Easy Consistent Character Method - Stable Diffusion Tutorial (Automatic1111)

fictional girlfriends have been popular since the dawn of time we started drawing them on the walls of our caves then we began carving them into Marvel statues and now we can just create them through a few methods using generative Ai and this tutorial will be showing you a workflow for creating your own AI girlfriend so you can drive the Boomers mad and drive the eag girls out of business at the exact same time this workflow is based around using purely prompts to drive a consistent character and it's partially driven by luck in terms of getting the right detail but it's also one of the simplest methods that doesn't involve delving into complex tools to get a good result ultimately this is just one of many techniques that I'll upload so if you aren't a fan of this method subscribe hit the Bell icon so you're alerted when the next video lands but enough chitchat let me give it to you bite-sized now there are a few things you will need to replicate my results I'm using the absolute reality checkpoint as it creates very realistic images and allows for a greater degree of variety compared to some other checkpoints that experimented with in the past I'll be downloading two ups scalers Ultra sharp and super scale as these are fantastic for realistic images and even though super scale is Superior in detail the ultra sharp can be good for anime based models I'll be using both the bad dream and unrealistic dream embedding which works well with the checkpoint to produce better results and is placed within the negative prompt luras are optional but I like to play around with them to push the realism and style off my images and I'll be using two instant photo for a more photographic look and dark light for better lighting I'll be installing after detailer which can help us control how our character looks as we're adjusting the in painting stage rather than the image generation stage and it can be useful for correcting artifacts in our image Haku IMG will be useful for editing the image to try and get that photograph effect and you can install this from the extension tab by searching for its name and to finish I'm using a clip skip of two alongside the MSE 840,000 V and I'll link everything in the description so this workflow will revolve around using a name to drive consistent faces and then using prompting techniques to drive additional detail for a unique look to do this we need to realize that stable diffusion can be very stereotypical when it comes to names if I were to use a name like Sophie we'll get a Caucasian person compared to using a name like Park which will give us some more Korean aesthetic or Ina which leans more towards the middle East or even babatunde which sounds more African I think this may be because stable diffusion has names associated with people and its data which allows you to reference celebrities and in this instance find the face associated with a name or a combination of names but unfortunately one downside of this method is that different names from the same culture won't necessarily result in a diverse set of faces as seen in this image well I've gone with a typical British name but we've got similar faces across different names and speaking of which you can also combine celebrity names to get unique faces whether that's through specifying their names as a prompt or in my case using prompts alternating to switch between the specified names every step you can see how we're getting some blending and you can add the celebrity name twice to increase the prominence of that person leading to more unique looking faces while maintaining that consistency once you decide on the name and the face that you like we can then focus on prompting in those additional details I'm using some standard prompts for Quality such as photo realistic and raw photo and some negative prompts including open mouth so our character's mouth is closed you can change these to suit your preferences now let's start with after detailer as we can use this to drive some adjustments to our character at the in painting stage which can help us get a more unique face I'll be using face eye and hand model so we can imp paint those areas without manual work on the default settings the face and eyes will use a separate width and height at 768 and in the faces prompt section I'll add in soft lip skinny and filum for that dip beneath the nose now moving on to our prompt box I'll add in Asian white within square brackets with a pipe between them so every step it will switch from Asian to white as we're alternating the prompts I've opted to go for a white jumper gray pleated skirt and in terms of the hairstyle I'll go for curtain bangs now something you'll probably notice is that we're getting actual curtains in our image due to the name of our hairstyle and the fix for this is to delay the implementation of the hairstyle with square brackets so we have less steps to implement that prompt over the image generation's total steps now will be a good time to add in the background which will help further remove the curtains and I'll add in a modern neighborhood background now another problem is that our character is appearing sharp and looks like she was photoshopped into the image a brilliant trick for fixing this is to delay your entire prompt using the square bracket method delaying everything by about two steps by adding our background at the start of our prompt so we generate our background first and then the character and everything else afterwards this technique is amazing as it will absolutely change the composition of your image helping your character feel like a part of the scene you can also add in a fisheye lens prompt to somewhat distort the lens which can help add some visual interest to the image lastly I'll add in those two lwas I downloaded using a low waiting so we can see whether or not they improve the overall image I'll also add Asian into the negative prompt as the instant photo Laura can make your Generations look more Asian by default and I wanted to counteract it a bit now at this stage if I generate a series of images you can see that we have that consistency while also keeping those extra details we specified and this is all driven by that name we provided and if you find you'll get in the same face despite by changing the name try lowering the after detailers the noising strength for the face model I also decided at this stage to download an additional checkpoint called realistic vision and I use this with the refiner at 60% to help bump up the realism and at this stage all that's left to do is prompt in the details you want and generate a series of images until you get an image you like there will be limits to how much you can control through prompting alone especially regarding the color bleeding and I'm ke to keep this as a prompting Focus method rather than bringing in control net and other tools as this will drastically change the results and increase the levels of complexity now the only thing left to do is to take an image you like and run it through a few filters to try and replicate a photograph effect you can do this in or any photo editor of your choice but I'll be using hacku IMG as it's simple and can be used within automatic one different photographs have different looks but a popular way to trick the human eye into thinking that something is a real image is through imperfections such as film grain playing with the exposure and adding some blurriness I personally like to add a bit of exposure noise and maybe some contrast and a tiny bit of blur to lessen the sharpness and then I'll send the image to the extras Tab and upscale it by two using the same upscaler and that brings us to the end of this workflow which was hopefully easy to follow and connect us a great starting point for generating images before bringing in other tools like control net a big thanks to the supporters and if you found this helpful drop a like And subscribe this is bite-size genius and I hope you enjoyed

Channel: Bitesized Genius

Views: 30,845

Keywords: stable diffusion tutorial, stable diffusion tutorial for beginners, stable diffusion tutorial video, Stable diffusion tutorial 2023, ai artist, a1111 tutorial, stable diffusion ai art, controlnet automatic1111, stable diffusion online free, stable diffusion gui, stable diffusion browser, create consistent characters stable diffusion, consistent characters stable diffusion, stable diffusion prompt guide, stable diffusion tutorial anime, Stable diffusion tutorial img2img

Id: 4KaTyjSK03A

Length: 7min 39sec (459 seconds)

Published: Tue Dec 26 2023