5 Stable Diffusion Tricks That Will Make Your Art Instantly Better

if your stable diffusion art sucks then these tips might help you in this video I'm going to go over five steps that will help your art be better let's go number five use image to image tab when someone thinks of stable diffusion they think huh it's that text image tool like mid journey and this is also the sole reason why a lot of people are focused on the text to image Tab and not the image to image tab but in truth this tab has a lot of power that you're not using first off did you know that you could load an image into this tab then use either of these two image recognition tools deep Guru can recognize the tabs of an anime image and clip is used to recognize the tags of a real image let me load this anime image I generated and hit interrogate deep boru and there you go there are my tags I can load a real picture and click on interrogate clip to get its tags as well the second feature is the sketch feature here I have loaded a very basic sketch of an apple wait can you even call this an apple this looks like a drawing of a two-year-old but if I load my negative prompts here and then write Red Apple and hit generate you can see how it converted that sketch to a nice art it has its flaws but it's better to do a few generations to get the optimal output I can even improve it by using this marker like tool to draw a sketch of a table below it and then write apple on table pink background and there you go now this is nice isn't it in paint is also a very powerful tool you can paint over anything and change it into anything you want in my previous anime art video I painted over the top body of the art I generated and completely changed the clothing just by writing a few words you can check that video for more information but just remember how in paint isn't a tool you need to forget it can really add details you need after an art generation and is really helpful if your initial generation missed a few details number four use hires.fix to correct faces ninety percent of the time stable diffusion is going to end up giving you a really bad face and this is where hires.fix comes into play and helps you correct your messed up face if you write a normal prompt and hit generate you'll probably get a really bad face now if you click on that hires.fix button and then choose an upscaler like RS certain then hit the generate button you'll get some insanely good eyes here it's very important to keep an eye out for the settings if you don't want a lot of pressure put on your VGA a or if you want the generation to be done fast increase the higher steps but you'll have to sacrifice quality if you want more quality but don't mind the extra time and resources you can keep it zero change width and height to what you want or use the upscale buy box right here hire stat fix is much better than your usual restore faces option in fact here's an image with restore faces on and here's the same image of that same prompt with hires.fix on you can see the difference right number three use vaes if you have good art but the color of the art is the problem the vaes are the solution ever heard of color grading or color correction vaes pretty much help you do that where do you find vaes go to the site civet.ai and in the search box type vae then you'll get a list of vaes available for you to download take a look at this vae you can see how it adds more color to the original image and tries to give it a more photorealistic feel after downloading a vae you need to go to your stable diffusion web UI install directory there go to models and then click on the vae folder and paste your vae in here now go to your stable diffusion interface and click on settings on the left side navigate to the section called stable diffusion and in there select sdvae if it doesn't appear make sure to click the blue refresh button you might also notice there are two options called automatic and none if your vae has a similar name to the checkpoint or model you're using then automatic will load that vae or if you select none it will select the vae integrated to the checkpoint make sure to click the apply settings button to save the settings now each time you generate an art that particular vae will be applied to the generated art some interfaces allow you to change vae with each generation so keep an eye out for that too and one downside of this is how the vae might affect the generation so be careful to check any comments the creator has given in the description page number two use the Styles box let's say you generated a very nice art from a certain prompt but then you you close stable diffusion and someday wish to regenerate that same prompt or modify that prompt a little bit or you are tired of typing the same prompt every day and you want to keep it saved so you can just paste it every time you start stable diffusion so as an example I have a set of negative prompts that I often use now I'll take my cursor over to the save icon right here and then click it once I got a new box asking me to enter the name of the style how about I type negative prompts and if you're using this for a specific generation you could include that as well press enter and the style will be saved now if I go and reload the UI and go to the Styles box you'll see our saved style I click on it then click on the paste icon right here and there you go you've got your negative prompts you can save any large or small prompt this way and make the process much easier number one write prompts smarter not harder when comparing stable diffusion with mid-journey or when anyone mentions stable diffusion everyone thinks oh big prompts there's a huge misconception in the AI community that in order to produce good art and stable diffusion you need to write really long prompts well my friends you're still living in the Stone Age stable diffusion especially with the release of automatic 111's web UI has become much more advanced to prove my point here is a very very long prompt to generate a woman wearing a red dress in a ballroom background if I hit generate you can see it generates a nice art of a woman with all the details we mentioned let me hit generate again and here's another beautiful art so that proves that this prompt works but imagine how long it would take to write this prompt it's more than 90 characters long what if we can simplify this prompt alright let me show you step by step how I simplify this prompt you can see in the original prompt there are a ton of keywords about how art should be realistic and good quality but most models come with art trained in good quality so we only need to focus on realism rather than writing an essay about realism in the prompt I'll just type realistic then select the keyword and press the Ctrl and up Arrow key this increases the weight of the keyword and puts more Focus us on that keyword next we don't need to mention how the art should be high quality over and over again we can simply enter one keyword and be done we also don't need these perfect face beautiful face keywords because we already control any bad quality art with our negative prompts so all we need is the description of the woman and what she should be wearing then we just need to specify the camera angle for the shot and at last the background in one single phrase there we go we reduced a 94 word prompt to 35 words and if I hit generate and wait a few seconds I get an artwork of the same quality as before see no detail changed if I hit generate again there you go even nicer art the whole point of this experiment is to show how advanced stable diffusion has become sure if you're using its original model then you'll need these keywords but if you're using a well-trained model downloaded from civet AI then there's no need to put more effort writing prompts it's just about including what you want and balancing the weights you can even keep the negative prompts the same trust me guys prompts generation can go a long way and there are a lot of small tricks you to learn I'll be doing a video about it very soon so stay tuned and be subscribed to the channel so that's it those are my five tips hope it helped you and you learned something as always if you learned something make sure to hit the like button and stay subscribed for more awesome AI content like this comment down below what your favorite tip is thanks for watching AI gang and I'll see you all in the next video [Music] [Music] thank you

Channel: The AI Outline

Views: 5,721

Keywords: stable diffusion, stable diffusion tutorial, stable diffusion ai, stable diffusion ai art, stable diffusion art, stable diffusion installation, stable diffusion browser, stable diffusion gui, stable diffusion beta, stable diffusion online free, stable diffusion guide, stable diffusion install, stable diffusion prompts, stable diffusion better models, learn stable diffusion, learn stable diffusion fast, stable diffusion video, install stable diffusion tutorial

Id: rqcyfB49us4

Length: 8min 2sec (482 seconds)

Published: Sat Jun 24 2023