Transcript of Create CONSISTENT CHARACTERS from an INPUT IMAGE with FLUX! (ComfyUI Tutorial + Installation Guide)

the best method for creating consistent characters just got even better you all LED my previous video about creating consistent characters with my AI character sheet workflow but you all had one question what if I already have a character can I generate all this from an existing image and the answer is yes you can here is how it works now you can input an image of your character's face describe the rest of their body with a prompt and the workflow will automatically generate a character sheet showing your character from multiple angles it'll also create different emotions that you can easily customize and place the character in various environments with different lighting conditions that you can adjust with a simple prompt all these images are then saved out allowing you to train a custom Laura of this character for example Aura will allow you to generate endless consistent images of your character using simple prompts you can now also load in style lauras into the workflow so whether you want to create vintage 2D Disney characters or transform yourself into a 3D Pixel Character any style is possible except for the Minecraft movie style that it didn't work so in this video I'll guide you through the workflow and in the end I'll show you how to install this free workflow on your own computer step by step like for the last version we're using a character sheet that displays the characters's bones in the open Poe format as our foundation this allows us to generate the character from various angles using a tool called control net but this time I've enhanced this workflow by incorporating a tool called pull ID pull ID extracts the fasal structure of your input image and into integrates it into the generation process resulting in very consistent outputs the entire workflow runs automated inside of com UI a note-based interface for AI models and while it might seem overwhelming for beginners I'll guide you through each step however I still recommend you're getting familiar with comi before working with these workflows because they are quite Advanced I've created three versions of this workflow so let's start with the flux version I asked on my Discord for characters that I could try this out with and I got this woman right here which I guess was generated in mid Journey so let's create a character sheet for her I just open my comu ey window and I just drag and drop the workflow into the com UI window this is the full workflow and we are going to work from left to right let's zoom in on the left corner here you can use this one to activate or deactivate the other groups you can see I activated them now so here they are but for now we don't need them so we can deactivate all of them and the IDE is pretty much that we generate a character sheet here and then all the other steps the upscaling the emotions and the example images will be generated after that automatically based on the character sheet so let's keep only the first group activated and give our character a name down here you need to decide if you want to use an input image we want to use an input image so this should be yes and this should be no if you don't want to use an input image this should be no and this should be yes down here the models will be loaded and you can double check if you have the correct ones here when you first run the workflow some of these notes might be read even though you have the correct model in this case you probably just need to click on the name here and just reselect it and next you need to load in a character here you can just drag and drop it in here and preferably it should be a frontal image of a face like a portrait style image and it should be well lit down here in the apply poll ID you can set the strength and I like to keep it at 95% % if you want to go for a more stylized look for example a Pixar character you should reduce that to maybe like around 80% or something next we need a prompt for the rest of the body and I've broken up the character sheet prompt into three parts first up here you probably don't have to change that in the second group here you can add the style of your image and I'm just adding amateur photography and shot on iPhone for more amateur style kind of look because like putting DSLR and professional photography there can create some fake looking plastic skin with flux and down here you put in the character prompt and this is the description for your character what I like to do to find the right prompts is to use an captioning tool like for example Joy caption Alpha you can just use that on hagging face for free drag and drop the image of your character right here scroll down and click caption and after a few seconds you will get a very very detailed description of your character so I use this very long description as my foundation for the prompt copy it over but I then like to shorten it a little bit and I also delete the parts of the prompt that are not necessary for example the background and all this stuff we can also describe the parts that are not in the image so we can say for example she's wearing jeans and brown boots when we're done with a prompt we can go to the control net setup you don't have to change anything here you just need to import the post sheet right here finally here are the sampler settings and you don't have to change anything here if you're using a version of flux Dev so that's pretty much it we can just now click Q prompt and you can see this worked really really well though the boots on this one look a bit different but that's fine if you're not happy with your result you can try to adjust your prompt or you could also try out different seats for some slightly higher quality you can also increase the steps for example 235 once you like the image that you see here you can go back to the beginning here and activate all the other groups and make sure when you activate that last one here to deac activate the second yes here again so now we can just click Q prompt and the workflow will pick up where we left it two steps are happening in the second group the first one is an upscaling step and this will already increase quality by a lot if you have the time you can also set this to four if you want but I recommend two and in the second step in this group it will go through all the individual faces and just generate more detail for them if you want to save out all the individual faces as individual images from this workflow you can also activate this note down here using control B and if you then click Q prompt you can see you now have all the individual face images here and you can select the best ones for Laura training for example this next group here will just save out all the individual poses automatically in the next group here it generates different emotions for your characters and I just put some example ones in here you can super easily change them by just adjusting these sliders here let's say say we want to have her look in the other direction so I just turn that around let's make her blink and smile and I click Q prompt and you can see it changed the expression immediately in the final group here it will generate some example images of your character in different environments and you can change these environments by changing the prompts under the images here if you're not happy with your results one thing you could do is just come down here to the left corner and change the seat the this will then generate a completely new set of images and in the end you can just pick the ones that you like the most oh these look really cool these four Images will be based on the emotions that you generated here so you can see this is number one number two three four and five and if you want to change one of these emotions you can simply change that down here let's say we want to have her look down a little bit so in this image we could take emotion number four so select that from the drop drop- down menu here click Q prompt again and it will only change this image here and now you can see she looks down you can also select all these blue nodes here all the control Nets and click controlb it will generate new images of your character based on the prompt alone and this worked really well the advantage of deactivating the control net is that you can then prompt for specific poses like for example we C we could make her dance maybe she's dancing with her arms up in the air let's try that and it just worked and if you don't like these results you can always change the prompt prompt for more specific clothes here for example or try out different seats so this worked great with this AI generated character but now I want to try something else I want to turn myself into a Disney Pixel style character so for that I go back to the beginning of the workflow here and this time I'm using an image of myself next I'm SE searching for a flux Laura with the style that I'm going for I just download that go into my com VII folder and just put this Laura file in here in comi I click refresh and now I come down here to these Laura activate that by clicking contrl B and when I now click on this Laura name here I can select the Laura let's make that a bit stronger so maybe like 79% next I need to change the prompt so for the style and quality I'm going to put in in the style of pixel 3D animation high quality octane rendering Pixel Character stuff like that and for the description I keep it really simple I can deactivate the other groups again to see if the character sheet looks cool Q prompt and this is looking pretty good so now I can activate the other groups Q prompt now it's done and it looks really really good like the poses look good and I'm always amazed how well the faces come out even for stylized characters and the example images and different lighting conditions turned out really well as well now if you want even better quality you can get the advanced versions of these workflows on my patreon there I included upscaling setups for all the important parts of the workflow it will for example scale up all the lighting images the different emotions and you can even upscale all the individual images of the character's face from the character sheet giving you maximum quality for Laura training as I said these workflows are very resource intensive and take a lot of time to compute so let's now continue with the free fast version for flux I simply drag and drop it into the workflow and you can see it pretty much looks exactly the same but the models are different for this we're using ggf models for flux and you can get them inside of the comi manager just go to the model manager and search for flux and here you can see all of them now there are a lot of different versions and it's hard to pick the correct one one for your system you can check out this flux ggf guide on civit AI that I will put in my video description and here for example you can see the version number and how fast it is and how much you're paying the price with quality so I want to check out the Q4 version where the inference speed is supposed to be very fast but the quality is only acceptable so let's find out what acceptable means so I download this version right here by just clicking install once that's done let's also download the clip encoder with ggf so I'm also installing Q4 here once that's done just click refresh and I'm just going to select the model here and the clip encoder let's switch that out for this one now we can do even more to speed up this workflow so I'm downloading this alimama creative flux turbo model right here and you can find that link in the installation guide and put it inside of comu I model luras and I like to rename this one flux eight step just so I know what it is in comu I click refresh and use this Laura loader right here to select it so let's run this thing and compare the outputs well I mean you can kind of see me in there and it's looking like a Disney Pixar character so I'm honestly surprised that it worked so well we can also maybe try the moderate version here that is supposed to have good quality and the best balance between performance and accuracy and yeah okay that's definitely better that's progress I also created an sdxl version of this workflow and you can produce some pretty amazing results with this one as well now I also want to try to turn myself into a Pixar character I'm using the wildcard turbo model we don't need a Laura for this and this already works at eight steps so it's also super fast for the sdxl version the pull ID seems to be a bit stronger so for more stylized characters I like to reduce the strength a little bit we can also use a negative prompt with sdxl and I already put one in here but the structure of the workflow and how it works is exactly the same compared to the other versions so let's just click Q prompt and this only took a couple of seconds it's looking pretty good and it's done and honestly it's looking pretty good and also I think the lighting tests look really good as well so definitely don't sleep on the sdxl version just because it's older now let's train some luras with the data sets we just created for this I'm going to use flux gym and the easiest way to install flx gym is via Pinocchio it's a oneclick installer for a lot of different AI models and I'm going to start up flux gym and let's start with a Pixar version of myself that I generated with flux Dev first of all we need a name for our Laura and I'm just going to use mixar and I'm also going to use this as my trigger word the trigger word should be something that is not a commonly used word I want to train flux def so I select this one I have 24 GB of vram but you can get as low as 12 gabes I've tested that it will take a longer time but it works that's all we need to do here and now we put in our data set so let's collect all the images that we want to use for training and I definitely want to use all the full body Images I want to use all the different emotions and I definitely want to use some of these lighting tests here so let's select the best ones I think this is looking pretty nice I like the subsurface scattering in this image here so let's take that one as well that is good too and this one is great as well this is already more than enough but I like to go to the Head poses that we generated here and what I like to do is I like to add at least one like profile view we can just drag and drop that in there already added the trigger word to all the images here and next we can automatically caption these images with Florence 2 and this is pretty fast so after Florence 2 was done I just made sure that I added like the perspective of the image so for example here I I want to make sure that um flux understands that this is a back view of the character this is a profile view I added the lighting conditions and I also added like the emotions for these uh images here and yeah the lighting conditions here and now you can just click start training when the training is done you can go into your Pinocchio folder go to API flux gym outputs and there you'll find all your luras there are different versions so it saves out different versions as it's creating these luras and we want to use this last one here go into my comu folder go to models luras and I put it here and now in comy ey we can click refresh and I load in this workflow this is just a standard image Creation with flux workflow so in the beginning here we load in the models just like for the character sheet and here we can load in two luras one is deactivated let's keep that that way and next I want to load in the mixar model so I use a prompt like this a young blonde man named mixer remember we need to put the trigger word in there too so mixer walks through the mystical forest with giant mushroom let's cue the this prompt and this is looking pretty cool I've also included this noise injection setup from last video again so this will just add a little bit more detail to the generation helping especially with like cloth and hair when you like an image you can also activate the upscaling setup right here using contrl B and just let it do its magic so now that you know how to use these workflows let me show you how to install them on your own computer to help you follow along I also created this free guide so you can double check if you have all the correct models and put them in the right locations first you need to install a comi and unfortunately while I was making this video comu I released a new version of the tool that uses a new python version that is not yet compatible with the workflow so make sure to use this provided link to download a compatible version if you already installed comi in the past you should be good so click the link scroll down and download this ZIP file here while this is downloading you can already install git click on this link and download the 46 bit version I've already installed it but you just need to follow the default installation steps once it's downloaded you can right click and extract all once it's done you'll find this folder in the extracted folder and I'm just going to rename that demo this is your comfyi directory and you can put it anywhere you want on your computer next you need to download the com view I manager just click the link in my guide scroll down and here just right click save link as and go to your comi directory open it and save it inside here to install the comi manager simply double click on that file and congratulations you installed comi but since this workflow uses pull ID we need to do a few extra steps for example we need to install pH X lip for this go go to your python embedded folder inside your confi directory and type in CMD into the address bar and click enter and this will open this CMD console here you just need to copy in this command and press enter when you see this line you know that it's finished and you can close the command window next we need to install inside phas for this we can stay in the python embed folder and I want to run the python application to double check if we have the correct version this will open another window and you can see here we are using python 31.9 close the window again and click on this link here you need to download the file for our python version and since we are using 3.11 we need to download this version here 311 once it's done we can come back to our main confu ey directory and type in CMD here and press enter now you can copy the first part of this command up until here copy and paste press space and now go to the downloads folder or to the place where you downloaded the inside Face File right click copy as path and put it here finally press space again and copy this our nnx runtime and press enter and once again when you see a line like this you know that it's done and you can close the window finally you can start com UI by clicking run Nvidia GPU and com UI will start in your browser to load in my workflow you can simply drag and drop my workflow file into the com UI interface and you can see that it loads it but there are a lot of different nodes missing but to install them we can simply go to manager install missing custom nodes select all of them and click install and once it's done just click restart this might take some time as it's finalizing the installation process for the notes and once it's done you can see the workflow is here and we don't have any red notes but before we can use these workflows we need to install the missing models you can find a list of all the models you need in my guide but the easiest way to install them is not to manually download them from here and put them in the correct folder we can actually do that in comu ey so let's go to manager model manager and search for flux Def and I'm going to use this one here as my main model so just click install once that's downloaded let's download the clip models and we need two of them and we need this one here and we also need this one here and install the them both next we need the vae so search for vae and install this one here next we need the control net model next we need an upscaling model and you can get this here click on the link click on Download Here download this model go to comu ey models upscale models and put it here next we need the pull ID model and for flux just use this link here click download then go to your main directory comu ey models and now you need to create a new folder call that pull ID and put the model into this folder next you're going to need Antelope V2 so click on this link download the zip file and while that's downloading we can already create the folders for this one we need to go to comi models and create an inside face folder so type in inside face go into the folder and create another folder called models and finally inside of that folder create a new folder called antiope V2 and extract the files from inside the antelope V2 zip that you just downloaded into this folder finally that's all you need to do for the flux version when you first use this one of these workflows you might need to reload them so just go to the load diffusion model and click on it and select the correct version Oh and if you want to see previews while generating you can go to manager and I'm using latent to RGB so that while it's generating you can already see if your prompt is working I hope you enjoyed this video and this workflow is useful to you if you would like to support my work and gain access to the advanced workflows and exclusive example fights like all the data sets character sheets luras and prompts I used and created for this workflow consider supporting me on patreon your support makes this channel possible so thank you very much and see you next time