[Stable Diffusion] Prompt Sharing and Learning Thread

Mr-Fox

Well-Known Member
Jan 24, 2020
1,401
3,794
I just found out that you can use camera raw non destructively converting the image to a smart object (y)
Thank you for the tip. I like to watch onYoutube to learn more about Photoshop. I didn't think of making it "smart". TBH I have not used that function much and need to learn more about different clever ways we can "convert" the layers for our purpose.
 

fr34ky

Active Member
Oct 29, 2017
812
2,170
Thank you for the tip. I like to watch onYoutube to learn more about Photoshop. I didn't think of making it "smart". TBH I have not used that function much and need to learn more about different clever ways we can "convert" the layers for our purpose.
I didn't think of it either, somehow I touched something and the option to turn on and off the layer of camera raw filter was available, then I searched could I do a non destructable camera raw and found the answer. That youtuber is great, I also watch him when I want to learn something.
 
  • Like
Reactions: Mr-Fox

Mr-Fox

Well-Known Member
Jan 24, 2020
1,401
3,794
FML! Bruh, clarity is lit!
View attachment 2428106
Yes Clarity is indeed "Lit" :love: . I tried to make work with my project but it simply wouldn't.
Clarity is superior and generates beautiful images. Another one is that I want to explore a bit more. It has a few other "versions" as well.






They are all different flavors of merges with . An there is also a specific Lora.
 
Last edited:

Jimwalrus

Active Member
Sep 15, 2021
935
3,450
Just a quick test, taking Sepheyer's original but a completely opposite approach to the settings.

Same prompts, same seed, same model, same aspect ratio. Everything else is pretty much maxed out i.e. 150 steps vice 20; Hi-res x2, 30 steps with denoising at 0.7; CFG of 30, corrected back to mimic 7 with Dynamic Thresholding.
00026-470340181.png
Very different, not too sure I prefer it, but an interesting demo.

Taking a mid-position approach:
85 steps; Hi-res x1.5, 15 steps with denoising at 0.35; CFG of 13.5, corrected back to mimic 7 with Dynamic Thresholding.
00027-470340181.png

I'll get back to work now!
 

Mr-Fox

Well-Known Member
Jan 24, 2020
1,401
3,794
Just a quick test, taking Sepheyer's original but a completely opposite approach to the settings.

Same prompts, same seed, same model, same aspect ratio. Everything else is pretty much maxed out i.e. 150 steps vice 20; Hi-res x2, 30 steps with denoising at 0.7; CFG of 30, corrected back to mimic 7 with Dynamic Thresholding.
View attachment 2428844
Very different, not too sure I prefer it, but an interesting demo.

Taking a mid-position approach:
85 steps; Hi-res x1.5, 15 steps with denoising at 0.35; CFG of 13.5, corrected back to mimic 7 with Dynamic Thresholding.
View attachment 2428861

I'll get back to work now!
I really like the last one. A small tip, put "detailed teeth" in the prompt to improve them. I'm sure you already know but it might help someone.
 

Sepheyer

Well-Known Member
Dec 21, 2020
1,532
3,637
Just a quick test, taking Sepheyer's original but a completely opposite approach to the settings.

Same prompts, same seed, same model, same aspect ratio. Everything else is pretty much maxed out i.e. 150 steps vice 20; Hi-res x2, 30 steps with denoising at 0.7; CFG of 30, corrected back to mimic 7 with Dynamic Thresholding.
View attachment 2428844
Very different, not too sure I prefer it, but an interesting demo.

Taking a mid-position approach:
85 steps; Hi-res x1.5, 15 steps with denoising at 0.35; CFG of 13.5, corrected back to mimic 7 with Dynamic Thresholding.
View attachment 2428861

I'll get back to work now!
Thanks for testing! Earlier on I ran these tests to determine the sweetspot for the number of steps. And the results I got convinced me that 20 is really the sweetspot. The checkpoint and the sampling method had far larger impact then the steps.

Given that steps are increadibly expensive: 150 steps x 10 runs is the same as 20 steps x 75 runs, I trust having more candidate-renders is where it is at. Again, thanks for testing this out!
 

Sepheyer

Well-Known Member
Dec 21, 2020
1,532
3,637
Looks like a more realistic John Person (The Pit) bimbo milf. I would love to see her take on the gardener and repairmen etc or the principle at the school etc.. :D
Can't wait until we can do scenes. OMG, imma be needing a tonn of tissues for wiping my drool out.

00056-1320762134-(realistic photo of gorgeous Playboy playmate model from Norway).__(anfas, di...png
You don't have permission to view the spoiler content. Log in or register now.
 
Last edited:

fr34ky

Active Member
Oct 29, 2017
812
2,170
That's just the style of John Person. He makes them look a bit Square jawed.
View attachment 2428938

No traps here.. :LOL:
I swear, not only that vignette but that scene has to be one of the most glorious on porn comics ever. Basically that comic is what brought me to porn art.

Yes Clarity is indeed "Lit" :love: . I tried to make work with my project but it simply wouldn't.
Clarity is superior and generates beautiful images. Another one is that I want to explore a bit more. It has a few other "versions" as well.






They are all different flavors of merges with . An there is also a specific Lora.
Deliberate is great, love it a lot, funny that I created my own lora of deliberate, didn't know the lora even existed.


On a different topic, just a theory, I think that adding 'child' (and/or any synonymous) on negative prompt may 'open' the checkpoints to more revealing positions and dressing. The reasons are quite obvious. I've noticed that watching some prompts on civitai.
 

Mr-Fox

Well-Known Member
Jan 24, 2020
1,401
3,794
I took a leaf out of Jim's book..

Some testing.

With Hypernetwork : , weight 0.5
00040-1320762134.png

With weight 0.5
00041-1320762134.png

Adding "(bulging cameltoe:1.2)" to prompt. You can also try something like prominent pupis or prominent cameltoe etc..
00043-1320762134.png
 

fr34ky

Active Member
Oct 29, 2017
812
2,170
Can't wait until we can do scenes. OMG, imma be needing a tonn of tissues for wiping my drool out.

View attachment 2428926
You don't have permission to view the spoiler content. Log in or register now.
Man I swear that when some points start to connect, porn games are gonna jump 10 years in graphic quality and most of the games we see in this forum will look like pacman.