[Stable Diffusion] Prompt Sharing and Learning Thread

Mr-Fox

Well-Known Member
Jan 24, 2020
1,401
3,802
Tried making a LORA of Bo-Katan but for some reason SD thinks she's really old lol any way to fix it?

You don't have permission to view the spoiler content. Log in or register now.
Try to specify the age in positive (18, 20, 25 years old) and in negative add "wrinkles", "old" etc.
You can even add "young woman", followed by specified age.
Like so:
Photography, young woman, 20 years old, bo-katan, full body view, etc.
 

Hunter9836

New Member
Apr 27, 2023
4
2
Hey, what's up guys? I'm seriously hooked on this checkpoint and don't even wanna try anything else. I've experimented with a bunch of other models and even mixed some up, but still can't replicate this one on my own. Would really appreciate any help I can get. P.S: [1] Maybe Model NED is in the mix , and hipoly3Dmodel lora used [2]I've already gone through all the info available on this forum, but still struggling to nail this style View attachment 2576525 1682584101109.png 1682584532036.png 1682584688026.png 1682584767572.png 338975387_552047427052158_1194769038138438799_n.jpg
 
Last edited:
  • Like
Reactions: Pogo123

Mr-Fox

Well-Known Member
Jan 24, 2020
1,401
3,802
Can you please tell me the name of mix or the ckpt used???
Use a checkpoint aimed for semirealism like deliberate, experience etc. Specify the style in your prompt. Keyword "fantasy".
Write in the positive what you wish to see in your image and write in the negative what you don't wish to see in your image.
Look around on civitai for user sample images in the style that you want and you will find a prompt that you can start out with.
 

Hunter9836

New Member
Apr 27, 2023
4
2
Use a checkpoint aimed for semirealism like deliberate, experience etc. Specify the style in your prompt. Keyword "fantasy".
Write in the positive what you wish to see in your image and write in the negative what you don't wish to see in your image.
Look around on civitai for user sample images in the style that you want and you will find a prompt that you can start out with.
I'm comfortable with the prompts, loras and embeds... But struggling with the Base Model itself... Can't seem to nail the image to perfection like the images I posted. FYI I've tried NED, REV, chilloutmix, Perfect World etc.
 

Mr-Fox

Well-Known Member
Jan 24, 2020
1,401
3,802
I'm comfortable with the prompts, loras and embeds... But struggling with the Base Model itself... Can't seem to nail the image to perfection like the images I posted. FYI I've tried NED, REV, chilloutmix, Perfect World etc.
What is it specifically you don't achieve? Where did you find those images? They look like they could come from civitai and if so you should know the checkpoint and the prompt.
 

Hunter9836

New Member
Apr 27, 2023
4
2
What is it specifically you don't achieve? Where did you find those images? They look like they could come from civitai and if so you should know the checkpoint and the prompt.
Nah bro its from Social Media, I'm trying to recreate the whole aesthetics of the images I posted, to perfection. which I am not able to achieve 100% on my own.
 
  • Like
Reactions: Mr-Fox

Nano999

Member
Jun 4, 2022
166
73
Recenetly I just tested over 200 checkpoints xD

btw a quick question. Does it make a difference if you put you <lora:NAME:0.6> right after trigger words in the middle of promt or in the end?

e.g. (trigger word is trigger1)

milf, outdors, trigger1, <lora:NAME:0.6>, looking at the viewer, grin, beautiful landscape, Egypt architecture

OR

milf, outdors, trigger1, looking at the viewer, grin, beautiful landscape, Egypt architecture
<lora:NAME:0.6>


I am doing it in the following way:

milf, looking at the viewer, grin,
beautiful landscape, Egypt architecture
outdors,
trigger1,
<lora:NAME#1:0.6>
<lora:NAME#2:0.6>
<lora:NAME#3:0.6>
<lora:NAME#4:0.6>


I see no difference, but maybe it might go wrong at some point?
 

rayminator

Engaged Member
Respected User
Sep 26, 2018
3,126
3,187
Recenetly I just tested over 200 checkpoints xD

btw a quick question. Does it make a difference if you put you <lora:NAME:0.6> right after trigger words in the middle of promt or in the end?

e.g. (trigger word is trigger1)

milf, outdors, trigger1, <lora:NAME:0.6>, looking at the viewer, grin, beautiful landscape, Egypt architecture

OR

milf, outdors, trigger1, looking at the viewer, grin, beautiful landscape, Egypt architecture
<lora:NAME:0.6>


I am doing it in the following way:

milf, looking at the viewer, grin,
beautiful landscape, Egypt architecture
outdors,
trigger1,
<lora:NAME#1:0.6>
<lora:NAME#2:0.6>
<lora:NAME#3:0.6>
<lora:NAME#4:0.6>


I see no difference, but maybe it might go wrong at some point?
use this Additional Networks extension
https://f95zone.to/threads/stable-diffusion-prompt-sharing-and-learning-thread.146036/post-10255639

then you just use <NAME> just need to remove the lora: and :0.7 or no tag depending on what you are trying to create
 
  • Like
Reactions: modine2021

Mr-Fox

Well-Known Member
Jan 24, 2020
1,401
3,802
use this Additional Networks extension
https://f95zone.to/threads/stable-diffusion-prompt-sharing-and-learning-thread.146036/post-10255639

then you just use <NAME> just need to remove the lora: and :0.7 or no tag depending on what you are trying to create
It works both ways. The Additional Network extension is just another way of using lora's. It is limited to 5, but if you can't get the result that you want with 5 or less then you are probably doing something wrong.. The only benefit with it is that you can separate UNet/Text Encoder weights and fine adjust both. It's a bit overkill for most since generating images with ai isn't that consistent and predictable. But if you have the need to tickle your OCD go for it.
 
  • Haha
Reactions: sharlotte

Mr-Fox

Well-Known Member
Jan 24, 2020
1,401
3,802
Nah bro its from Social Media, I'm trying to recreate the whole aesthetics of the images I posted, to perfection. which I am not able to achieve 100% on my own.
Well look through civitai, I have seen many similar types of images there. The prompt is what will make it happen for you and the checkpoint will facilitate it. Some checkpoints will be easier to work with than others but it will not be the thing that makes an image possible or not possible, only easier and relatively more consistent. Then of course lora's and similar can of course make it easier as well, as long as it has been trained by someone who has a clue. Don't assume that it was the latest version of the checkpoint that was being used in those images and it's not for sure that it was the same checkpoint in all images.

I just discovered , it looks like it can be a good candidate, at the very least worth a try. I have seen highpoly lora being use in combination with , it's worth a try also.
 
Last edited:

Nano999

Member
Jun 4, 2022
166
73
use this Additional Networks extension
https://f95zone.to/threads/stable-diffusion-prompt-sharing-and-learning-thread.146036/post-10255639

then you just use <NAME> just need to remove the lora: and :0.7 or no tag depending on what you are trying to create
Additional networks are buggy and not working with some loras, I do not recommend to use it and they are also extending the promting time writing, as all promts I have they are pre-written in .doc

My original question was ignored
 

Mr-Fox

Well-Known Member
Jan 24, 2020
1,401
3,802
Additional networks are buggy and not working with some loras, I do not recommend to use it and they are also extending the promting time writing, as all promts I have they are pre-written in .doc

My original question was ignored
No one is ignoring your question, it's just that no one knows either way for sure. The only way is to test it for yourself and you did and couldn't see any difference. Essentially I think you answered your own question. It would be my guess also that it doesn't matter where you put it in the prompt since it's not a normal prompt token.