- Aug 4, 2020
- 2,402
- 3,088
- Oct 29, 2017
- 812
- 2,170
There is a filter and a category bar under the search box.
You still have to bang your head a lot to get results even with the checkpoints that are supposed to do the stuff they say, that's why you rarely will see a good blowjob picture for example.
Some styles are quite easy to get with any checkpoint, and some checkpoints are a little too hostile for porn, so at some point you have to compromise and see what thing is harder to get, usually the porn part is harder to get in a stylistic checkpoint than a style in a porn checkpoint.
- Sep 15, 2021
- 930
- 3,423
I think the problem with generating porn is that it requires two (or more) humans to be in actual contact. This is something that SD inherently struggles with as it has no concept of how human bodies interact.
It's seen some pictures of that sort of thing, but it tends to get itself really confused and merges people into one another.
It's looking more and more like the best way to go is having trained models (LoRAs probably) for each position and / or type of intercourse and using those in addition to a model that handles nudity well.
- Jan 24, 2020
- 1,401
- 3,794
There are blowjob specific checkpoints
You must be registered to see the links
You must be registered to see the links
- Jan 24, 2020
- 1,401
- 3,794
- Jan 24, 2020
- 1,401
- 3,794
check out the scripts. I just learned about it by Naytahn in the URPM discord. You can use it to generate a grid with images comparing different parameters that you set ex prompt or step or CFG etc.
Prompt matrix
X/Y/Z plot (steps)
X/Y/Z plot (Sampler)
- Oct 29, 2017
- 812
- 2,170
I tried them already, the interracial is decent, the other so so. I even tried one paizuri model a couple days ago and it was a total pain to use, not easy to get any results at all.
Most of the times the guys from the comments just copy and paste the prompts from the checkpoint creator and they call it a day.
I personally don't like to do that because it doesn't add anything, if I can't make my own ideas work with the checkpoint/lora, then it's not very useful/versatile.
Just saying that don't think you won't have to bang your head, making them work and making only 1 picture look good is not as easy as it seems, specially with that kind of poses.
When I said 'porn' I just meant erotic, nude hot poses, other than just standing pretty girls. Putting a sex picture together that looks alive is another monster, we will get there for sure, but it's really hard.
- Oct 29, 2017
- 812
- 2,170
I use the scripts for generating that kind of grids, even tried the x/y/z scripts, they are good when you already have a concept and want to get the best looking picture for a final release.
What I meant is, sometimes I'm trying 2 different girls to see which one looks better using the same prompts. Or sometimes I'm using a girl and trying 2 different poses or body types at the same time, going from one to the other, the only way I could find is using 2 different tabs...
It's like going through 2 different branches...
I have a GPU with 3Gb of VRAM and I can generate images without problems (as long as I keep the resolution at 512x512 or its equivalent aspect ratio) The key is in the parameters of the 'webui.user.bat' file. Right-click and select 'Edit'. The content of the file will appear in the Windows notepad. Look for the line of code 'set COMMANDLINE_ARGS=' and after the equals sign write the following: "--medvram --opt-split-attention" without "". Save the file and run the .bat again. It works like this for me and I haven't had any problems, unless I want to raise the resolution to more than 512x512. I hope it helps you if you still haven't solved it.
- Dec 21, 2020
- 1,531
- 3,618
Who can find a black woman close to this prompt:
Strangely I have such slim pickings, I can't even. And the end result is no where near what the prompt is suggesting. I.e. from time to time there are pretty ones, but nobody close to the prompt.
Strangely I have such slim pickings, I can't even. And the end result is no where near what the prompt is suggesting. I.e. from time to time there are pretty ones, but nobody close to the prompt.
You don't have permission to view the spoiler content.
Log in or register now.
You don't have permission to view the spoiler content.
Log in or register now.
- Oct 29, 2017
- 812
- 2,170
But she has like 8 fingers in each hand xD I'd rather bang your 5 finger girls.
Try this checkpoint for black women:
You must be registered to see the links
I use it a lot for any race but some guy posted some awesome black girls with that model a while a go.
- Jan 24, 2020
- 1,401
- 3,794
- Oct 29, 2017
- 812
- 2,170
I just tried that blowjob pose embedding and it sucks bad even using all of my tricks. I'm shocking nobody trained more bj pose embeddings, it's either impossible or people just don't like blowjobs (?)
Edit: it sucks but not in the good way
Just adding more stuff to the topic:
While looking for more BJ checkpoints reading the comments someone asks how was the embedding trained, and the guy answers:
'The other was something I just tuned of another model by hand (moving sliders in an embedding editor). This was trained from scratch using images I collected, cropped, and resampled.'
So, 2 conclusions, 1) there is an embedding tuner, which would be cool to try, and 2) people just upload whatever shit they make into that page with very low quality testing, so best thing to do is download the most popular high quality checkpoints and don't lose much time on this kind of low quality stuff.
Edit: it sucks but not in the good way
Just adding more stuff to the topic:
While looking for more BJ checkpoints reading the comments someone asks how was the embedding trained, and the guy answers:
'The other was something I just tuned of another model by hand (moving sliders in an embedding editor). This was trained from scratch using images I collected, cropped, and resampled.'
So, 2 conclusions, 1) there is an embedding tuner, which would be cool to try, and 2) people just upload whatever shit they make into that page with very low quality testing, so best thing to do is download the most popular high quality checkpoints and don't lose much time on this kind of low quality stuff.
Last edited:
- Jan 24, 2020
- 1,401
- 3,794
I did some testing.That prompt looks a bit suspicious in relation to the resulting image. You can fake that stuff by putting it in img2img and change the prompt text and then generate with 0 denoising. It's possible that he just got very lucky or did something else that doesn't show up in the meta data. Or Idk..
Did some rounds with tweaking in between.
"beautiful black young woman from Africa with (beautiful face), (((enormous breasts))), plump figure, big hips, small waist, thick thighs, big afro hair, large hoop earrings.
posing for a picture on the street smiling hands in the air.
she wears (((tight Yellow long-sleeved shirt))) and (((tight black jeans))).
the location is (street, evening, depth of fields). digital painting, hyperrealism"
The result:
There are of course more tweaking and improvements to be made. I suspect there is a lora needed such as hyperBottomHeavy or similar
Last edited:
- Oct 29, 2017
- 812
- 2,170
May be the guy just took some random pic and put it in img2img with some prompts, that will never give you the original picture using txt2txt from zero.
- Dec 21, 2020
- 1,531
- 3,618
"Tense" and "alert":
You don't have permission to view the spoiler content.
Log in or register now.
You don't have permission to view the spoiler content.
Log in or register now.
- Jan 24, 2020
- 1,401
- 3,794
- Oct 29, 2017
- 812
- 2,170
I didn't understand the post completely eitherI thought the style looked familiar..Derrrp.. completely misread your post..
- Dec 21, 2020
- 1,531
- 3,618
