What I WISH I knew when I started (Stable Diffusion)

Video Statistics and Information

Video
Captions Word Cloud
Reddit Comments
Captions
when I first started using stable diffusion I wasted so much time just fiddling with different settings different props this and that so I'm going to save you that time by showing you some of the really important tips that are going to let you bring your ideas to life as quickly as possible and I'm going to show you one tip at the end that I can't believe I didn't know for literally months but it is a game changer number one is to start creating a Keepers folder from day one so whether it's something you generated yourself or you just found out online all of that metadata The Prompt and all of the other details about the image are usually saved in that image so what I would recommend for everyone is to go into your outputs folder create a new folder and I call it keep for prompts I give it a little underscore so it's always at the top and then that way when you're when you encounter something cool save it in here this is just a couple Generations I was doing while I was creating this video this one I quite like so I'm going to drag and drop it into the keep for prompts folder once you have a catalog of images you can always go into the PNG info tab here and drag and drop your images right here and then you'll see the positive prompt negative prompt and all the parameters for that generation from here if I want to iterate on that I can just say send to text to image instead of a cat got Emperor let's do a dog god Emperor it's a pretty cool result and I didn't have to waste time building anything from scratch one important thing to keep in mind is that when you're looking at PNG info it will tell you the model hash and the name of the model if it's available when you send this to text to image or image to image it won't necessarily change to that model so pay attention to that if you're not getting the results you expect the next one is when you inevitably start playing around with other models rather than the base 1.5 read the description of the model the Creator oftentimes includes some helpful language on how to properly prompt this model and they'll give you some examples many of them are fairly basic and will just give you a good starting point with a good negative prompt but some of them have entire user guides for them this is the RPG model and if you open this PDF you'll see there's an entire document on how to properly prompt this model this one even includes a workflow of starting with a seed going all the way to a high-res image these creators really know their models so feel free to click into here steal some prompts or drag and drop the images you like into your keep for prompts folder for later next is to start building up a library of styles in this drop down here you can select a style that you've created select it and then hit apply selected Styles and that'll just send it over to the prompt so instead of lightning striking the tree of life let's do a portal to the underworld swirling with energy there we have something pretty cool already so what I recommend is when you're in PNG info you found something you kind of like you send it to text to image if I want to reuse a style like this in the future I can just go to save style and here I recommend giving it also the name of the model because not all models will be able to use the same types of templates that we're creating here so I'll call it dreamlike and let's call it epic lightning hit okay and now this style will pop up into this menu right here if you're starting from scratch and you have zero prompt written right here you can just come in here select epic lightning and then hit apply selected Styles and it pops right back up I also recommend creating just generic type of styles so this is just like a standard positive only prompt that if I'm starting from scratch I can just throw this in here and I'm no no I'm going to get something fairly high quality and fairly realistic and of course you can do the same thing for negative prompts if you end up with a prompt that you save the style and you don't want it anymore there's nowhere in the user interface to do that so I'll show you how you can delete that first go to your main stable diffusion web UI folder scroll down to styles.csv open it with notepad or some other text editor and you'll see all of your Styles here with starting with the name of the style quotations for the positive prompt then the negative prompt and then a return so here at the very bottom if I want to get rid of this I can just backspace over that make sure I'm not deleting anything else from another prompt save it after that if you go into stable diffusion you'll see that the style is still there once you hit refresh the style goes away the next tip is about sampling methods so for the most part you really don't need to worry too much about what sampler you use a lot of people will just stick with Euler a and they get very good results the other popular one is this DPM plus plus 2m Keras or dpm2 Keras those tend to be popular but I want to just point out ddim because it is unique because it gets results really quickly with a small number of steps so I'm using my cat God Emperor prompt here and with just 10 steps you can see how incredibly fast the generations are and the results are not bad either they actually look quite good so especially early on when you're like writing a prompt and iterating on stuff really quickly and you just kind of want to want to see what that result is I recommend switching over to ddim run like a small batch count sampling steps of 10 is just fine do your generation and then you you'll get a selection of different images you can choose from and then maybe you take the seed from one of those images change to a different sampling method and continue or just keep using ddim it's really fast and then finally this last one is a bonus tip I can't believe I didn't know this for her so much longer but when you're working on prompts sometimes you want to add additional weight to a prompt you either do that with the parentheses around it one two or three sets of parentheses or you see that colon you know Point 1.2 or something like that to quickly change the weight you can actually just highlight highlight some words hit control hold control hit the up Arrow or down arrow and it will adjust that weight up and down so insanely useful to be able to just quickly change the weights of things so that's it there are a million more tips I could have included but I wanted to keep it fairly short so leave your favorite tips in the comments and I'll see you in the next video
Info
Channel: One Giant Leap
Views: 32,111
Rating: undefined out of 5
Keywords: stable diffusion, ai, stable diffusion tutorial, controlnet, ai logo, stable diffusion logo, logo tutorial, artificial intelligence, midjourney, dalle-2, dalle, stable diffusion text, ai typography, stable diffusion typography
Id: d-0FneYRggM
Channel Id: undefined
Length: 6min 29sec (389 seconds)
Published: Tue Mar 14 2023
Related Videos
Note
Please note that this website is currently a work in progress! Lots of interesting data and statistics to come.