[Stable Diffusion] Prompt Sharing and Learning Thread

[Jimwalrus]

I've just seen a video from Sebastian Kamph that says that it's not always good to inpaint after the upscale, funny because I learned to do that from him before, he doesn't say when it's good and when it's not good to do it. There is a lot of experimentation going on it seems.

I've never been at the cutting edge of anything before, it's fun!

 

[sharlotte]

Yeah AI stuff is really starting to get better and better. With my GTX1070ti 8GB VRAM that's a few years old, I'm able to do stuff but can see the limitations. After 1 hour of doing generation, using inpaint to change faces (

You must be registered to see the links

by AI entrepreneur) , I tend to run out of memory (CUDA out of memory), works fine when I restart the gui but somewhat frustrating.

I did generate my own lora, from 148 images of a well known pornstar which you should recognize below. Intensive bit of work but satisfying to get some results.

Anyways, here are some of my recent stuff, just doing varied prompts and girls to see how things work:

 

[Divine Cyberman]

I've just seen a video from Sebastian Kamph that says that it's not always good to inpaint after the upscale, funny because I learned to do that from him before, he doesn't say when it's good and when it's not good to do it. There is a lot of experimentation going on it seems.

Yeah, there is still a lot to figure out. From my experience so far, inpainting an upscaled image is most useful for improving the blurry/deformed look on faces but I'm still not sure if it's really useful when fixing other problem areas like hands. Also, I think the usefulness of upscaling before inpainting also heavily depends on the composition of your image. For example, a full body, group, or distant shot image should in theory benefit more from the "upscaling first" workflow than a close up portrait.

 

[Sepheyer]

Elegance model:

You must be registered to see the links

Spoiler: More

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

 

[Sepheyer]

Musings: looks like with AI as a creative tool the role of a human creator changes to "curation". And then the better form of curation becomes creating one's own models that conforms to the view of the curator/creator. I.e. if I like "playboy style" and I seek models that capture that look, then only natural that it becomes a necessity for me to create my own models that have: right characters, right backgrounds, right poses, right lightning.

 

[Mr-Fox]

Musings: looks like with AI as a creative tool the role of a human creator changes to "curation". And then the better form of curation becomes creating one's own models that conforms to the view of the curator/creator. I.e. if I like "playboy style" and I seek models that capture that look, then only natural that it becomes a necessity for me to create my own models that have: right characters, right backgrounds, right poses, right lightning.

There are many checkpoints with the magazine cover look or playboy look. A different option to make your own from scratch is merging, i.e. curation of the checkpoints that suits your purpose the best.

You must be registered to see the links

You must be registered to see the links

 

[Sepheyer]

Spoiler: Prompts

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

 

[Mr-Fox]

You must be registered to see the links

Sweaty muscular cheerleader in sports arena. Prompt by AIArtjak at civitai experience comment section, slightly edited by me and adding some lora's etc.

 

[Mr-Fox]

You must be registered to see the links

You must be registered to see the links

Apparently txt to audio is now also a thing. I have tried it out and it's a cool concept but as a musician I'm not terribly impressed with the resulting audio files so far.

Prompt:neoclassical metal electric guitar intro in the style of guitarist Jeff Loomis, followed by guitar riff section in the style of guitarist Tosin Abasi, 2 minutes long.

I was probably a bit optimistic here...

00133-2999632165.wav
00134-2999632165.wav
00135-2999632165.wav
2023-02-21_21-13-15_joined.wav

Phew.. all can relax.. no artists will lose their career... lmao

 

[deltami]

anyone find that Controlnet is good but not that good? especially on NSFW focused model like urpm/ realistic vision, mostly the legs are more twisted and arms are more fused than direct text2img.

, or just my setting is not well toned enough？

all experience share welcome！

 

[Mr-Fox]

Sometimes it's nothing to do with the extension or lora or etc. Sometimes it's the resolution, sometimes it's just the nature of the beast so to speak. I have not tried messing with the controlnet yet. My experience with SD so far though is that it's like trying to describe the world to a blind person that will then show you their interpretation of the world that you describe. Obviously a blind person doesn't know how the world is supposed to look like and just does their best to understand your description. Use a lot of negative prompts and weight them also. Lets see if it makes an improvement. If the prompt doesn't give the desired result, then start to go through all the settings and try to find anything that makes an effective difference.

 

[Sepheyer]

I am really feeling this elegance model.

 

[Mr-Fox]

I am really feeling this elegance model.

View attachment 2409231

Yes I have tried it also and I agree, it's great. Awsome image. The stuff dreams are made of.

 

[Mr-Fox]

Btw have you tried the

You must be registered to see the links

for the pokies? I think it would fit this image.



And there are also lora's for bigger everything also.. It's like ordering food, you'd want "extra eerrryting"..

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

 

[Sepheyer]

This guy on deviant asked for a pregnant milf. Not that I am into this, but boy, do I appreciate a lucky pull from the static:

 

[Sepheyer]

Btw have you tried the

You must be registered to see the links

for the pokies? I think it would fit this image.

View attachment 2409305

And there are also lora's for bigger everything also.. It's like ordering food, you'd want "extra eerrryting"..

You must be registered to see the links

View attachment 2409310

You must be registered to see the links

View attachment 2409324

You must be registered to see the links

View attachment 2409331

You must be registered to see the links

View attachment 2409336

You must be registered to see the links

View attachment 2409337

OMG, FML! Finally a motivation to figure LORAs out. Thank you bro.

 

[Mr-Fox]

OMG, FML! Finally a motivation to figure LORAs out. Thank you bro.

Don't forget to sift through my post about this:
https://f95zone.to/threads/stable-diffusion-prompt-sharing-and-learning-thread.146036/post-10093830

 

[Mr-Fox]

Ah shiiz.. Forgot this one.

You must be registered to see the links

All lora's pages has some good tips and info along with good prompts to try out so visit them all. Not only to download.

 

[Mr-Fox]

A short while ago fr34ky talked about not using the restore face function. I just learned about VAE.
Variational autoencoders (VAEs).

Why do we need VAE?

Variational autoencoders (VAEs) are a deep learning technique for learning latent representations. They have also been used to draw images, achieve state-of-the-art results in semi-supervised learning, as well as interpolate between sentences.

SD 1.5 comes with a VAE but there is a new improved verison.

What does it do?

It gives better and sharper details such as eyes and face, faster.

Get it here:

You must be registered to see the links

press download.. Place it inside "Stable-Diffusion\stable-diffusion-webui\models\VAE".

Go to settings, Stabble Diffusion.
Press Reload SD VAE.

And choose "vae-ft-mse-840000-ema-pruned.safetensors".
Also "Ignore Selected VAE for checkpoints that has their own" seems like a good idea, it was checked by default but if it isn't make sure it is.
Press "Apply settings".

So does this mean we don't need to use "Face Restoration" anymore? Let's try it and see. Try with and without, also try the different types and with different settings.
In settings, "Face Restoration" you can choose either to use "codeformer" or "GFPGAN" and how much it is being used.
They give a slightly different result, so try both.
" CodeFormer weight parameter; 0 = maximum effect; 1 = minimum effect " so to decrease the effect increase the number.
Then there is postprocessing as an option, here codeformer is also being used. Yeah I know a bit confusing because the same name.
In settings , "Postprocessing" you can "Enable postprocessing operations in txt2img and img2img tabs ".
You can choose between again "codeformer" or "GFPGAN". Or both..
It will show up in the UI for txt2img and img2img.
I'm using codeformer at the moment with Visibility:1 and Weight:0,4 while testing.
I'm not an expert, only learning as I move along.

 

[Sepheyer]

FML. Her left eye makes me want to set my hair on fire.

Would anyone have a fix for this particular image?

Naturally there are good ones in the mix too:

Spoiler: Illustration

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

But I wonder about fixing something that you are absolutely in love with:

Spoiler: Prompts

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