Anal Helper LoRA (Not a slider)

Also, what would an anal slider even look like? Near 0 strength the dick is just jammed against her perineum? 😂

@Loraholic I know you didn't mean to step on anyone's toes.

But... I don't really get rationale behind charging 7k for a model or making people wait 14 days. When I made my Pony models, I couldn't wait for people to use them.

Knowing how small your data set was here, I'm feeling inspired to train LoRAs locally again. I'm going to start looking for greedy buggers and try and beat their LoRAs

yeah this is way better. 7k for a lora, k3nk is delusional af. Ive trained tons of loras, thats way too much.
This works really well.

@jj43797771 He wants to be one of those elite users in the creator program that makes... let me check my notes here... <$70 per month median

@jj43797771 actually you know what, I'm wrong. 9 people have already downloaded it. 63k buzz in the bank.

@pogo looks like 9 idiots to me then haha

@jj43797771 I would say that's true if people are paying $ for buzz to buy loras

@pogo Early access is clearly labeled. My peeve is when people make 99% girl checkpoints (loras, too, but making black box checkpoints is worse) and don't bother to post even the tiniest comment that the training data is all-girl. Now that is a waste of someone's time. One person, who I won't name, even had the nerve to try to justify this by saying "I make these for myself. If anyone else likes them, that's fine." Subject change evasion. It takes a fraction of a second to put something like "all-girl" somewhere (comment, description, title) or a simple ratio or percentage (95% aF / 5% aM). aF = attractive female, aM = attractive male.

Civit needs to understand that people are busy and don't like to have their time wasted by downloading mystery content that turns out to be all-girl, when they're not looking for that. Imagine if it were the opposite: "NSFW Improved Realism Lora" = 99% male training data with female sample images and not a whiff of notice anywhere from the creator. There would be a riot, especially if many checkpoints were like that.

@ss9999 You mean a LoRA that has men in the samples but not in the training data? That sounds frustrating. But, I think in the case of Krea2 at least, most people are using multiple LoRAs at a time. It should be expected that you would at least need some kind of NSFW helper like SNOFs for hardcore images.

That's the case for me anyway.

@pogo The most common issue is for a checkpoint or lora to have almost only female sample images (plus maybe one or two token images with old men), which forces people like me to ask what the percentage of attractive males is in the training data. If Civit would simply have creators put that info in during the submission process then it would save the minority like me who aren't looking for all-female or 90% female 10% male content a lot of time. I think there should be two boxes: % attractive young females % attractive young males. Then, people could even search for loras and checkpoints and exclude results based on this. It's a lot better than having very little idea what a lora or checkpoint does and it's better than having people title them as if they're general-purpose when they're actually designed to produce female content specifically. If someone makes an "Improved Realism" checkpoint or lora and the training data is 85–100% female, why shouldn't people know that?

I applaud people taking the time to make improved female genitalia loras and such for the community but it's not content I want to spend time on. So, why not called the lora "Improved Female Realism" or put a note about the training data percentages in the description? This has been a long-standing problem. For instance, with SDXL checkpoints, there was BigAsp 2, which had male training content. Derivatives of that, like Big Lust, had so much female bias that I would have to put in like 40 negative prompt words to try to deal with it. It suggested that there was practically zero male training data but the creator never bothered to consider the time and energy of anyone not looking to generate females. If it were an isolated/rare thing I wouldn't complain but it has been going on for a long time. The title "Big Lust" does not tell anyone it's only for generating women.

@ss9999 wouldn't it be pretty clear if the "Improved Realism" LoRA was mostly women, if the sample images were all women? And if it showed men as well, you would see the LoRA fail to enhance the realism of the men to the same effectiveness as it does women.

It honestly isn't a problem I run in to. My recent LoRA used exclusively women and, with my workflow of multiple LoRAs, I have no problem generating men as well.

Add a male-focused LoRA to your stack and you're good to go bud

@pogo "wouldn't it be pretty clear if the 'Improved Realism' LoRA was mostly women, if the sample images were all women?"

No. All it really shows is that the creator is either catering to the larger portion of the audience and/or prefers to provide female sample images. I have seen the same creator, for instance, post nothing but females for some versions of the same checkpoint. Look at the same checkpoint for a different model and there are young males in one or more sample images. A particularly annoying practice is to post fat/old/homely males along with a parade of young gorgeous females. This doesn't tell the user if the male training data, if there was any, involves any young attractive males.

Guesswork shouldn't enter into things. Creators should simply let people know what percentage of attractive females are in the training data, versus attractive males. It's not much to ask. It takes a few seconds at most to post two numbers. Civit should have this be part of the submission process. Or, they could call it Improved Female Realism? Why is that even controversial? All it does is acknowledge that generating females isn't the thing that everyone in the world wants to do, which is a reduction of chauvinism.

"It honestly isn't a problem I run in to. My recent LoRA used exclusively women and, with my workflow of multiple LoRAs, I have no problem generating men as well."

It's pointless to download a large amount of content if the training data for that content is exclusively or overwhelmingly female if the goal isn't to generate females. Being able to produce some male images from that content doesn't make that better. A person is still wasting their time and storage, plus they'll have to fight model bias.

@ss9999 Do you see my recent image posts? They all use my Semi-Pro LoRA that was exclusively trained on women. I can still get dicks and dudes.

Honestly, I don't think you're using all the rational abilities available to you.

1)You see mostly or all women in the samples: probably all women in the dataset.
and
2)Everyone using Krea is using multiple LoRAs enabling different content. Use snofs, use realism engine, you will be able to generate dudes no matter what else you use.

I can only give you the keys, bro. I can't make you drive.

@pogo You're arguing that it makes sense for checkpoints and loras to be black box mystery content that may have nothing to offer the person who is spending time and energy downloading them, testing them, and storing them — all so that some people can pretend that everyone wants to generate female content in order to get an ego boost.

What I am arguing is that people be able to know what's in the tin — not every tiny detail but the important basics, like if the training data for custom content is exclusively or overwhelmingly female. No guesswork, no evasive smug BS like "I make these checkpoints for myself and if anyone else likes them that's fine."

This entire conversation happens because people were complaining about one lora creator using early access, something that's clearly labeled. That something that is extremely clearly labeled like early access is a problem but having things be extremely unclear (and not just a few things but huge numbers of them over many years) isn't seems like mistaken prioritization. But, since I'm not looking to generate 1girl content, who cares if my time and storage is wasted, eh?

@ss9999 "You're arguing that it makes sense for checkpoints and loras to be black box mystery content"

No, I am not. I'm arguing that there are clear signs that will show you what the content is, and that you're ignoring them.

"so that some people can pretend that everyone wants to generate female content in order to get an ego boost."

That's schizophrenic. You're doing an insane amount of reading in to anonymous strangers intentions, when you can spend that effort reading in to the sample images and drawing reasonable conclusions about the dataset.

"something that is extremely clearly labeled like early access is a problem but having things be extremely unclear"

I'm the one that started the thread. I brought up K3NK in reference to his "Only-Anal" model. And to be clear, it is obvious to anyone who looks at the sample images what the LoRA is all about. You can even see watermarks in his sample images. What does that tell you? It tells you that the dataset included pictures with watermarks.

@ss9999 And I'll say it again. There is no LoRA on this site that will stop you from making hetero content. Every person using KREA2 to make hetero images is using multiple LoRAs, ones that enable and enhance hetero images.

Go look at my images for Pogo Semi-Pro and tell me my LoRA can't be used to make images with dudes and dicks.

@pogo "I'm arguing that there are clear signs that will show you what the content is, and that you're ignoring them." Enjoy being wrong, then.

@ss9999 you're an actual moron.

it works

@Loraholic Awesome, can't wait! Not all heroes wear capes – some make darn good LoRAs 😁 Thank you so much!

@Loraholic 
Ok. Thanks for the very sarcastic message.
The face doesn't change with realism engine or snofs as they mask the face data out during the training.

@Loraholic And which has a male being penetrated. : )

... I don't understand why you'd want to do that...

Maybe. But, probably not. You could use a negative value, but the OPPOSITE of anal is not vaginal. It's no penetration at all. And stronger, no dick, and stronger no nudity. You would need to train to a NoAnal lora... And again, I just don't understand that kind of thinking.

I envy you. Do you want some vaginal sex? I have plenty of that. But in exchange, give me some anal)

@Loraholic I don't think it always works that way. For example, yesterday I tried to insert a dildo into anus. I had to correct the prompt multiple times! But I wasn't getting anal penetration, and I also wasn't getting a visible anus! The girl was facing the camera. Although in the prompt, I described anal penetration several times and in different interpretations. But I hope my ordeal is now over. After all, I'm here to download this Lora)

@Loraholic Your advice makes sense. While 6GB models need to write the same action or item 50 times, this is strictly forbidden with Krea. I'll try your theory in various situations. I've written countless prompts, but even today I'm losing track of some details. Thanks for the advice!

@Loraholic it seems to be very common with most of the general-purpose Krea2 NSFW LoRAs that they tend strongly toward anal penetration when attempting prompts of different styles for vaginal whether it's with penis, fingers, dildo or anything else. Maybe it's a preference of the LoRA creators using lots of anal training material, or some bias of the base model itself.

Do you suppose you could try to create a LoRA which makes it more probable for the insertion to be vaginal? I haven't found any reliable way otherwise and I see there are a few other comments asking for this kind of a possibility.

@Loraholic ok it's your choice. Just saying, there's like a dozen anal loras by now and probably about the same for oral (and dozens of different penises, it seems), but very few for those of us who prefer the female genital except a couple of very specific poses. Oh well, maybe it's a matter of trying to narrow down an optimal combination of models and loras with what we have, unless someone makes one that works properly for vaginal.

@critissues Lora's lack of vaginal sex is due solely to the fact that 99.9% of people can create such a photo. If that were a problem, Lora would have been created immediately.

@Loraholic You’re such a LoRA-holic! 😂 It feels like your days have 30 hours, while I'm stuck with only 24.

It looks like my problems were partially caused by stacking certain Loras, maybe trained mostly with solo images, which were affecting the appearance of the vulva. They probably made the model move away any objects that would obstruct the view of it.

>wants a lora to help specifically with vaginal penetration
>clicks on a lora called anal helper

ok