Stable Diffusion Tools: Master the Art of Stable Diffusion

today's video is brought to you by tensor art tensor art is a free stable diffusion based AI image generator you may have seen image generators like this before and felt a bit intimidated by all the tech lingo I'm using tensor art to demystify all that text speak and explore the magic behind the AI image generation process today's guide is for beginners so open up tensor art and follow along make sure to use the invite code making the photo together we'll talk about models and Laura's sampling methods steps and scales all those things that make stable diffusion based image generation so flexible by the end of this video you'll be a pro you'll know what the options do and if they are the right fit for your AI generated images tensor art is sponsoring this video and you'll get a good look at the many tools that are at your fingertips once you wrap your head around the tools you'll see how stable diffusion models can make more personalized works of art if you're new to stable diffusion it's an open source AI technology that creates images from text prompts you give it a text like a sunset over the ocean and it generates an image it uses a process called diffusion to generate the images diffusion is a process of adding noise to an image and then gradually reducing the noise over time stable diffusion is open source this means that anyone can access and use it stable diffusion forms the backbone of many AI image generators like tensor art there are a lot of ways to personalize stable diffusion but some of those powerful tools aren't very intuitive for a beginner once you know how to use the tools you'll find stable diffusion to be flexible and it gives you a lot of control over your images let's jump in with models stable diffusion version 1.5 or 2.1 are base models the newest version is stable diffusion XL but a lot of creators like the ascetic look of version 1.5 the base models are a good place to start but what are these other models offered by tensor art to get a specific style or subject create a fine tune or create a specific model for different types of images for instance if you like to create vintage cats you train a model on a bunch of pictures of vintage cats now whenever you prompt you'll get back vintage cats if you prompt for a high fashion model you'll get an image that includes vintage cats anyone can train a stable diffusion model and many creators share their models for you to use some models are better at generating specific types of images such as Landscapes or portraits and tensor art has dozens of these models some of the most popular focus on creating realistic images anime cartoon and fantasy images popular models are dream shaper anything and open Journey choose a model that matches what you're trying to create if you need a model that is realistic then you may want to use a model that was trained on a data set of real world images if you want anime choose a model trained on anime style images but bear in mind that many models are created in response to not saved for work filters creators want to make edgier images with sex or Gore these are often blocked by the mainstream AI image generators like mid journey and Leonardo AI speaking of fine tuning let's talk about Laura's if you've tried generating images and ended up with unrealistic picasso-like results with weird hands and face distortions then Laura's are your new best friend they're small files that tweak details in your models there can be loras for anything poses clothing emotions art mediums specific objects one of the most popular Laura's in tensor art is add more details you can adjust how much of this fine-tuned style you want to add using an easy slider in tensor art you can even add more than one Laura and mix the Styles together vae is optional and usually improves fine details like eyes think of it as the icing on the cake that helps your images stand out with more vibrant colors and crisper details in tensor art the default is automatic vae so it's trying to choose one that will make your images look the best but you do have choices the names are very descriptive so you'll need to do a little bit of experimentation a detailer or after detailer enhances details especially in the face and hands it detects the faces and hands and then in paints or fills in any missing or blurry areas a detailer also uses a face detection model to identify and correct any facial distortions or artifacts sensitivity lets you control how sensitive the tool is to faces more sensitivity and it will be looking very closely for faces even if they're in the background and you have an extra prompt box if you want to make adjustments to the face like adding freckles or fixing this two-headed vintage cat in tensor art you can click a detailer on an image or open the a detailer tab in the prompt box you describe what you want to see but in the negative prompter box you describe what you don't want to see I have an entire video on how to use negative problems in stable diffusion there are some common types of negative prompts some of the most common have to do with body distortions for instance node deformed or disfigured hands or faces extra limbs you know that sort of thing and many creators use the same negative prompts on all of their images image to image means that you give the AI an image along with the text prompt to show the AI what type of image you want by dragging an image into the box you're saying to the AI I want my image to look like that without changing the interior design prompt I'll add an image with some colors and a brick wall notice that the AI picked up on the general layout of the room and some of the colors though it blanked a little bit on the brick wall denoising is a way of telling the AI how much you wanted to pay attention to the image one means to ignore the image and really do its own thing zero means to essentially replicate the image as closely as the AI can default in tensor art is 0.5 and that's a really good place to start so use low denoise if you just want slight variations to your image set it high if you want more variability I have a couple of videos on how to use images as part of your prompt lenet is a specialized form of image damage prompting it lets you capture a pose or a composition from an existing image it's a whiz at detecting edges or human body poses and fuses that information with your prompt you're not copying the whole image just the essential composition or body pose this leaves you free to change other image details upload an image with a pose or composition you want to mimic in the text prompt be specific about the alterations you'd like to see when you upload a reference image you can see the pose or details extracted from the original image popular control net models are open pose and Candy Edge detection but with tensor art you're a little bit spoiled for choice aspect ratio is the shape of your image the default aspect ratio in tensor art is two by three portrait but you don't have to stick with this you can make your image wide by switching to a landscape orientation or you can click on custom and use the sliders to specify an aspect ratio tensor art will make an image up to 1024 by 1024 pixel Square while we're talking about aspect ratios let's talk about the high res tool hi-res comes to the rescue when you're creating non-square images the AI prefers making Square images and a Default Resolution of 512x512 pixels want a tall or long image or one with higher resolution that's when you need high res fix essentially High roast fix first crafts a low resolution image and then it scales up this image to the resolution or the aspect ratio that you want this process significantly reduces the chances of oddities in your final piece like multiple heads or repetitive patterns denoising and high-res fix operates like it does with the image damage prompting it's your say on how much variation you want between the first low resolution image and the high resolution output the defaulted tensor art is 0.5 and that's a safe bet for starters as for the best upscaler there's no clear winner it's a game of trial and error to find your sweet spot for the next tools you have to know a little bit about what's going on under the hood of the AI image generator AI image generators sculpt images from noise the AI starts with the random mess of noise and chisels away at it step by step until we're left with a gorgeous clean image there are many ways the AI can make an image from noise the process used is called the sampling method in tensor art the default setting is a Euler a but you have other choices Advanced disabled Fusion creators will notice a difference between the sampling methods but plenty of stable diffusion users keep the default but if you want to change the sampling method here are some popular choices steps are related to sampling method it's the number of passes the AI makes gradually reducing noise at each stage more steps equate to a gradual noise reduction like sandpapering wood for a smoother finish it's precise but demands more computer power fewer steps are quick and efficient but the leap between stages might lead to some hiccups in the image generation tensor art sets you at 20 steps by default and that's a balance that usually yields good results the CFG scale is also known as The Prompt guidance scale you're telling the AI how faithful to be to your prompt it's all about finding that sweet spot between Fidelity to your idea and the quality of the image lower CFG values might give you a stunning picture that's a bit off prompt while higher values stay true to your prompt at the risk of a lower aesthetic appeal especially if you haven't used any style words in tensor art the CFG scale ranges from 0 to 30 with 7 being the typical setting consider the CFG scale as your creativity control it's your way of guiding the ai's imagination if the default setting of 7 isn't hitting the mark don't be afraid to play around with this slider to achieve your perfect balance when the AI kicks off the image generation process it Blends your prompt with a random image or seed for most creators the random seed works just fine if you run a prompt several times the image will be different because the AI is picking a fresh new random seed each time but if you get an image that you really like you can keep that seed lending a certain consistency to your future creations let's take a trip back to one of my previous interior design images I'll set everything up exactly as it was and copy the seed this time though let's switch to a different room because I use the same seed the rooms have a unified aesthetic making them look like they belong to the same stylish home remember that while most folks stick to the default settings you have the power to tweak these controls and experiment tensor art has all the tools you need to master AI image generation you should feel more comfortable navigating the robust toolbox that tensor art offers and have the knowledge to Leverage The Power of stable diffusion to create your unique art each tool and concept we covered today holds a key to personalizing your AI image generation and making it a true reflection of your unique Creative Vision so grab your newfound knowledge and open up tensor art and start exploring remember it's all about practice and exploration and not being afraid to make adjustments and try new things let us know your favorite tips and tricks for creating AI images in the comments below share your images and don't forget to tag us at making the photo this is Jen at making the photo let's make something amazing together [Music]

Channel: Making the Photo: AI Edition

Views: 43,443

Keywords: AI art, AI image, Creating stunning AI images, Digital art, Making the Photo, Stable Diffusion, Stable Diffusion negative prompting, Stable Diffusion prompts, Synthography, Wits End Photography, digital photography, how to use tensor.art, image prompt, negative prompts, prompt craft, stable diffusion ai, stable diffusion examples, stable diffusion prompt guide, stable diffusion tutorial, stable diffusion video, tensor art image2image, stable diffusion, ai art, tensor art

Id: zQjQbXJQuc8

Length: 13min 9sec (789 seconds)

Published: Thu Jul 20 2023