Why everyone else's Stable Diffusion Art is better than yours (Checkpoint, LoRA and Civitai)

so if I told you to make these images in stable diffusion using the base standard models you'll have a very hard time unless you're very good at prompting and the reason for this is that the standard stable diffusion models such as SD 1.4 or St 1.5 a very good all-rounders but they don't excel at specific tasks so stuff like photorealism or comic book art studs now the way to get around this is to use custom models so one good website to get these custom models from is a website called civetai.com and when on this website you'll primarily be working with one of two files checkpoint files or Laura files and the difference between the two if you think about the standard stable diffusion model you've been using up till now think of that as a standard car with the checkpoint file we're sort of saying we're changing the entire core and we're using another one with a lower file on the other hand it starts saying we're keeping the same car however we're just modifying it and this analogy will make more sense as you actually start to use these models so first to install them click on a Model that looks interesting to you so I'm going to go with the realistic Vision so as the name suggests really good for making realistic looking images and just scrolling through you'll be able to get a sense of what this model would be able to produce and if you scroll further down you'll be able to see what the community was able to make using this model and so the way to download this model is simple you just press the download button here and so while that's downloading you want to make a note of the trigger words that the model uses now with checkpoint files I found that the trigger words or how many or if they use them are all is different from model to model so stuff like the deliberate model doesn't use any trigger words inkpunk diffusion only uses one and realistic Vision uses multiple now as a rule of thumb and what I found is if you have no trigger words you don't need to worry about anything if you have one trigger word you'll need to include that word in your prompt for it to actually activate the model if you have multiple trigger words it doesn't necessarily mean you have to include all of them in your prompt sometimes you may have to only include one or maybe none at all so in the case of this realistic vision the trigger words here just influences the final style of the actual image and so the best way to know how to actually use these trigger words and what they do is just to go on the example images and just flick through and look at the prompts on the right and see if they're using the trigger words and how they affect the actual final image so now that the model has finished downloading what you want to do is go on your stable diffusion folder go to models stable diffusion and then paste in the model file inside here once that's done go back to your stabled fusion and then click on the red button over here show and hide extra Networks go to checkpoints because that was the file we downloaded then click refresh and you should be able to see the realistic vision model over here so now just clicking on it will change the model that you're using so right now I was using stable diffusion 1.4 it's now going to change to realistic vision and now what I'm going to do is paste The Prompt that I saw from the image earlier on the example page then click generate and there we go this was the image I was able to generate of course it doesn't look exact because I didn't exactly follow the the generation data that he used but you get the idea it's very good for creating realistic looking images so now moving on we have Laura files and the way to set up blower files is slightly different so now we have the Studio Ghibli Laura file and using this file will allow us to create images in the style of the famous animation movies that Studio Ghibli makes so to begin I'm just going to click the download button over here and so while that's downloading alongside the trigger words that you'll need to remember you'll also need to pay attention to the base model that is being used so in our cases SD 1.5 and the reason we needed to pay attention to this will become apparent once we've downloaded the Laura and start using it once the lore is downloaded you'll want to go back into your stable diffusion folder go on models but this time go into the lower folder and paste in the file over here now going back to stable diffusion you'll want to click on the red icon again go on Laura click refresh and now you should be able to see The Studio Ghibli style card now clicking on the card what will happen is that the model itself won't change instead you'll just get this piece of text and it will say Laura Studio Ghibli style offset one and so you will always need to include this text alongside your trigger word and also alongside your prompt to get the results you want and so I managed to copy and prompt and generate the image but the image itself did not look exactly how the other Laura files examples did and this was because remember how I told you you had to pay attention to the base model that was being used so over here right now we're using the realistic Vision model alongside the Studio Ghibli Laura file and that's all fine however it was not intended for it to be used like this so what it was intended was for us to actually use the SD 1.5 checkpoint alongside the Studio Ghibli file to actually get the results we wanted and there's nothing wrong with using different checkpoint files with lore files that were not intended It's just sometimes that would lead to unexpected results and sometimes it will make the image sometimes look even better so it's really a case of trial and error and just so to further demonstrate my point I was able to find this example image and over here even though this is using the Studio Ghibli lore file the model that is being used is the abyss orange mix 2. and this just goes to demonstrate that you don't need to exactly follow the base model that is being used you can't mix and match and so that is all and I'll see you in the next video

Channel: Neo Professor

Views: 37,500

Keywords:

Id: GBYD-CyTt7w

Length: 6min 15sec (375 seconds)

Published: Thu Apr 27 2023