[Stable Diffusion] Prompt Sharing and Learning Thread

FreakyHokage

Member
Sep 26, 2017
327
480
Yes you can post links, no problem. If you look around in this thread we do it a lot. I'm familiar with "badhands" however Ti's has been very problematic in my experience so I choose to stay away from them for now.
They tend to take over too much and then linger in the result before they have been "purged" even after you remove them. Hypernetworks has been the same so for now I stick to Loras. I'm aware that there is a new type of Lora or a couple named , at civitai it has the category name . I have not messed with these yet but I have the Locon extension so it's only a matter of time.
I've only messed around with a few Lycoris. I have one now that does Princess Diana I made some pretty decent images last night but I still need to work around with it lol. But I do know what you mean about ti's I have the age slider added and with too many prompts it messes all up. 00012-2989291205.png
 

FreakyHokage

Member
Sep 26, 2017
327
480
So any good automatic plugins or LORAs to handle eyes? Because a lot of my eyes come out messed up.
I was having issues with eyes as well and this video helped me out. I've been using hirex.fix and it's gotten better. I know you can use inpainting to do it as well but I'm still learning how to use inpainting and img2img.



 
  • Like
Reactions: Mr-Fox

me3

Member
Dec 31, 2016
316
708
Stacking embeddings and loras will often screw things up for you as well. The more poorly done they are the more "wrong" things they affect. Just because it says it'll just give a hairstyle or clothing it very often causes impacts with other things.Which is why people should put much more effort into the job when training them, unfortunately many just go for the simple shortcut route and you get stuck wondering "WTF made that happen"
 

Jimwalrus

Active Member
Sep 15, 2021
902
3,339
Stacking embeddings and loras will often screw things up for you as well. The more poorly done they are the more "wrong" things they affect. Just because it says it'll just give a hairstyle or clothing it very often causes impacts with other things.Which is why people should put much more effort into the job when training them, unfortunately many just go for the simple shortcut route and you get stuck wondering "WTF made that happen"
Abso-fucking-lutely!
I do find though that it's not as bad if you combine embeddings and LoRAs i.e. if you use one TI and two LoRAs it's less likely to screw things than three LoRAs.
I haven't done any empirical testing though, that could just be a result of the chances of hitting a bad one increasing because more people are generating LoRAs these days than TIs.
 
  • Like
Reactions: Mr-Fox

Sepheyer

Well-Known Member
Dec 21, 2020
1,528
3,598
Imma drop here my first ~10 attempts using control net in ComfyUI. Finally got over my laziness and tried to see for myself what's up.

The motivation was to get those arms behind the head: a pose that I almost never can achieve using the text prompts.

Welp, the results are pretty fucking great considering I am using everything off the shelf.

What's especially notable are the very final results in the set - check out the ease when capturing full-body, down to the heels. That's a fucking gamechanger (for me).

Granted, there are fused parts, but for the time being IDGAF, I am ecstatic of how well it works. Finally, the ComfyUI page says the control net comes with performance hit, but saw no performance hit whatsoever - may be they just forgot to update the page from the very first implementation of the control net.

a_10632_.png
You don't have permission to view the spoiler content. Log in or register now.
 
Last edited:

Mr-Fox

Well-Known Member
Jan 24, 2020
1,401
3,793
Imma drop here my first ~10 attempts using control net in ComfyUI. Finally got over my laziness and tried to see for myself what's up.

The motivation was to get those arms behind the head: a pose that I almost never can achieve using the text prompts.

Welp, the results are pretty fucking great considering I am using everything off the shelf.

What's especially notable are the very final results in the set - check out the ease when capturing full-body, down to the heels. That's a fucking gamechanger (for me).

Granted, there are fused parts, but for the time being IDGAF, I am ecstatic of how well it works. Finally, the ComfyUI page says the control net comes with performance hit, but saw no performance hit whatsoever - may be they just forgot to update the page from the very first implementation of the control net.

View attachment 2696495
You don't have permission to view the spoiler content. Log in or register now.
Holy smokin fudge on a stick.. That's just awesome. Great job. :D(y) I have been putting it of for a long time myself also but I very much intend to start messing with control net stuff as well.
 
  • Red Heart
Reactions: Sepheyer
May 13, 2023
35
31
Stacking embeddings and loras will often screw things up for you as well. The more poorly done they are the more "wrong" things they affect. Just because it says it'll just give a hairstyle or clothing it very often causes impacts with other things.Which is why people should put much more effort into the job when training them, unfortunately many just go for the simple shortcut route and you get stuck wondering "WTF made that happen"
This is becoming an issue with Civitai, I'm afraid. I've noticed more and more Loras with someone going: this doesn't work well, LOL, which sort of raises the question: why are you putting it out there? I could see if someone asked: why it wasn't working, but a lot of them are just chucking stuff out there.

Imma drop here my first ~10 attempts using control net in ComfyUI. Finally got over my laziness and tried to see for myself what's up.
those are noice images! Great job!

i've been looking at Comfy UI, but I got my start on Automatic1111--so how is comfyUI? Does it give you more control?
 

Sepheyer

Well-Known Member
Dec 21, 2020
1,528
3,598
This is becoming an issue with Civitai, I'm afraid. I've noticed more and more Loras with someone going: this doesn't work well, LOL, which sort of raises the question: why are you putting it out there? I could see if someone asked: why it wasn't working, but a lot of them are just chucking stuff out there.


those are noice images! Great job!

i've been looking at Comfy UI, but I got my start on Automatic1111--so how is comfyUI? Does it give you more control?
A1111 has more features, but I switched to CUI because it let me do landscape-oriented images like this using its composing feature:

a_05676_.png

Then I stopped using landscape, but using CUI stayed as a habit.
 

Sepheyer

Well-Known Member
Dec 21, 2020
1,528
3,598
Holy smokin fudge on a stick.. That's just awesome. Great job. :D(y) I have been putting it of for a long time myself also but I very much intend to start messing with control net stuff as well.
Here a starter pack for the poses:

pose sholder wide 001.png pose arms behind head 022.png pose bent 001.jpg

Here is the tool I use to generate these:

Important: the defailt stickman comes with wide shoulders (~40 units), I change them down to 25 units. The prompt for the body-ratios appears when you click on a stickman's nodes.

Here daring Ms. Sanchez, the physics teacher, to do a split.

a_10737_.png
You don't have permission to view the spoiler content. Log in or register now.
 

Mr-Fox

Well-Known Member
Jan 24, 2020
1,401
3,793
Here a starter pack for the poses:

View attachment 2698581 View attachment 2698585 View attachment 2698586

Here is the tool I use to generate these:

Important: the defailt stickman comes with wide shoulders (~40 units), I change them down to 25 units. The prompt for the body-ratios appears when you click on a stickman's nodes.

Here daring Ms. Sanchez, the physics teacher, to do a split.

View attachment 2698578
You don't have permission to view the spoiler content. Log in or register now.
Thank you. Now that you are using the open pose editor, you might be interested in pose packs. There is a category for poses on civitai and there are other places as well. I found out about this long time ago and I have been meaning to try it out but haven't gotten around to it yet.


Examples:
1686839540821.png 1686839568047.png 1686839597182.png
Source:

Examples:
1686839793155.png 1686839861897.png 1686839886696.png
Source:

Examples:
1686840010611.png 1686840022666.png 1686840044305.png
Source:
 

Sepheyer

Well-Known Member
Dec 21, 2020
1,528
3,598
Thank you. Now that you are using the open pose editor, you might be interested in pose packs. There is a category for poses on civitai and there are other places as well. I found out about this long time ago and I have been meaning to try it out but haven't gotten around to it yet.


Examples:
View attachment 2699613 View attachment 2699626 View attachment 2699628
Source:

Examples:
View attachment 2699630 View attachment 2699636 View attachment 2699637
Source:

Examples:
View attachment 2699642 View attachment 2699643 View attachment 2699644
Source:
Thank you! I raided them hard!
 
  • Yay, new update!
Reactions: Mr-Fox

me3

Member
Dec 31, 2016
316
708
Lycoris was mentioned and i was getting rather bored trying to beat the AI into understanding that skin come in a massive variance of tones and body parts can have very different "builds" on the same person...so far it's winning...

Grabbed the huge amount of 6 fairly low quality 20-30 year old photos and threw them at the AI. Some other things got thrown back, digging through it there was something that looked slightly similar. If anyone is able to identify it without cheating I guess there's some hope...
(considering the prompt is like <10 words total i guess a few things could have been fixed there)

00077-818029826.png

one of the training images is in the spoilerand the .safetensors should be attached, but be warned it has A LOT of issue, it's really touchy about sampler and seed, and i've only tested it for a few images on one model.
You don't have permission to view the spoiler content. Log in or register now.