[Stable Diffusion] Prompt Sharing and Learning Thread

Anon1341 · Apr 14, 2023

my first work with SD and dreamshaper.
I spent about 4h of learning stuff, generated about 2k images in total :X I hope Ill be better at this....

Does anyone know a good way to inpaint a transparent clothes while keeping the previous things like, e.g. a tattoos and so on?

orginal image promnt:

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

final result:

Elefy · Apr 14, 2023

Sepheyer said:
this is a prompt sharing thread

This thread is for our praise

https://f95zone.to/threads/ai-art-show-us-your-ai-skill.138575/

Anon1341 congratulations for your 4h effort and welcome

picobyte · Apr 14, 2023

riaguas · Apr 15, 2023

miaouxtoo said:
StepSisWasNeverStuck here's another few for you. You'll have to change the model to chilloutmix and rearrange the text.
Probably apply an IG filter on top as well or use phtoshop to change lighting.

Get png info from PNG INFO tab

View attachment 2535885 View attachment 2535886
View attachment 2535900 View attachment 2535894

Oh - also try your luck with the hyperbreasts LORA. And you'll have to compare text with the IG channel to see how you can make it more like that.
Probably enforce the "slim thicc" or prioritise thicc and play with that.

GL. I'm going back to messing with other styles.

I tried to replicate the image on the bottom right but despite copying alll prompts and using the same model, it sill doesnt look quite as yours, how can this be?

Mr-Fox · Apr 15, 2023

Alanrigas said:
I tried to replicate the image on the bottom right but despite copying alll prompts and using the same model, it sill doesnt look quite as yours, how can this be? View attachment 2547873

It could be a setting that you are not aware of and/or a different VAE etc.
It can also be that he is using a different version of the same checkpoint and/or Lora's
This is just the nature of SD, it's not always possible to get an exact replica but you should be able to get something very similar.
A tip is to load the image miaouxtoo posted into PNG info and then "send to text2img". It's a much more convenient way than doing it manually and some settings that are "hidden" in the menu's is going to be added as override settings. Unfortunately not all settings will be transferred. You have to change Face Restoration to Codeformer manually if you have it set to GFPGAN.
Here's my result:

miaouxtoo · Apr 15, 2023

Alanrigas said:
I tried to replicate the image on the bottom right but despite copying alll prompts and using the same model, it sill doesnt look quite as yours, how can this be? View attachment 2547873

Do you have the negative loras? And it looks like you might have changed some of the negative prompts? I can't tell ... I didn't look at mine haha.

Mr-Fox · Apr 15, 2023

miaouxtoo said:
Do you have the negative loras? And it looks like you might have changed some of the negative prompts? I can't tell ... I didn't look at mine haha.

I took a look at his image and he didn't replicate the settings, he was not even using hires..
Negative lora? I didn't see any in the data, only the negative tags.

Did you mean negative embedding? "easy negative" I didn't notice this before.

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

docar8 · Apr 15, 2023

miaouxtoo said:
Do you have the negative loras? And it looks like you might have changed some of the negative prompts? I can't tell ... I didn't look at mine haha.

Dude I read through all the pages (1-61) and the knowledge is really overwhelming. I noticed the discussion on this model and tried some stuff myself. But I want your opinion on this. Do you think the model used in these two are the same?

miaouxtoo · Apr 16, 2023

Mr-Fox said:
I took a look at his image and he didn't replicate the settings, he was not even using hires..
Negative lora? I didn't see any in the data, only the negative tags.
Did you mean negative embedding? "easy negative" I didn't notice this before.

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

oh yeah - my bad! I meant negative embedding

I was just skimming messages and not really paying attention haha

miaouxtoo · Apr 16, 2023

docar8 said:
Dude I read through all the pages (1-61) and the knowledge is really overwhelming. I noticed the discussion on this model and tried some stuff myself. But I want your opinion on this. Do you think the model used in these two are the same? View attachment 2549208 View attachment 2549220

I vote yes for same model. It might be a different upscaler, or like photoshop after or something... the top one is just more golden-hour / warm tones

agenc · Apr 16, 2023

how can i create this image, i am using SD can you help me

osanaiko · Apr 16, 2023

agenc said:
how can i create this image, i am using SD can you help me

1. make sure you have a nvidia GPU with at least 6gb VRAM, ideally more. (it's possible with an AMD GPU but this is less common so you'll have to find your own way)
2. find and follow some instructions to install the Automatic1111 local install
3. watch instruction videos on how to use stab diff webui
4. maybe download some alternate generative models like AnythingV3 or Kotomix. You get these from Civtai.
5. practice practice practice
6. go deeper by reading through public shared info in places like this:

You must be registered to see the links

Read the first post of this thread, it has lots of info to get you started.

Mr-Fox · Apr 16, 2023

agenc said:
how can i create this image, i am using SD can you help me

First question is what kind of rig do you have? Amount of vram? Just so we know if you are hardware limited or not.
Where did you get it? If it was on civitai or a forum then maybe there are some clues in a comment wich checkpoint etc that was used. Other than figuring out wich checkpoint to use it's a process of working on the prompt, describing what you wish to see in the image and the negative prompt what you don't want to see in the image.
You can go to civitai and look for a checkpoint that has similar sample images and if they have shared the prompt, use it as a starting point. We can help you figuring out the settings and give suggestions about the prompt, but you need to get the ball rolling yourself. Besides it's much more satisfying to do things on your own

*Civitai is having some downtime atm, they are working on getting it up again.

You must be registered to see the links

Sharinel · Apr 16, 2023

I wish I could get the photos to show the whole body rather than just the face/upper torso. Sometimes it shows a bit more but very rarely the whole person.

I've tried adding in things like high heels or long legs to try and train it to show full body, but no dice

Prompts in the png, checkpoint is Hyperv1

Mr-Fox · Apr 16, 2023

Sharinel said:
View attachment 2550680
I wish I could get the photos to show the whole body rather than just the face/upper torso. Sometimes it shows a bit more but very rarely the whole person.
View attachment 2550690
I've tried adding in things like high heels or long legs to try and train it to show full body, but no dice

Prompts in the png, checkpoint is Hyperv1

Try to add things like "cropped" "closeup" and anything else you can think of that describes what you don't want in the image and place it in negative prompt. In positive put things you want to see in the image such as "whole body view" "distanced" "zoomed out" or anything else you can think of that describes what you want to see in the image. People often forget to use the negative prompt and is only focusing on the positive tags and misses out on the most powerful way of using the prompt. The most powerful way is to use positive tags and negative for the same visual aspect.

Example

Positive Prompt: Something Small
Negative Prompt: Something Large

Then when you also add weight to it, it's dynamite stuff

Positive Prompt: Something Small:1.2
Negative Prompt: Something Large:1.2

You have put a very big emphasis on "Something Small" .

Sepheyer · Apr 16, 2023

Sharinel said:
View attachment 2550680
I wish I could get the photos to show the whole body rather than just the face/upper torso. Sometimes it shows a bit more but very rarely the whole person.
View attachment 2550690
I've tried adding in things like high heels or long legs to try and train it to show full body, but no dice

Prompts in the png, checkpoint is Hyperv1

So, about doing fullbody renders. Normally, if get a fullbody, including heels, then the face is messed up. If you get the face right, then the heels won't show. There is a workaround tho - it has been debated somewhere halfway through this thread.

What you do, is send the picture to inpaint, or somesuch, and get the SD to continue the image down from where the original was left off. This way you'll get the rest of the woman. Naturally, there is a gotcha as I think the quality drops notably.

picobyte · Apr 16, 2023

[Stable Diffusion] Prompt Sharing and Learning Thread

Newbie

Member

Active Member

Newbie

Well-Known Member

Newbie

Well-Known Member

New Member

Newbie

Newbie

New Member

Engaged Member

Well-Known Member

Active Member

Well-Known Member

Well-Known Member

Active Member

Well-Known Member

Active Member

Well-Known Member