Regional Prompter Extension In 15 Minutes – Stable Diffusion (Automatic 1111 | Tutorial)

if you've ever struggled with color bleeding or images that are not accurate to your prompt then you will want to use the extension called Regional prompter what this amazing extension does is allow you to divide your images into regions and then Place elements of your prompt inside of those regions giving you full control over colors descriptions write down to printing designs on clothing this can be done through non-destructive templates allowing you a modular way or swapping out prompts without ruining your entire image I'll be breaking down how you can use it with practical examples so you can spend less time reading and more time creating hit that like button and consider supporting through the links below but enough Shilling let me give it to you bite size first I'll give you a quick overview on what Regional prompter is and some of the more general settings before we dive into covering each section with practical examples Regional prompter is an extension for automatic one which allows you to divide your image into regions and then apply prompts to those regions with a greater degree of accuracy if you're familiar with the after detail of extension then this will operate in a similar way but instead of impainting it would use different techniques to apply the prompts to a specific part of the image for example on the left I have an image of a woman without using Regional prompter where details have been captured but not accurately to my prompts the image on the right is using Regional prompter and as we can see it's a far more accurate image from the rimless glasses to the dark blue trousers and even the pattern on the blouse first you should note that regions are separated using three words in all capital letters you can use and break or one of the special phrases provided by the template we'll look at later on if you want to use and with the normal functionality of combining two prompts into one then you can enable the option disable convert and to break the first setting we have is called active and this will determine whether the extension will affect your generated image or not basically the on and off button 1 you have two generation modes to choose from attention and latent attention is the default option and in my experience gives the best results but latent has the ability to better handle lawyers which may need to be separated into regions and seems to work better when using highways fix avoiding a runtime error you then have use base prompt and use common prompt which sounds similar but operate differently so listen closely to this section use base prompt will take everything in your prompt box before the first break keyword and apply it to every region of your image using a weight determined by your base ratio so for example with this prompt and the base ratio of 0.2 we can see that the first section of our prompt is multiplied by 0.2 and is less important than the rest of our regions which will be multiplied by 0.8 as these numbers represent portions of one use common prompt will take everything in your prompt box before the first break keyword and apply it to the start of of each of your prompts separated by regions or now of example prompts it will look like this in practice where that common prompt is at the start of each of our regions but in the nutshell you should use base and or common prompts to apply a group of prompts to your entire image such as Styles quality prompts the background for the entire image or Canberra perspectives the base ratio as specified earlier controls the strength or weighting of the use base prompt and you can use multiple base ratios separated by commas to influence different portions of your region but be careful as this can get Messy as there's no visual indication on whether you selected the right number of Base ratios in relation to your prompt so you may need to experiment you also have the option use comma negative prompts which allows you to have negative problems which apply to your entire image while also having negative problems which apply to specific regions of your image I don't tend to use this as things can get out of control quite quickly but instead as an option and will cover it in The Matrix portion of the video the last option I'll cover before we move on to the Matrix Tab and some examples is the debug option which isn't well documented but I think this is exclusively for people who are using an API also known as an application programming interface so I wouldn't worry about missing out on anything if you don't use this option but now that we've covered some of the basic settings you have a whopping free option to choose from for controlling your images and we'll be going through them section by section our first option is Matrix and this allows you to create a grid with horizontal and vertical rows which you can then use to allocate prompts to portions of your image and this is great for creating compositions you have horizontal vertical and random split modes which determines the direction The Matrix is divided typically for portraits you will want vertical so that each portion of your character is nested within a region inside the Matrix horizontal is useful for if you want multiple people in an image and random is great for if you're using both horizontal and vertical values within your divide ratio as it will often default to the option which works best you then have the Divide ratio where you can add or remove divisions within the Matrix using commas to separate values you can also create splits within divisions by using a semicolon between values and the extensions page has a visual diagram which explains how things are split up which hopefully makes things a bit clearer when you have your split mode and divide ratio finish you can hit the visualize and make template button which will do two things the first thing it will do is give you a template for laying out your prompts and I'll be using a simpler one so you know what to do the second thing it does is show you how your grids are laid out using different colors so it's all easy to see now for an example on using Matrix without any other options selected I'm going to use a split mode of vertical and a divide ratio of 1 1 which will give me three rows the idea is to use those three rows to separate the head chest and Beauty area into regions which we can then modify in the non-destructive way this is what the prompt looks like using the template we were provided with earlier on but you can swap out add Row for break or end depending on your preferences and this is the resulting image where you can see I've generated a series of images swapping out the colors of the shirt alone if I were to try and do this without Regional prompts to run the same seed you can see we get the one colors across our image and a lot less consistency overall but here's a question where do we place prompts for our background whether you could place it anywhere but if we do this we either get no changes at all all changes in only one portion of our image this is because our background prompt is currently being applied to a specific region of our image and you can see this if we use comparisons where I've put the background in a different part of our prompt in each image the effect becomes weaker and weaker and it's not the Lush Forest I was looking for to fix this we have two options we can enable use based prompt and visualize a new template which would include the ad base keyword we can then increase our base ratio to increase the strength of that background in our image as seen in these images but increase the value too high and you will lose the regions as the base prompt begins to take over next we can try the use common prompt option which will apply our background prompt to the start of each region we don't have the ability to control the weighting through the extension using this option and you would have to wait The Prompt for a stronger or weaker effect lastly let's have a look at the use comma negative prompt which will allow us to add a break after our group of prompts we want applied to our entire image then we can apply prompts to the next region before using another break to enter a new region in the negative prompts you should have a similar layout to the positive fonts as seen in my prompt box and in this image I'll produce the size of the hips busts and nose which are all areas specific to the region the negative prompt specifies I've included those additional breaks as we can use loads to pass from one region to the next so on the left we're specifying our adjustments and on the right and specifying that no changes are made so you can see the comparison our second option is mask and this option allows you to create a series of masks which can contain portions of your prompts this option has the benefit of allowing you away to visually Define regions and then apply prompts to those regions but as the disbenefit of feeling underdeveloped you have an image tab called do not upload here until bug fix and this is where air you can either draw or upload masks for use in your image but I would advise against uploading a mask in this section as they are resolving bugs your regions operate similarly to layers in Photoshop and this is how you navigate to a layer or region to use for drawing a mask modifying a mask or identifying which region your mask is located you also have a special region number which is -1 and when you paint on this region number it will fill the painted portions of your mask with a white color acting similar to an eraser tool you then have in paint width and then paint high and this is used to determine the size of the Mask which should be the same size as your image you also have two important options create mask area is the option you select to create the canvas in which you can begin drawing your mask ensuring that your region number is on one but any number other than -1 will work you can then select draw region Plus show mask which has two functionalities the first functionality is is to take the mask you have drawn on the canvas and add it to the collection of masks before taking you to the next region number you can view the created mask in the mask image box on the right hand side and the next option is to show The Mask commonly drawn on the selected region we can accomplish this by navigating back to region 1 and selecting draw region Plus show mask in which the mask image box on the right will show the corresponding mask drawn on that region number you can have up to 360 regions starting from Zevo but it's important to ensure that this does not spiral out of control as you will need to remember the order of your mask and then order the content of your prompts in line with the regions for example in region 1 I've masked the top part of the canvas where the moon will go and on Region 2 I'll be painting another mask for the character which will be the third prompt as the first is our base the second is our moon and the third is our character if we swap the promise order of the Moon and character then you can see our image isn't generating correctly as we're trying to draw a moon on the bottom and the character at the top which doesn't make sense within stable diffusion's data the final area is upload mask here because radio and I think this is the place where you would upload your mask instead of the box at the top as a temporary fix until they've resolved the bugs our last option is prompt and this will allow you to Define regions through prompts where a mask is created to Define regions of your image in my opinion this method is the most reliable for color of accuracy and stopping color bleeding looking at our options we have two prompt modes prompt and prompt X and when using prompt duplicate areas are added while prompt X overwrites duplicate areas while I'm not entirely sure as the difference is when using this on my new I believe that this ensures that any mask created for an area of your image overwrites the area of the previous mask a bit like layers in Photoshop so that objects which are overlapping one another can be drawn properly for this reason specifying elements of your image from large to small will ensure that you start with the largest mask and then overwrite it with the smaller elements and then we have the threshold where you can specify a waiting for each region of your prompts or use a single weight for the entire image moving on to the formatting there isn't really any template provided for this section but the layout isn't too difficult what you need to do is write out your prompts using non-descriptive terms for the parts you wish to control with regional promiser later for example here I'm describing tapered trousers long sleeve blouse long hair eyes Man glasses but notice I haven't specified any colors that is because my color will be specified after the break keyword where we can specify the colors we want with a strong weighting then use a comma and a reference to the prompt we want the color of assigned to and the break to end that region so for example we're doing dark blue comma tapered trousers to reference our tapered trousers and I'm using a base prompt as I find it gives better results but you don't have to use it for this method avoid using common negative prompts as it will take everything before the first break and apply it to the start of each region which can ruin the accuracy of the image you can also use descriptors in the regions such as where I'm specifying a chevron print for the blouse and you can see that the pattern is being applied accurately to the shirt with the weight in controlling how prominent the design will be now all of these prompts will require tweaking and experimenting to find what works for you but when you find a series of settings that give you what you want you can swap out the colors and descriptions relatively easily without destroying the overall image or worrying about color bleeding now that we've explored how it all works let's have a closer look at further use cases for this tool While most of my examples have been on anime checkpoints as they struggle with color bleeding the most you can certainly use this with realistic sharepoints as well for example let's imagine I'm a fashion designer working in New York City and meeting my avocado one toast drinking my pumpkin spice latte sitting on the balcony of my five thousand dollar per month studio apartment and trying to come up with a couple of concepts for clothing designs you can combine XYZ plot with wild cards for randomization swapping out multiple areas of your prompts in a single image then generating multiple images on different seeds to achieve an array of random designs which you can come back to for more inspiration a subscriber also asked me how you can achieve realistic looking clothing and something I wanted to mention quickly is that texture plays a huge role in how clothing looks for example I could specify a silk Blazer and this would give me a different result compared to cotton velvet or even leather when structured in my prompt I would personally include the texture in the base prompt rather than the region as this would allow you more freedom and the ability to modify the weighting of that texture I'll be using this extension for some of my workflow videos on both YouTube and patreon where we'll be making art so something you wanted to see wasn't covered here then I'll likely cover it in a future video you also have a preset section which to my knowledge doesn't work as I was getting ever messages when trying to use it but this allows you to save and load presets for your settings and options chosen finally we have the lower stop step options and the documentation on these is lacking to save the lease and testing them out there weren't any changes to my image but I've never encountered a situation where I've needed to touch any of these options so it shouldn't hinder your use of the tool but I hope this video has been useful and if it has consider becoming a supporter for upcoming additional content and ensure you like the video to help us hit the algorithm this is bite size genius and I hope you enjoyed

Channel: Bitesized Genius

Views: 11,048

Keywords: Stable Diffusion, stable diffusion prompt guide, stable diffusion controlnet, stable diffusion prompts, BitesizedGenius, Automatic1111, stable diffusion lora, stable diffusion extensions, stable diffusion embeddings, stable diffusion checkpoints, stable diffusion anime, stable diffusion scripts, stable diffusion video, stable diffusion img2img, Stable diffusion realistic, stable diffusion models, stable diffusion install, Stable diffusion tutorial install, regional Prompter

Id: WwATqUp5gBc

Length: 15min 14sec (914 seconds)

Published: Sun Aug 27 2023