GUIDE: How to Use NovelAI Image Generation

hello everybody and today I will be showing you how to use novel AI image iteration you are going to need two things before you even allowed to generate images you're going to need a novela account and you're going to need anomalia subscription of some kind uh I personally have scroll because I can't afford opus but there are some benefits to using Opus which I will discuss later in Syria you want to go to this little palette icon and click on it in order to get to the image generation or you can click this big button right here that says image iteration it does not matter because it will bring you to the same exact place which is novela.net image once you learn into this you will notice a couple things there's a prompt box that I just backed face something from there's a paint new image button there's an upload new image button and to the right screen there's an option to select your model there's image resolution settings there are also mod specific settings like the ability to turn off around quality tags and there's also the ability to buy analysts which I will get into what those do and how those function in a second or rather later video times relative you know uh anyway so we're going to talk about prompts first there are several ways you're approaching prompts first off you can just rely on the pre-programmed pre-recognized tags that are in the program such as the word Valley as you can see the AI has now generated a picture of a valley and in this case the viewer is looking out into the valley uh now you don't really just have to rely on the tags that suggests for you for prompts and you can get a lot more complicated for your props to this value for instance if I added the tag Forest the image will now have uh uh more focus on having force in the image so before I just had a Valley now it has a forested value more like but you can get even more complicated with this if you want to now just because your tag is not recognized does not mean necessarily you have to follow those tags you can really type anything in this box and it will likely suggested while we're on tags though as you can see these gray circles have various levels of intensity a darker gray means it more to know more about this subject so as you can see here it knows a lot about pink hair but not so much about stuff like pie so you can keep this in mind when using these uh tags that it recognizes if you're having trouble with the AI not really focusing apart on party image like you'd like it too you can just surround the tiger where you use with Open brackets which are usually located next to uh the mice and equal signs on most keyboards if you're using a mobile device it'll probably be in secondary uh keyboard menu and what that will do is it will focus on that aspect more than it otherwise would and I don't think there's hard limit to this as far as like I know from reading documentation there is no one metallic curly brackets to put on it but honestly with a lot of as of a lot of things in AI using too much may be Henry's now you may notice that I put my mouse cursor on this little long number that's right here and this is what's called the seed the seed is a numerical representation of the exact way in which the AI made your image so this means that if you want to go and reproduce an image that you need with the same exact scene and same with lab prompt you can do that for instance if I were to copy the scene and have this exact prompt and put in the seeds field it'll produce the exact same result so to show that what I'm going to do is I will type a completely different prompt into the prompt box such as uh pumpkin and now obviously pumpkin but if I were to go back and put the prompt that I had before into the power box with exact same seed boom that's the exact same amendments that we generated before now see now obviously you can see with the function the seed boxes so you can go back and previously uh prompts if you want to with the scene in the exact prompt uh now no if you if you uh type anything other than the exact scene and the exact prompt you will not get the same image you'll get a different image you want the same scene modifying the prompts a little bit the same C that you used last time or probably not result in the exact same image and it will probably change it a little in fact the further away you get from The Prompt when using the seed the more it will differ from what you originally generated since the sea is more of more reflective of the composition of the image than anything now I'm going to go over these models that novel AI has there's uh Navi in a full which I'm using because like uh using it there's no real reason why I picked the first video and then there's curating and then there's furry now all of them have their jam stops without anime full will allow for a wired wired unit uh Generations like it says there um and it's basically it's the full data set of what novel AI has compiled in this image model um curated is a lot more limited and it has some issues like if you're trying to do something really specific it'll probably not recognize that but in the full data set it probably will recognize it so essentially if you're having problems with curated to switch to full and now for furry well that should be obvious if you don't know what that means that's not the topic of this video so just go open a new tab search furry and come back to this video and now I'll move on to something more simplistically understand which is the image resolutions now uh image resolution will affect a lot of stuff it will affect honestly how large the image is but it also affect the composition of your image and how much analyst it costs to do all these things speaking of endless cost Opus users at least as of recording this video can have any images in the large category a large drop down a large subheading sorry jury duty completely for free but that's only for Opus users and everyone else will have to pay the full endless price for generating these images uh in order to demonstrate the effect that your image resolution has on uh your composition here is the same prompt that we've been using for the past couple of minutes with the exact same prompt down to the lever and also the same seed as you see the composition is now completely different because we saw SWAT resolutions uh it's more of a neighborhood type thing right in a single cabin with this kind of composition and it's totally framed differently because of the resolution so essentially this means that with different resolutions even with the same same prompt you will have a different image so be sure to know that now as you can see I'm sliding this little image uh accounts slider thing and depending on your resolution you can generate up to a maximum of four images per gallon uh I've tried to generate really small images and even that then uh you can only generate four at a time now I'm going to show you what four elements of time looks like and now we have all our images here uh all the images will be uh aligned four in a row with this resolution and such um all the images here they enumerate the scene of the one on the left by one the further right you go uh though during four memories of that time it's cool now you do want to know that the more the more images you request and the larger your images are the more analysts that you'll use up I'll get further into analysts at a later point in this video so stick with me as you may or may not have noticed by default there's an undesired content drop down which automatically selects low quality plus bad knot and you'll see undesired content undesired content just means exactly what you think it means it will steer the AI into not generating what you put in there uh this could be anything you want to be if you uh need to for example you can just type in like uh watermelon or something then they will avoid generating watermelons and uh down here it's it says add party tags with a Target this this appends masterpiece and one other quality tag to your image sorry not appends uh puts at the beginning which is not opinion uh obviously you can turn it off and on they're not going to force you to use this if you don't want to uh as you can see there's something called steps right here and essentially in the simplest form that I can explain it this is how many times that your AI reiters upon your image upon first generating it uh more steps can mean that it will get better but to a point like putting 99 or something like that will probably not be beneficial it'll probably just look the same as like 35 steps and so on scale is the degree to which the AI obeys what you put in the prompt uh lower scale would mean like through the letter by the book but like a higher one will mean creativity more creativity sorry and you may want that lower or higher depending on what your prompt is so you should really experiment around with uh steps and scale to figure out what you want for yourself uh down here there are Samplers now you do not have to worry about this as a casual user because the difference in these Samplers are very small and the novel AI team has decided that KU are ancestral is the best sampler to use for most users you should really only Tinker around with this if you really know what you're doing or you just want to if you're curious as to what all the different Samplers do which is a very hard to spot most of the time so you do not need to worry about that as a category user now you will notice that there's several things up here there's variations and there's an enhance which you can change settings of now for enhance this uh you can change the magnitude which I think it's probably self-explanatory this is an uh blows up the image and the magnitude which is how how much it changes the images in order to enhance it and like everything else this cost and less to do so be wise uh depending on the image it will just try to improve certain parts of it with some honest some automatic method which is not technically detailed in any of the documentation that I've seen uh I'll show you one example of enhanced doing its job as you can see here it did indeed enhance the image there's quite a few subtle but good differences between my original image in this image uh and ultimately in the end you're the one who decides if you want to use enhance or not and whether or not you think it's worth using analysts to enhance these images uh now on to editing images this works in much the same way as painting a new image which is why I will not actually show you how to use this because it's a bit redundant to making a new image uh hitting the variations button right here does essentially the same exact thing as the sharing parameters so I will not even show you what generating that many what generating damages these variations looks like now on the painting image this is where it gets interesting so this is as far as I know very rarely done when it comes to aim regeneration generators it's essentially just a normal painting program I mean you got color picking uh color selection on doing redoing uh selecting a racing and obviously drawing uh and you can also turn on uh touch tablet mode which obviously is not applicable for PC you only need that um but what sets this apart from just you know a normal editing program is that this will take in this is going to be used by the AI to generate something new from whatever you draw here I'm going to just make something really basic with this that actually really does not show off the tech that well because I'm not really an artist so I'm just gonna do like uh the sun over the ocean and you need to make sure that you uh keep in mind to describe what exactly is going on in the image that you just drew so the DNA actually draws what you want so for example uh mine my image is I'm a son of the ocean so in the prop box I'm going to describe it as such using the relevant keywords and such now I do realize this is not the best example for selling this off but it did indeed change my image with what I gave it as you can see the ocean is not no longer just a flat color the AI did add some wavelines into it and it did change the background and transparent height this is more of my fault that it did not change that much if you actually try to make a decent image it will change it a lot more and it will have more of an impact than you know just a really basic image of the sun over the ocean doesn't even really have a background now of course like everything else is cost analysts and uh you know I said earlier in the video now I'm going to show you how uh image you image or a below damage Works uh you didn't see it by selected image from my computer and just like painting New Image I will describe what is going on in the image plus any anything else I want the AI to do with my image into the prompt box you know it has some guides as to what I wanted to do with that image this photo I'm using is a black and white photograph of Jorge Luis bores so I am trying my best to describe that and also note that it's in black and white to that it kind of matches what I put it but also reiterates on it because you don't want it to be exactly that damage now as you can see it drew a pretty Dapper young man and a Sue and die which is basically what the original images but of course every it does not match it exactly because you don't want to match exactly speaking of there are these uh sliders down here strength is a measure of how close uh the AI will try to fall in the original image I do not recommend lowering this than what Ari is by like a lot because what ends up happening is that the the image will be very poor quality at levels like 0.1 stuff because they were just trying to match the image but also have a little bit of AI generation in there and in the end it'll really look distorted and blurry uh noise it's just a measure of how much detail the AI is allowed to generate putting this too low might make it smudge some details but putting it way too high can also caused some weird artifacts and some weird patterns to show up in the image so you want to strike a bouncer uh now I'll show you a little neat thing uh that the AI not only AI developers have called prompt mixing it takes the two tags that you type and combines the qualities or as they say vectors uh damage and you can do this by typing two tags and separating them with little uh long line that should be under your backspace key on your keyboard and what that will do uh we'll try to combine the hand dog now as you can see in my case the image I got did not really combine them that well but you can obviously see that it tried to put a little bit of a dog in this image now if you want to go on a documentation website which I will link in the description below there are way better examples like they did like a frog cat and it mixed together pretty well uh uh one other thing to do with this uh if you put a colon and then a number between I believe negative one and a hundred you can weigh the aspects of the image differently what I mean by that is you have the AI focus a lot more on the cat party image for instance then the dog part of the image which will create some interesting results as you can see that's way different from what it generated before the cat is outweighing the dogs weight so it's more cat-like than dog-like now I'm going to move on to a few UI based things that have nothing to do with every generation but will help your experience on this right side you may have noticed that all your images that you generated are showing up now don't worry this service does not store your images in any way in fact if you refresh the page all these images will disappear completely and there will be no way to get them back but uh you can clear individual images from your history because let's say I don't know you during something kind of Saucy or a bit risque and look someone looks over your shoulders like oh crap you have a really weird image on your screen well you can just come over here to top right image click the little gray X and that duration will be deleted It'll ask you are you sure and press delete it and it will delete it uh and obviously you can click on any image within this history bar to go back to it and uh go back and edit it put a variation on it etc etc here's another neat little trick that you might find useful you can use the shift key and Enter key to add lines to the prom box like I just did Ctrl shift enter on the other hand makes new lines that are intended for prop mixing so you don't have to type the little long line every time another sub trick having to do with this Ctrl shift enter thing is if you use uh a tilde which is a little squiggly line that's on the very not a very top of near the top left of most keyboards uh it will repeat the prompt plus any text you have in this newly created box that you created with with Ctrl shift enter so I'll show you that I will generate an image real quick uh now obviously it just repeated your problem because there's two cuts in the image now on to the thing that I've been wanting to talk about for the entire this video obviously as everyone knows AI is not free nor is it cheap is not cheap to run it's not flavor meaning that the developers have to make some kind of money somehow and I mean you already have to pay a subscription to use not only in the first place so you should probably have no issues accepting the fact that you have to pay for analysts or just wait for your subscription to automatically fill back up your endless now if you go to the top right you can see that you have multiple options for buying in this uh there's a 379 option as uh and this buy is that many analysts there's 649 option and 10.99 option for really eleven dollars now I used a lot of paint steps but usually for uh tablet and scroll it's a found steps to put them on for free and you have to buy the rest aimless are used for basically everything in existence when it comes to image iteration unless you're open so much to some image resolutions will and need to be free but for all everyone else uh the image resolution will affect your analysts cost number of images roll factory analysts cost uh trying to get variations we'll also use analysts editing the image will use analysts enhance will also use analysts and everything you can think of Minsky's Atlas the only thing that does not use any analyst whatsoever is the downloading of images and just UI functions in general do not use any analysts because I mean why would they speaking of downloading images you you have unlimited image downloads and since that's all there is to say I'm going to end this video by telling you that I will be putting a few helpful links in the description there'll be obviously the novela website uh link to the novel I Discord which has a bunch of helpful moderation developers who actually engage with the community and uh are willing to answer questions and such all along and then there's a link to the documentation for the Amber generation which was made by novel am themselves so if you have any further questions or Curiosities then you should go there I recommend checking out the prompting section because I did not explain it near as well as they do in the documentation and since that's everything that I have to say about this see you another time and Happy generating

Channel: AppulMan

Views: 48,318

Keywords: NovelAI, Image Generation, Artificial Intelligence, ai tutorials, full tutorial novelai image generation, How-To Guides, Tutorials, novelai, novel ai

Id: JQVNYvlfQAg

Length: 25min 42sec (1542 seconds)

Published: Thu Oct 13 2022