[Stable Diffusion] Prompt Sharing and Learning Thread

[Nano999]

Ensure that you've installed the

You must be registered to see the links

, you can do this from the extensions tab. After installing you will need to restart AUTOMATIC1111.

Installation:

1. Open "Extensions" tab.
2. Open "Install from URL" tab in the tab.
3. Enter URL of this repo to "URL for extension's git repository".
   This -->
   You must be registered to see the links
4. Press "Install" button.
5. Restart Web UI.
6. Put the LoRA models (*.pt, *.ckpt or *.safetensors) inside the sd-webui-additional-networks/models/LoRA folder.
   Alternatively inside: Stable-Diffusion\stable-diffusion-webui\models\Lora
7. Go to "settings" "Additional Network" paste the path for your Lora folder
   ex. G:\Downloads\Stable-Diffusion\stable-diffusion-webui\models\Lora
   and click "apply settings".
8. In text2img or img2img expand the "Additional Networks" menu near the bottom and select "Enable" then select a Lora for "Model1" and adjust "weight1" , 0.8 is a good starting point. You can combine several Lora's but the result might vary. Some Lora's has specific key words or phrases for the prompt ex "shirtlift" or "braless" etc.
9. Create awesome images .

Thank you for explanation.
I am a bit confused. So I tested 5 images as below with 2099334373 seed
Image 3 by default should have weight of 1, right? because of the value at the end <lora:samdoesartsSamYang_offsetRightFilesize:1>
but it's different when comparing with image 5? why so? The weight is the same - 1?

Or was it doubled, like first time weight 1 from the prompt and for the second time from Additional Networks?
So it applied Lora 2nd time to the image 3? And applied it by 0.65 to image 3 in case of #4 image?

neg: (worst quality, low quality:1.4)

 

[Jimwalrus]

Thank you for explanation.
I am a bit confused. So I tested 5 images as below with 2099334373 seed
Image 3 by default should have weight of 1, right? because of the value at the end <lora:samdoesartsSamYang_offsetRightFilesize:1>
but it's different when comparing with image 5? why so? The weight is the same - 1?

Or was it doubled, like first time weight 1 from the prompt and for the second time from Additional Networks?
So it applied Lora 2nd time to the image 3? And applied it by 0.65 to image 3 in case of #4 image?

neg: (worst quality, low quality:1.4)

I'm not sure, I don't normally use the separate Network Modules weighting option - I do it manually with just the LoRA name and :weighting in the prompt. I don't know if it stacks, averages, multiplies or what!

 

[Mr-Fox]

Thank you for explanation.
I am a bit confused. So I tested 5 images as below with 2099334373 seed
Image 3 by default should have weight of 1, right? because of the value at the end <lora:samdoesartsSamYang_offsetRightFilesize:1>
but it's different when comparing with image 5? why so? The weight is the same - 1?

Or was it doubled, like first time weight 1 from the prompt and for the second time from Additional Networks?
So it applied Lora 2nd time to the image 3? And applied it by 0.65 to image 3 in case of #4 image?

neg: (worst quality, low quality:1.4)

I'm not sure what you are asking. But if you wish to replicate an image you need to replicate everything that has been used for that specific image. The same checkpoint "model", the same sampler and upscaler if used, and the same seed and all the same settings and resolution, the same additions such as lora's with the same weight etc and of course the prompt. If you are creating an image yourself and not replicating anything, the settings is up to you. You need to try out for yourself wich settings yields the best result.
The only way for us to know anything about an image is for people to post them with the meta data (generation data) intact.
This way we can say if anything is changed between two images.

 

[Mr-Fox]

If you are using a static seed, meaning in the seed box is a number series ex: 123456789, then the image should be the same or very similar each time you press "generate" as long as you don't make any large changes. If it has -1 instead, it means that the seed will change each time you generate an image and will give completely different images.

 

[Nano999]

Jimwalrus understood it correctly, the question was why image #3 and image #5 are not identical, as they have the same settings, but for the image #5 the weight of 1 was specified in the network tab. (so it was set 1st time in the promt and second time in network, so I thought it was doubled or smth as the imgs are different)

Spoiler: My images

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

 

[Jimwalrus]

Also, please be careful with the prompt "girl" - SD regards "girl" as being a female under about 16 or so. If you want an adult woman, specify "woman" in the prompts. The exception is "1girl" or "2girls" etc when in conjunction with "woman" prompts. On their own they will produce, as you've found, a young girl. That's obviously something to be very, very careful about!

 

[Jimwalrus]

Jimwalrus understood it correctly, the question was why image #3 and image #5 are not identical, as they have the same settings, but for the image #5 the weight of 1 was specified in the network tab. (so it was set 1st time in the promt and second time in network, so I thought it was doubled or smth as the imgs are different)

Spoiler: My images

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

Yeah, looking at the PNGInfo for each image, it seems to be additive when using the Network Modules weighting. I'd advise against using that tbh. Too easy to forget it's on when changing other things.

 

[Mr-Fox]

Yeah, looking at the PNGInfo for each image, it seems to be additive when using the Network Modules weighting. I'd advise against using that tbh. Too easy to forget it's on when changing other things.

I feel it's much more convenient to just use a slider rather than typing in the prompt. When I have found the amount of weight I want to use I just leave it while I fine tune other things.

 

[Nano999]

Well.. even if delete from the promt weight value, the picture #5 will not revert back to #3

 

[Nano999]

I manage to revert it, but look at the info
I broke the Lora promt name and got the image #3

cmd could not find that samyang lora file because of incorrect name, but actual result was applied with that lora xD

Well, I figured it out, if you want to use networks need to set weight to 0 in promt
The picture #5 is a picture with weight of 2, so when you specifice weight in promt, like 1 and then later 1 in networks the applied promt to the pic will be 2.

PS although it's strange that when you break the lora like in the pic below it still works as lora:0 in promt and if you delete this line completely it will be a different #3 pic, but logically it should not, as previous lora line was broken

 

[Mr-Fox]

If you use the extension I posted the install instruction for. There will not be any confusion.
you will simply enable Additional Networks and then chose a lora you wish to use and then set the weight with the slider.

 

[Mr-Fox]

well I don't care if she has one extra finger, she's gonna to be my waifu regardless


(prompt by

You must be registered to see the links

).

 

[Nano999]

If you use the extension I posted the install instruction for. There will not be any confusion.
you will simply enable Additional Networks and then chose a lora you wish to use and then set the weight with the slider.
View attachment 2452146

Fair enough, you don't need to specify anything in promt, just using networks, very convenient.

Btw I checked the result of setting value to 0 for lora in promt + setting network value for 2 and deleting promt completely, pictures will vary slightly
(both with weight of 2)

 

[Nano999]

Btw is it possible to reproduce civitai results?
Like this one:

You must be registered to see the links

I put the image into PNG info and send text2img, my result:
Also that image uploaded some settings called "Override settings Clip skip: 2", but they are hidden somewhere or what, If I empty the field, the name and the field itself will dissapear?

Not sure what did I miss here

Original:

My try:

 

[Mr-Fox]

Btw is it possible to reproduce civitai results?
Like this one:

You must be registered to see the links

I put the image into PNG info and send text2img, my result:
Also that image uploaded some settings called "Override settings Clip skip: 2", but they are hidden somewhere or what, If I empty the field, the name and the field itself will dissapear?

Not sure what did I miss here

Yes it is possible but very difficult. You can set clip skip in settings Stable Diffusion at the bottom.

Prompt:beautiful, masterpiece, best quality, hiqcgbody, anime, 1girl, sam yang, halter cut dress, mascara, flirty, blunt bangs, portrait shot, looking at viewer, glossy lipstick, nightclub, intricate details,

Negative prompt: (painting by bad-artist:0.9), easynegative, (worst quality, low quality:1.4), watermark, text, error, blurry, jpeg artifacts, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, artist name, bad anatomy, deformed iris.

You need:

You must be registered to see the links

,

You must be registered to see the links

,

You must be registered to see the links

both lora's with weight: 0.5


and these settings:

Size: 448x640 Seed: 3686603218, Steps: 28, Sampler: DPM++ SDE Karras, CFG scale: 7, Clip skip: 2,
Hires steps: 20, Hires upscale: 1.75, Hires upscaler: Latent (bicubic antialiased), Denoising strength: 0.55

 

[Nano999]

well I don't care if she has one extra finger, she's gonna to be my waifu regardless


(prompt by

You must be registered to see the links

).
View attachment 2452214

Also tried to recreate your waifu, but got an error for lack of 8x_NMKD-Superscale_150000_G upscaler
Could you share a link where to dwnld it?

Spoiler

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

Upscaling changes the end result so much...

 

[Mr-Fox]

He is using the old version of NED fyi. You can see that it's the same model hash.

 

[Sepheyer]

Ensure that you've installed the

You must be registered to see the links

, you can do this from the extensions tab. After installing you will need to restart AUTOMATIC1111.

Installation:

1. Open "Extensions" tab.
2. Open "Install from URL" tab in the tab.
3. Enter URL of this repo to "URL for extension's git repository".
   This -->
   You must be registered to see the links
4. Press "Install" button.
5. Restart Web UI.
6. Put the LoRA models (*.pt, *.ckpt or *.safetensors) inside the sd-webui-additional-networks/models/LoRA folder.
   Alternatively inside: Stable-Diffusion\stable-diffusion-webui\models\Lora
7. Go to "settings" "Additional Network" paste the path for your Lora folder
   ex. G:\Downloads\Stable-Diffusion\stable-diffusion-webui\models\Lora
   and click "apply settings".
8. In text2img or img2img expand the "Additional Networks" menu near the bottom and select "Enable" then select a Lora for "Model1" and adjust "weight1" , 0.8 is a good starting point. You can combine several Lora's but the result might vary. Some Lora's has specific key words or phrases for the prompt ex "shirtlift" or "braless" etc.
9. Create awesome images .

Thanks for this! I finally got over my laziness and ran LORA for the first time. Yea, great guide! Them LORAs are dangerous.

Spoiler: More

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

download: "NMKD Superscale"
And unpack it inside Stable-Diffusion\stable-diffusion-webui\models\ESRGAN
You need to restart SD for it to show up in the upscaler menu.
Though you would get pretty similar result with SwinIR_4x tbh.

You will also need

You must be registered to see the links

Place it inside Stable-Diffusion\stable-diffusion-webui\models\hypernetworks
go to settings and hypernetworks and select mjv4-hypernetwork and set the weight to 0.3 and press apply.

I'm running my GPU as hard as it will allow me so I hope you have something equivalent or more powerful.

 

[Mr-Fox]

Here's my generated image of the one you posted from

You must be registered to see the links

Very small things can make a difference. I don't have the same version as he. Like I already said he has the old version with the model hash:d8722b4a4d