | Stable Diffusion Checkpoint | Civitai

@themancalledmojo999 I will assume that if I'm just pursuing horny generation, I should go with this one after seeing the general praise and well, it's got pornmaster in the name, lmao. Currently finding out how to mess with google colab, and I think the paid gpus can definitely play with the full 10gb version.

Don't know if you'd know, but while this model uses danbooru tags, would adding in artist tags influence the art style at all? Or would that need to be a specific resource you'll have to find and download

Thanks for putting up with me btw

@BobTheBurner you can use artstyles. for example simply "ghibli-style". but you get nearly no good results if you use any detail prompts. it gets mixed too much and characters have too low similarities to those styles. just experiment with artists names and compare them with results on danbooru

@BobTheBurner some good advice that i found out right now: don't use brackets and numeric weights at the same time and in combination. white or strange looking images get generated 

@Stapfschuh02  I'll make a note to experiment a bit, and I'll also note down about seeing if an order of tokens(whatever the prompt keywords are called specifically) affects generation. I'm trying to plan ahead as much as I can because I'm gonna make my money's worth off of google

So if I'm doing (thing:1.2), if I want to do something like brackets, I should just do the same with a lower weight like .8 instead of brackets.

@BobTheBurner exactly. i found this out on a pretty "rude" way. i switched checkpoints without thinking about those brackets and weights i used and just wanted one image. i was wondering why i suddenly got these strange results on this checkpoint that never happened even once before. after i looked at my prompt box i saw the problem. that's why i wanted to inform you about this.

@Stapfschuh02 Thanks very mush, so, the more nitty gritty questions, what sampler/schedule did you use with this pornmaster model during your testing? Been using DPM++ 2M Karras at 30 steps, cfg 7 just now

You bother with clip skip at all? I just left it at default. I skipped hiresfix and am doing the images at 1024 by 1024 as it's large enough for my tastes unless it's better to do 512 by 512 and use hires? Also, you use refiner at all?

I actually am curious if the shape of the image affects quality depending on what we're trying to generate?

@BobTheBurner i always test my prompts and everything else with default settings. but my standard resolution on images is 512x1024. love to play with higher resolutions on sd and lower resolutions on sd xl. for tests, to see what i get.

I have a 3080 with 12 GB, I don't think I need it.

Fixed now!

@theally it works now, thank you! Does this mean that other Illustrious models will also become available in the generator again?

Looks like PONY standard VAE works fine https://civitai.com/models/660673?modelVersionId=785437

Or SDXL VAE https://civitai.com/models/296576?modelVersionId=333245

The images look 100% the same with any of them

using il v1

Fixed it one of my prompts was causing it.

（ENSD：31337）可以算是上古级别的东西了，不必理会，不会影响生图的质量

非要的话，只能说ComyfUI没有这个东西，硬要复现webUI的图的话可以用Inspire里的KSampler，具体自己研究。

@openmn793 感谢大佬解答。

Just use huggingface or liblib/tensor (but just allow DL)