How to make an AI Instagram Model Girl on ComfyUI (AI Consistent Character)

welcome to Economist today I've got a super exciting tutorial lined up for you by now many of you might have heard about AI models making the buzz on Instagram what sets aana apart in the sea of thousands of AI accounts the answer to that question is it's her consistent face while switching outfits and locations and that's exactly what I'm going to teach you today we're going to create our own AI model from scratch and will guide you step by step to achieve similar stunning results don't worry if you're new to comfy UI this method is beginner friendly and easy to replicate before we dive in remember to check the description box for all the resources used in this tutorial if you're new to comfy UI I recommend watching our video on setting up the program and installing comfy UI manager for those already familiar with comfy UI make sure to update the program upon opening my workflow you'll get a prompt to install missing extensions just open the manager click on install missing custom nodes and install them all and restart your UI if you haven't subscribed yet please do also smashing that like button really helps us out let's beat that algorithm together now let's get started and create something [Music] amazing first up load a default KFI workflow we're aiming for a super realistic output so we'll use The Cutting Edge realism engine sdxl checkpoint model you can find the link for this in the description below for our character we want a woman with blonde hair next up in our K sampler settings we're going for 25 sampling steps and acfg of s for the sampler we'll select the DPM Plus+ SD e caras we'll set our image Dimensions to 1024x 1024 pixels for a clear detailed view time to generate our image although it's great we need a closer front view for more detailed facial features so let's generate a few more [Music] images to simplify this tutorial we'll generate a face you might find familiar don't worry the final character won't be an exact copy but will borrow General [Music] features this one here is great for our needs with its clear lighting and features now let's up scale it using ultimate SD upscale to do that we need to connect the character image next we add an upscale model loader after that we select an upscale model I'm going to go here with 4X full Hardy let's connect it with ultimate SD upscale we do the same for the model the positive the negative and the AE for the upscale settings 20 sampling steps 5 CFG use DPM Plus+ SD cares again I'll adjust the danway strength to 0.2 and set the tile Dimensions to 768 pixels now let's start the upscaling [Music] process and there we have it our character's face reference is all set moving on let's take our character creation to the next level we'll be using an IP adapter combined with the control net method this awesome combo is key for achieving higher Precision in our character stability and consistency it's like giving our AI character a superpower to remain consistent across various images start by adding a load image node this is here our Target outfit that we want our character to wear it's a long sleeve shirt in a denim [Music] skirt next we need to load also our reference character face what we need to do here is to add two IP adapter apply nodes and then connect them with an IP adapter model the first one will use IP adapter plus face and the second one just the IP adapter plus next we'll add a clip vision and we connect it to both IP adapter apply node let's connect our character face image with the first AP adapter and the body with the second adapter don't forget to connect your checkpoint model with the first IP adapter apply to now let's chain connect our IP adapter's model output after that we'll move to add at ttn masks copy the outfit image right click on it and select open in mask editor here we want to mask only the face and hair area of the image next we do the same for the body area [Music] mask now we connect the the masked images with our IP adapter apply nodes and we need to lower just a little bit the weight of the IP adapters to give freedom to our model I guess 0.85 is enough for both for our checkpoint model I'm going to use an SD 1.5 model epic realism will work just fine it's optional but if you want to add Aura you can do that too I'm using epic realism helper Laura in my case let's connect the lur loader now with the IP adapter instead of the checkpoint loader also we connect the IP adapter model output to our K sampler all right we're going to stick with our current prompts for now we'll Circle back to them a bit later but first let's tweak our image Dimensions to create a vertical portrait setting it to 512x 768 pixels will be perfect for what we need now it's time to introduce control net into our mix let's add a control net apply node to our workflow this node will be the bridge connecting our current clip text boxes and our key sampler next we need to load a control net model for this we're going to use open pose it's fantastic for understanding human poses I've got a specific pose in mind that we'll use so let's load up in an image loader node let's type a random seed number here and change it into fixed let's increase the sampling steps a bit this will help us refine the details of our image at the same time we're going to reduce the dno strength to 0.75 we need to deactivate the upscaler temporarily to do this simply hold down the shift key and select the nodes you want to deactivate then press control and B on your keyboard okay let's hit that Q prompt button and see what we've [Music] created and there it is as you can see the IP adapter has done a fantastic job in keeping our character's face consistent but take a look at the outfit it's not quite what we were aiming for no worries this is a simple fix we just need to tweak our prompt a bit [Music] by describing in more detail what our AI model is wearing we can guide the model towards a more precise representation of our character's [Music] outfit awesome now we're starting to see images of our model with not only the reference face we chose but also the outfit we envisioned but why stop here let's generate even more images until we get perfect representation of our character a simple trick to get a variety of results is by changing the seed number each new seed number brings a new possibility almost like rolling the dice for different facets of our character when you're satisfied with your character image it's time to upscale it reactivate ultimate SD upskill using control plus b key now click on Q prompt button to generate a highquality [Music] image great now let's change only the pose to see if our character will maintain the same face features and the same clothing and just like that you can see the magic of our workflow while while the pose changes our character remains wonderfully consistent this consistency is key to creating a believable and engaging AI character let's upscale the image to get an even better [Music] look and here's another cool aspect by tweaking the positive prompt you can add a whole new range of emotions to our character want to see her smiling looking worried or even shocked just adjust the prompt accordingly you can also change the background let's place our character in a living [Music] room get get ready for this the possibilities with our AI character are truly endless it's like giving life to your imagination in our next example let's switch things up a bit we'll change our character's outfit to gym clothes and transport her into a gym setting picture her in a dynamic pose sporting in different outfit but with the same consistent recognizable [Music] face this is where the real fun begins we're not just changing clothes or scenery we're crafting a whole new scenario while maintaining the essence of our character it's like she's living different chapters of a story and we're the authors that's it for today if you find this tutorial helpful smash that like button and if you're not subscribed yet please do thank you for watching and see you in the next video [Music] oh

Channel: Aiconomist

Views: 72,578

Keywords: ai art, stable diffusion, stable diffusion tutorial, stable diffusion roop, stable diffusion inpaint, midjourney consistent character, stable diffusion controlnet, stable diffusion img2img, ai consistent characters, consistent characters stable diffusion, inpainting stable diffusion, ai influencers, ai influencer, ai instagram model, ai face swap, create ai influencer, ai tools for content creation, how to create ai influencer, faceswap, hyper realistic ai images

Id: HPSu6BWTHlc

Length: 12min 8sec (728 seconds)

Published: Sat Dec 16 2023