[Stable Diffusion] Prompt Sharing and Learning Thread

[devilkkw]

You have to experimenting and find what work better for model you are using.

This is an example of starting prompt without triggering any artist and negative i use for every generation

Spoiler: Prompt

You don't have permission to view the spoiler content. Log in or register now.

Waste the result, this is just a starting point, just try adding somethink like artist and see how it change, make sure artist is at the end of prompt, to not losing composition.

Spoiler: Prompt with Artist

You don't have permission to view the spoiler content. Log in or register now.

Spoiler: Result with some artist

You don't have permission to view the spoiler content. Log in or register now.

The artist name is in image name. You can mix artist, but remember to keep them at the end of prompt.

 

[Mr-Fox]

Thanks, I'm trying something like "bad hands, disfigured fingers, bad fingers, deformed, disfigured, fused fingers"
I don't know any artist for example, but I will try it.

Maybe krenz cushart, fkey, shal. e

It seems controlnet caused my weird fingers.

You must be registered to see the links

 

[daddyCzapo]

Tbh it turned out pretty good. It came with a watermark tho, i'll just photoshop it out Spent almost 30 minutes changing up the seed

Spoiler: Prompt

You don't have permission to view the spoiler content. Log in or register now.

 

[modine2021]

don't know what happend. was enjoying myself but now nothing happened but this message even after setting back to default values:


NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.
Time taken: 15.92s
Torch active/reserved: 3242/3676 MiB, Sys VRAM: 5237/8192 MiB (63.93%)

 

[sharlotte]

Modine2021, there's a long thread on git on this;

You must be registered to see the links

people tried different things to get rid of the error. I'll let you read and try to fix - good luck!

 

[modine2021]

Modine2021, there's a long thread on git on this;

You must be registered to see the links

people tried different things to get rid of the error. I'll let you read and try to fix - good luck!

oh i fixed it already after my other post ...i fixed it by editing the webui-user.bat file:

set PYTHON="C:\Users\SomeBody\AppData\Local\Programs\Python\Python310\Python.exe"
set GIT=
set VENV_DIR=
set COMMANDLINE_ARGS=--xformers --no-half-vae --disable-nan-check
git pull
call webui.bat
save all

 

[Sepheyer]

WTF Is the Latent Space?

Latent space is the storyboard. So, it goes like this. You have a script that you want to make into film.

You slice and dice the script into scenes and how you want them filmed. The document that captures this transition is a storyboard. Then, from that storyboard the film crew and the film machinery do the capture that eventually gets edited and produced into a film.

We just went: script -> [encode] -> sctoryboard -> [decode] -> [film]. IRL the encoder is the director and the decoder are the film crew and everything in the postproduction (plus beancounters, studio, etc.).

With stable diffusion the encoder are your samplers (i.e. Euler A, Heun, etc.) and the decoder are your VAE (yes, I know VAE is encode/decode, but fuck it).

Script (Prompt)​	Storyboard (Latent)​	Film (Clipspace/Pixelspace)​
​	​	​

The reason why we even have latent space is because it is much cheaper and faster to manipulate. The efforts to manipulate the latent space are one-millionth of what the pixelspace costs to manipulate. Exactly the same reason why we have storyboards.

That's how I map "latent space". Would you have a better example?

 

[Mr-Fox]

Story time with uncle seph.. I love it

 

[Mr-Fox]

Oh my god

Can you give us an update how it went with fixing the hands? I'm curious.

 

[modine2021]

View attachment 2487631

Spoiler: Prompt: CUI Meta

You don't have permission to view the spoiler content. Log in or register now.

i dont understand how to get the prompts form this image you used?. still finding my way around

 

[Sepheyer]

i dont understand how to get the prompts form this image you used?. still finding my way around

I use a different version of the interface, called Comfy UI:

You must be registered to see the links

And this is what it looks like once installed: link

 

[devilkkw]

After some try and test, i decided to ask you something.
I'm testing many prompt and many checkpoint i've made. Until i've discovered something strange and really great (i hope) on my models.
The trick is simple,and maybe someone know it or discovered it.
At the end of prompt just add ", Digital Compositing" so the prompt result like:
prompt, Digital Compositing
On my test seem these word add more consistence in prompt, and do better character/subject design.
Is possible is only my impression, so if you want try this trick and report if do difference or not is really appreciated.

You must be registered to see the links

 

[Nano999]

Any tips on what to type in promt or what Lora to use to prevent male characters magically turning girls after highres.fix

 

[fr34ky]

Any tips on what to type in promt or what Lora to use to prevent male characters magically turning girls after highres.fix

maybe 1girl 1boy in positive and 2 girls in negative, that if you are talking about 2 people pictures, you have to be more specific

 

[devilkkw]

What is the difference in results, can you show?

Is a result show of prompt, visually.
Actually i'm switching to " torch: 2.0.0+cu118 " and have to fix some library to make all functionally again, losing xformers but seem work great.

Finally made it work with torch: 2.0.0+cu118 and cudnn8.8

 

[miaouxtoo]

I've been testing a bunch of highres fix upscalers.
Looks like "Latent (Nearest)" for me is best for realistic skin & more detail in some areas for photo-type upscaling.
I've tried all the others, and there are obviously variations, but hard to notice.

My laptop craps out over 1176 x 1760 ish for latent nearest. Latent (Nearest Exact) doesn't work as well for me either.
There are realistic scratches on the left one's latex bra.



There are some issues with the boobs on the one on the right, but it's super interesting examining the differences.

My title for each of these is:
"Hey, don't forget about me..."

 

[modine2021]

I use a different version of the interface, called Comfy UI:

You must be registered to see the links

And this is what it looks like once installed: link

oh that looks like a shader creation in 3d programs..ugh...it tests my patience in 3d. ... thanks anyway

 

[sharlotte]

Experimenting with the various control net settings, using a photo as original and own created LORA for the face: nothing beats early morning yoga!

 

[Mr-Fox]

I've been testing a bunch of highres fix upscalers.
Looks like "Latent (Nearest)" for me is best for realistic skin & more detail in some areas for photo-type upscaling.
I've tried all the others, and there are obviously variations, but hard to notice.

My laptop craps out over 1176 x 1760 ish for latent nearest. Latent (Nearest Exact) doesn't work as well for me either.
There are realistic scratches on the left one's latex bra.

View attachment 2490492 View attachment 2490494

There are some issues with the boobs on the one on the right, but it's super interesting examining the differences.

My title for each of these is:
"Hey, don't forget about me..."

Don't forget to check out

You must be registered to see the links

.
I would recommend

You must be registered to see the links

and

You must be registered to see the links

.
For Ultrasharp, choose to download in zip form and then unpack in Stable-Diffusion\stable-diffusion-webui\models\ESRGAN.
NMKD should be put in this folder also.

 

[Mr-Fox]

My first posts using

You must be registered to see the links

by our own devilkkw.



You can see more in this post.