[Stable Diffusion] Prompt Sharing and Learning Thread

Mimic22

Newbie
Jul 2, 2018
31
11
Understood, I am in search for this model :

Its name was AngleDroit - V1
 

kaamist

Newbie
May 22, 2023
71
31
Could any suggest a Positive prompt to use in SD 1.5 -A1111 , to get dark brown large areolae and thick sharp pointed round nipples ??!!
 

Sepheyer

Well-Known Member
Dec 21, 2020
1,610
3,844
Could any suggest a Positive prompt to use in SD 1.5 -A1111 , to get dark brown large areolae and thick sharp pointed round nipples ??!!
If the model you are using is not giving you that, changing the prompt will not help. You will need to add an appropriate LORA to your workflow to get those specific items out.
 

kaamist

Newbie
May 22, 2023
71
31
If the model you are using is not giving you that, changing the prompt will not help. You will need to add an appropriate LORA to your workflow to get those specific items out.
Like that adetailr of perky nipple?? this one
 
  • Like
Reactions: Sepheyer

osanaiko

Engaged Member
Modder
Jul 4, 2017
2,742
5,112
look on civtai for models with relevant tags. i downloaded some nipple related SD1.5 LORA from there previously.
 

kaamist

Newbie
May 22, 2023
71
31
look on civtai for models with relevant tags. i downloaded some nipple related SD1.5 LORA from there previously.


I hope this is the one,, but i read some in the commentthere is one more lora to detect where the boobs are too ?? Could any share that lora too
 

cesid

Member
Nov 2, 2019
196
77
I am new to AI, I want to ask some question:

2/Can AI build porn video scenes?
Videos are much difficult than Images. With A1111 or Reforge the videos are just many images that become gif and AI doesn't seems to remember anything and give consistency. It requires much time also. Tried both SVD and animatediff.
There are other video generator that can be run locally such cogxvideo, hunyuan or mochi but those don't support human interfaces like A1111 but they use machine language like ComfyUI. Robots and mentats said that those are very good.
 
  • Like
Reactions: Sir.Fred

kaamist

Newbie
May 22, 2023
71
31
i've 2 Questions, which i wish the forum will help me,,,

Currently I do try learn SD thro sagemaker,,, I plan to build my pc early this April once all the GPU's are released... What should be my choice of GPU , since i'm building with 9700X AMD CPU .. Many benchmarks (Peget or Tom's Hardware) refer NVIDIA will be the best,, But I've a budget constrain of around 600 USD only.. So give your suggestions ...

Second Is there any LoRas for Breast deduction and Adding Large Perky Nipples ??? If yes, do share the links if possible..
 

Sir.Fred

Active Member
Donor
Sep 20, 2021
910
4,402
nVidia is really the only choice, as you need good CUDA performance. Just get the card with the most VRAM you can afford. Don't know about SD Loras since I really only do Flux but you'll almost certainly find what you're looking for at civitai.com or tensor.art.
 

kaamist

Newbie
May 22, 2023
71
31
nVidia is really the only choice, as you need good CUDA performance. Just get the card with the most VRAM you can afford. Don't know about SD Loras since I really only do Flux but you'll almost certainly find what you're looking for at civitai.com or tensor.art.
the more Vram the better performance, I'm thinkin of going RTX 4070 Ti Super .. Since i dont have any idea on the RTX 50 Series and its pricing is hell high for my budget...
 

devilkkw

Member
Mar 17, 2021
334
1,132
the more Vram the better performance, I'm thinkin of going RTX 4070 Ti Super .. Since i dont have any idea on the RTX 50 Series and its pricing is hell high for my budget...
Seem series 50 have low price than series 40. Also i hope the 40 series get lower when series 50 is released.
 
  • Like
Reactions: kaamist

zues01

Newbie
Apr 19, 2018
51
115
Hello Hello. So, looking for some insight here. There is an AI Hentai site I used recently - Betterwaifu and one of the models really caught my eye. I'm a sucker for Tifa pool scene stuff and was really happy with the results but incredibly limiting. So ive been on the hunt trying to recreate the same results to allow more creativity.

Betterwaifu had the model listed as "DucHaiten Real v2.0". The only advanced settings available are DPM ++ 2M SDE Karras Sampler, 35 Steps, and CFG Scale 7.

The prompt I used was straightforward. "Tifa, gigantic breasts, naked. At pool. During the day. Tifa is wet, dripping, sweating. Hair wet."

Over the past few days Ive been downloading various DucHaiten Checkpoints, different Tifa Loras, etc. No real close luck yet so I figured I come here and see if any eagle eyes can spot what models/Loras I should be looking into to get similar results. Thank you for your time and help. Appreciate it.
 
  • Like
Reactions: Sepheyer

osanaiko

Engaged Member
Modder
Jul 4, 2017
2,742
5,112
The prompt I used was straightforward. "Tifa, gigantic breasts, naked. At pool. During the day. Tifa is wet, dripping, sweating. Hair wet."
If I understand correctly (and a quick google search backs me up) you must use commas or semicolons to separate the terms in a prompt. Other punctuation *does* have an effect on the output because make a difference on the text content as it is converted to tokens, but it doesn't separate the token stream into separate "conceptual terms". in the case of your example, the subphrase "naked. At pool. During the day. Tifa is wet" is treated an one single prompt term, and the tokens that it converts into would be very unlikely to have close match to the sort of labels that would be baked into the model.

The reason why you might still be getting a "good" output from the betterwaifu site vs doing it locally with theoretically the same model and params, is that their web interface might be converting the periods to commas without you knowing, before the prompt is fed into the tokenizer. That's just a theory however, it could be something completely different.
 

zues01

Newbie
Apr 19, 2018
51
115
If I understand correctly (and a quick google search backs me up) you must use commas or semicolons to separate the terms in a prompt. Other punctuation *does* have an effect on the output because make a difference on the text content as it is converted to tokens, but it doesn't separate the token stream into separate "conceptual terms". in the case of your example, the subphrase "naked. At pool. During the day. Tifa is wet" is treated an one single prompt term, and the tokens that it converts into would be very unlikely to have close match to the sort of labels that would be baked into the model.

The reason why you might still be getting a "good" output from the betterwaifu site vs doing it locally with theoretically the same model and params, is that their web interface might be converting the periods to commas without you knowing, before the prompt is fed into the tokenizer. That's just a theory however, it could be something completely different.
Appreciate the feedback. Thank you. When I have some free time, i'll give a better go at the prompts.
 
  • Like
Reactions: osanaiko