Screenshot with parameters was shared to show you the example of a prompt. Isn't it obvious? =)

@aine_captain Can you share these options please ? I'm new but learning fast, I hope :)

@hoblin That wasn't the question but I thank you to have shared some example with your own prompts, that helped me a lot. 

@hypopo Settings → User interface → Quicksettings list
You can add any settings there. I'm using this: sd_model_checkpoint, sd_vae

@hoblin Thanks

@aine_captainThanks 

@aine_captain OK, Thanks anyway.

@aine_captain All good! I've since switched to Automatic1111. Everything works great! One quick question however, it seems like eyes often come out beady, white or entirely black and no matter what I write as a prompt it's sort of hit or miss. Did you have any tricks to getting clear faces? (Am on Silicon and can't get Codeformer to work :c)

@aine_captain Thanks! Most of my images were distorted and I finally found a reason why.

Also you might be interested in that. I've had limited success with this model on 2.1. For example, I copied your prompt for the redhead wearing a copper armor to test the model and it turned out great (i can send you a picture on discord if you want). Perhaps one of the reasons it haven't worked for me most of the time is that the dpm++ 2m karras sampler is not available on SD 2.1

@aine_captain I simply downloaded liberty_main.ckpt and and dropped it into a models folder for ckpt files and it worked

@aine_captain my bad then

Take a look at mine prompts from the first review, you'll get an idea how Liberty prompting works. It's not hard. If you don't like to change your prompting style there are version of Liberty with broken CLIP. It will behave as you expect. Take a look at the description, it has all the info you need. ⬆️

It’s interesting how everybody asking for a prompts but nobody cares to leave a review.

@hoblin would love to leave a review if I knew how well a model works. Interesting how everybody wants a review left by people who haven't figured out how to make a model worth bothering with yet.

@hoblin Yes thanks Hoblin to have shared yours and I got that "natural language" works better than using key words. But even in my mother tongue I've never been good at describing precisely a face nor an attitude, so imagine in english and also consider that Captain's mother language is spanish :)
At least, what would you recommend to have a realistic photography and not a drawning ?

@aine_captain Thanks! And I can't even imagine trying to mess around with Stable Diffusion in Spanish, despite the fact that I speak it fairly well and understand it even better. Ni me lo puedo imaginar. Hat's off to those of you doing it in your second language!

@hypopo Hello, I'm russian speaking Ukrainian =) The good prompting cames with the practice. I even created an app to help me to experiment with different words. And, given that I love to share my findings, I appreciate the right of others not to do that. Would be great if it helps somebody else to improve their skills https://github.com/hoblin/prompt-master

@hoblin I'm trying to install your app but I can't run the prompt master app, http://localhost:8080/ is not accessible with the message ERR_CONNECTION_REFUSED.

I followed all installation steps except the third one "Follow the instructions in the yfszzx/stable-diffusion-webui-inspiration repository to generate previews for the words in the prompts." as it was a bit confusing and I really don't know what to do here !
Any idea ?

@hypopo Try to follow instructions without exceptions. There are a threat on UD discord which might help.

@hoblin well I gave up with your app, but I found a workaround with the wild cards extension and text files filled from https://describingwords.io/
I use a generic prompt with the wildcards, then I can refine the one I like.
It worth a try if you didn't already ;)

@hypopo So many people gave up installing Prompt Master, so I decided to convert it into a web service =) Stay tuned!

Use the same seed and description of the subject. I've been successful doing that to keep the same character in the image. Too many changes to the part of the prompt describing the character and it may change, and changing seed will absolutely change the character. The sampling method too for that matter.

@aine_captain While I'm asking dumb questions, is there any difference between the Pickle and the SafeTensor?

@aine_captain Oh wow, didn't know that! Thanks a ton for telling me! Hmm, is there some way to inspect these pickle files for scripts?

@vegetarian_crocodile not sure if linking to HF or even Github is kosher with the rules here, but Google "Pickle Scanner" and you'll find a variety of links to such a tool along with tutorials on how to install and use!

Please read answers to this the same question on other comments. Thank you.

Good question, but I don't think he'll tell you. You ask too much

@aine_captain Thanks for the answers, that's what I thought. The details you provided really help.

What is block merge? How do you use it in automatic webui and why is it a game changer?

Do you know why the stable diffusion 1.5 ckpt model and 2.1 ckpt model give pickel warnings on huggingface? Why do official models have pickle warnings? Obviously safetensor files are the ones to use but just curious if you know. Even the 1.5 and 2.1 ckpt files must be safe though from huggingface right?

@aine_captain Thanks for the info. Can you please clarify a few things for me?

1) How do you tell which layers of the AI network contain which concepts so you know what you can keep/merge? Is it just trial and error, or can you actually tell somehow?

2) Please let me know of your findings with block merges after you have figured things out. I'm interested in learning about what you discovered and what the best extensions you ended up on.

3) Is there a way to tell which trigger words are used/learned by a model from the model itself or do you have to rely on the author mentioning them when they post it?

4) You must have downloaded the SD 1.5 model from hugging face. Is that correct? I'm assuming you are using the safetensors version. Do you mind telling me the checksum of the file from webui. Do you also have the SD 2.1 model from huggingface? What's it's checksum if you don't mind.

Thanks again. Best of luck with the block merge :)

@aine_captain Thanks :)

It might be because of the prompt. Don't you mind to share one (both positive and negative)?

if you use lora, especially anime ones in combination with realistic, you need to lower the lora <name:[value]> sometimes as low as 0.1 - 0.3, and increase the value of the trigger words by the same amount, to get rid of face distortion.