Prompt Wildcards in Stable Diffusion or Dynamic Prompting is wonderfully random

With this custom script enabled in your local stable diffusion install, you can now add some variables to your prompts! This means that each dynamic prompt will be unique and you can really enjoy discovering some happy accidents this way, or use it to cycle through known prompts that you enjoy.
The big lists of scripts is here, but you can skip this and just go to the pages I jump to in the video.
github.com/AUTOMATIC1111/stab...
THIS IS WHAT YOU NEED:
Grab the dynamic prompting script here: github.com/adieyal/sd-dynamic...
You can download the starter variable lists here: github.com/jtkelm2/stable-dif...
If you don't have a local installation, check out this video: • Stable Diffusion 1.5 -...
Check out Dream Studio here: beta.dreamstudio.ai/
Here are some excellent references on CFG_Scale and Steps and the Sampler settings that might work best for the type of imagery you wish to create.
Stable Diffusion Sampler Studies:
docs.google.com/spreadsheets/...
Stable Diffusion CFG Studies:
docs.google.com/spreadsheets/...
Join my Discord server, where we create ideas together!
/ discord
New! Become a member to get exclusive access to perks!
/ @sedetweiler
Enjoy this video? Consider buying me a coffee! ko-fi.com/sedetweiler
Twitter: / sedetweiler
Instagram: / sedetweiler

[@techtalkmike4968]

I just started my journey into AI generated art about two weeks ago and your videos have been absolutely invaluable to me. What an addiction it has become! Thank you!

[@sedetweiler]

Great to hear! Glad they are helping you out!

[@dement242]

You can right-click on "Generate" for the option to "Generate forever"

[@sedetweiler]

Thanks for the tip!

[@skrotov]

omg)) i love you

[@KadayiPolokov]

Scott if you edit the winuser bat file you can add --autolaunch inside the brackets of the command line ARGs part and SD will automatically open up in your default web browser once loaded. Or just make the local address a bookmark as it doesn't change. Don't need to copy-paste it every time.

[@sedetweiler]

I have it at a shortcut, but wanted to show this method in case people didn't know about it. Thank you for the autolaunch tip!

[@litty4913]

Thanks alot for this, i was struggeling to find out how this works :)

[@ArielTavori]

FYI, you can leave the batch size at 1, right click the generate button and select generate forever. It's a fairly new feature... Also the maximum batch size used to be like 8 until recently. Now it's at least 100.

[@trent797]

Sounds cool, but how long before it fills up your hard drive? Let's say you have 100GB free. Would it fill that up if it ran overnight?

[@marcdevinci893]

My maximum batch size is 16 and don't see anywhere in the settings to change it. Would you know?

[@marcdevinci893]

This is really awesome! Thanks for sharing

[@mafallo]

exatctly what I was looking for, thank you very much

[@sedetweiler]

Sure thing!

[@TheH1st0ry]

Very helpful! Thank you!!

[@masterboxxx]

so clear and cool explanation , thanks

[@sedetweiler]

Glad it was helpful!

[@madbuldog]

Awesome video as always, love all of your videos helped me a lot on my job as a designer for Cover Book Illustrations and as concept artist and game art illustrator. Now for the Question: You, analog_dreams and Olivio Sarikas a weekly AI podcast when? I would love to listen to you guys exchanging ideas for use of AI and sharing your experiences.

[@sedetweiler]

Olivio and I have chatted a bit on Facebook but have not decided on a collaboration yet. I would be open to ideas!

[@HalkerVeil]

Love it. This will be fun to play with.

[@sedetweiler]

I totally agree!

[@aksisoylenemez]

that was really nice and informative

[@sedetweiler]

Glad you liked it!

[@jgodvliet]

Thanks for your tips 😉👍

[@sedetweiler]

Happy to help!

[@1DusDB]

Example of a prompt in near future: A __pictureType_ of a __noun__ looking __physicalAspect__ feeling __emotion__ wearing __clothing__ in __pose__, __picSettings 😉 And humans won't either have to create the prompt, just get from web some already made txt files... that could be already pre-loaded in the app... so we will just have to push the "generate" button... oh wait... neither that🤔 Anyhow, thanks a lot for your video, very helpful.

[@sedetweiler]

You are not far off! 🤣

[@themonotropicguitarist]

EDIT: Apparently, according to my smart friend, if you run into this problem you need to go to the File Extension Options within the File Explorer and go to the "View" tab and then find and uncheck "Hide extensions for known file types." this will show the .txt at the end of the file and you can remove it which fixes the issue... Hi, great video! I was wondering if you (or anyone else here) knows why I can't save the dynamic prompting RAW as a .py ? It will only save the raw as a text file and I can't find anything about this issue. I have put .py at the end of the file name and it still sits as a text file and does not pop up in the web UI. Thanks!

[@sedetweiler]

Hmmm, I have never had this issue. As long as it is ending with .py and not .txt you should be fine. You can also restart to be sure it sees the change or use the button at the bottom of settings to reload the gui.

[@LouisGedo]

@Scott Detweiler Thank you for creating such excellent content on AI art creation; I really appreciate that you're doing this for people like me who have zero programming skills. 🙏 👏

[@sedetweiler]

My pleasure!

[@3dtechnik]

According to the github page, this is an extension now, so it has to be installed differently. I just stumbled across this, because SD would not list the script in the dropdown after having it installed as demonstrated by you. After using git for the installation as an extension it works just fine (but the UI looks quite different).

[@soapsudtycoon]

It's now an extension which you can install directly from the extensions tab in the webui

[@Kampftroll]

If you hold ALT and click the RAW button, it will download automatically so you don't have to upen and click "Save As"

[@sedetweiler]

well that's awesome! thank you!

[@infographie]

Excellent

[@Starshipx1]

Cool work do you have a video on automatic1111 how to setup and run 3d there is not a lot of videos on youtube regarding this thank

[@anikandr]

Hello! Thank you for the wonderful tutorial! How to make the picture visible when generating, as on 5:32 minute? Thanks, I found it in the settings!

[@AresWuzHere]

Just discovered you and easily installed stable Diffusion thx to you. Very interesting stuff. One thing I am curious is how do I upscale video's. thank you very much.

[@sedetweiler]

Welcome aboard! I will be doing more upscale soon. I have one already using chaiNNer, which I think is a pretty amazing tool.

[@AresWuzHere]

@@sedetweiler very good, I'm watching that video right now.

[@cl4yt0n57]

I got two questions: 1. When using one of the wild cards, are you able to be more specific? Like choosing a specific artist name you want in your prompt from the wild card artist.txt. 2. Where did you get the wild card "emotions" from?

[@tetsuooshima832]

For me it's interesting only to discover new artists and styles. Other than that I don't want to add more randomness to this program. Almost everything creates randomness in there, the steps, the scale, the eta noise, the denoise strength, the CLIP layer, the model and even a different version of the same model, and so on... there is no end to it x) Actually, too bad I can't combine this with X/Y Prompt script, I could go wild and generate more than 30 styles at once for the same prompt, just because :p

[@sedetweiler]

I am sure that is coming! It might already be here, but might require some messing around to get it

[@creativi-tea398]

I followed the steps, saved the raw text to the script folder, edited the .Bat file and saved the others to a new folder titled wildcards. However when I load it up, The Dynamic prompting option isn't in the drop down for me. Not sure what's wrong. I went through the steps a second time to make sure and I did it exactly, yet no dynamic prompting option is available to me.

[@zifengjiang5899]

Same here, have you solve it?

[@DanCreaMundos]

Hmm well I can't seem to find the way to make the script show up in the list, followed the exact instructions, even tried to restart the components from settings but for some reason it won't show. As soon as I can try it I'll comment on my experience

[@ArielTavori]

Great videos! I've made a list o settings/locations like "on a beach/in a plane/on a crowded Street/in a cave... A list of male, and a separate list of female celebrities; some customized lists of adjectives for different uses; and after some experimentation, I've stopped using the artist names because I find the weights are too random, with some artists names having hardly any effect, and some completely overriding and ruining the rest of what would otherwise have been an interesting prompt...

[@soapsudtycoon]

The extension also has collections of wildcard files which you can use to create your own lists.

[@blackvx]

I think I will just add the Juice file. It's already challenging to funnel SD to generate something I would hope for. 🍋🍒🍋♠ Thank you!

[@sedetweiler]

I do love that file as well!

[@ZeroIQ2]

This is really cool, I wonder if it is recursive?, meaning that you could put __WORD__ in a text file, so you could use one prompt to use lots of variables at the same time. So what I mean is a text file with artists and another text file with nouns, then another text file called varcombos and in varcombos it would have a list of "__artists__ __noun__ ()new line) __artists__ __noun__ __artists__' etc... then in your prompt you just type __varcombos__

[@sedetweiler]

I don't believe so, but that might change over time as this is quite new.

[@chrisblue6095]

Any chance you'd share your "noun" list for the "too lazy to make our own".

[@sedetweiler]

There is one here that is a good start. www.kaggle.com/datasets/leite0407/list-of-nouns

[@Xbudz]

i'm getting an error: ModuleNotFoundError: No module named 'prompts'

[@jeffrussel7104]

I’m getting the same issue. And the Dynamic option isn’t coming up in the script bar.

[@Aiv3rs]

Hey is there a way to use a hand drawing and improve it with AI Generation? I tried to use one of my drawing in a prompt and I didn't get any good result :/ ( I am using stable diffusion)

[@ZeroIQ2]

This is where the X/Y plot is your friend. You can tell stable diffusion to create a grid of images using different CFG and Denoising values. If you go down to "Script" and select X/Y plot and then set "CFG Scale" for the X type and give it a value of "1-29(+5)", then set "Denoising" for the Y type, with a value of "0-1(+0.25)", then click generate, Stable Diffusion will generate a grid of images using lots of different values, hopefully one or more of the results are want you are looking for.

[@FilmFactry]

Scott, sharing back a tip for your word lists. Instead of artists/photographers, make a list of publications. AI understands the "editorial" look of Vogue, Architectural Digest, National Geographic etc. Field and Stream:-)

[@sedetweiler]

Great idea!

[@Pauluz_The_Web_Gnome]

I was also wondering, what gpu are you currently using? Because it's lightning fast! according to mine...for the same picture gen. with the same settings for the boy in a yard, took me 1m 36.79s. Arrrrrrgh! and it took you only 6 secs. !! OMG.. (a dream for me) My guess is, you have at least a RTX 3080 12GB vram? correct? I am on a budget so I want to buy this one, is that ok for SD? MSI GeForce RTX 3080 Ti GAMING X TRIO 12G? I think € 1200.- is still a lot of money! Maximum I have spent for a GPU in the past is max. 600 euro's!

[@sedetweiler]

I do use a 3080! Good guess!

[@Pauluz_The_Web_Gnome]

@@sedetweiler i've ordered a Zotac RTX 3080ti, and it will arrive today within a couple of hours!!

[@Pauluz_The_Web_Gnome]

Ok, good for you! Lot's of succes Pauluz! Thanks!

[@Pauluz_The_Web_Gnome]

Okay cool, thanks man!

[@temporallabsol9531]

So good right?!

[@sedetweiler]

It's amazing!

[@MrMeepzor]

When I discovered wildcards and dynamic prompts my mind was blown but it is ultimately disappointing as whatever options you set up, it will just render its picks to your prompt permanently. The variables you set up in your prompt disappear and are permanently replaced by whatever random item it chose from your wildcard list or dynamic prompt. If you want to randomize your options again you have to re-input all the variables into your prompt again, and each time it will permanently resolve its choices to your prompt. Why isn't there a version of this that allows your prompts to remain as-written and invoke these arrays or lists each time without having to re-write them into your prompt? Annoying. I want to be able to set up a prompt that produces different results each time, from within a range that I can specify. Instead this just does it once and then I have to set it all up again like dominoes.

[@jefffrilot9667]

Looks like this install tutorial is obsolete

[@sedetweiler]

Yup, this is moving at 800 miles an hour!

[@soapsudtycoon]

@@sedetweiler Might be worth having a look at the Magic Prompt feature which is a lot of fun

[@tetsuooshima832]

A

[@kallamamran]

As if SD is not random enough without "presudo random" words in the prompt 😆

[@soapsudtycoon]

Since the prompt is the single most important aspect of a generation, if you're going to vary one thing, that should be it. It's also really useful for artist discovery - e.g. A lonely woman on a windy hill by __artist__ and __artist__ (or equivalently {2$$__artist__})