Image To Image Tutorial In 13 Minutes – Stable Diffusion (Automatic1111)

image to image is one of the most powerful features within automatic one allowing you the ability to take an existing image and modify it through prompts via a process known as imp painting in this tutorial I'll be taking you through all of the options with examples so you can spend less time reading and more time creating a shout out to the supporters and once you've liked this video consider supporting as well but enough talking let me give it to you bite-sized so the biggest difference between the text to image and image to image tab is this image box which is where we place the image we want to modify this is done by dragging and dropping the image in question and you can then combine it with prompts and other tools to adjust it further The Prompt box operates the same but instead of generating a fresh image from your prompts alone it will influence how your existing image should change to match the prompt based on your denoising strength here I'm describing the same woman in the image but with some details adjusted and we get that smile coming through while keeping most of the image the same and as the reference the negative prompt tells the software what to avoid in your image and this works the same as text image in this example I've stated that I don't want any red and the higher the waiting the less red we get in our image interrogate clip is a powerful tool which will take your given image and suggest a series of prompts that will attempt to replicate the image you provided in this instance and receiving written prompts for a checkpoint that uses tagging so the results will differ greatly from the original Deep Buu is like interrogate clip but this will attempt to generate a series of prompts from an image based on damaru tags doroo is an image board for art mostly leaning towards anime gaming and some non-safe for word material because this is giving us tags we're getting a far better result which matches our image more closely than interrogate clip you then have copy image two alongside the list of locations you can send the image this is an alternative to dragging and dropping the image but either method will work so let's break down what each of these Tabs are sketch will allow you to draw on your provided reference image and then use those adjustments to influence the new image for example we could paint our clothing red and intensify that color on only the mass areas ignoring our hair and other nonrelated parts of our image in paint is one of the most powerful Tools in stable diffusion as this will allow you to mask a portion of your image and use your prompts to generate a new image in that specific mask location for example I can mask the eyes and specify a blue color and here you can see both the mask and the resulting image under in paint you will find some additional options unique to this section of stable diffusion so let's quickly cover these while we're here mask blur will apply a blur to the edge of the mask and if I bring up the mask on screen a sharper mask makes the in painting stronger in that area while a blurrier mask lessens the effect removing it completely at higher values but can also help blend the in painted area in some cases then looking at our mask mode we have two to choose from inate masked will ensure that only the mask portion of the image is filled with the newly generated content the impa not masked option would ensure that everything outside of the masked area is filled with the newly generated content the white area is what is mask while the black area is the unmasked portion because we're inverting our mask and this can be useful for masking everything except for the specified locations saving you time then for our mask content Phil will draw your prompted details within a blurred version of the Mast area of your image leaving some room for stable diffusion to interpret the details because you're making adjustments based on what is already there original will fill the mass portion of your image with your prompted changes based on the original content of the section to be altered this is useful because you're changing what is already there without blurring or doing anything to change it beforehand latent noise will fill the mass space with random noise and then generate an image within that space from that noise this is similar to when you generate an image with a low sampling step which produces a very noisy image but it becomes clearer the more steps you use latent nothing will fill the mased area with a flat color made up of the prompts you're using to fill the space and it will then build up the image from there based on the number of sampling steps you're using in paint area allows you to specify whether to modify the whole image or just the mask area when applying the in painting process this is because making certain adjustments like resolution changes even when using a mask can change the resolution of your entire image by selecting whole picture you can ensure that the resolution for the mask area is in line with the rest of your image but by choosing only mask you can ensure that the resolution you choose only applies to that mask portion which can help you get more detail on Lower resolution portions of your image only mask padding pixel determines how much of the area around the mask should be used to generate the new image which will help blend both the surrounding area and the image within the mask by better understanding the context of the image in paint sketch allows you to draw a mask or added ability to include colors which can help stable diffusion understand an objects much better by recognizing both the color shape and description of the Mask in this example I'm turning our shirt blue but even if I remove the blue from my prompt it still recognizes the color of the Mask in paint upload and allows you to use your own custom mask for your reference image which is more powerful than hand painting as you get access to the same features but you can create a custom mask in an image editor that's much more accurate and fit for purpose batch is used for generating multiple images At Once by specifying an input directory where your reference photos are located an output directory which is where the generated images will go and you can even reference the location of a folder for the impact mask and control net mask this is done done by specifying the folder locations for your outputs inputs and masks ensuring that your masks have the same name as the image you want to associate them with you can see on my screen how I'm using the name image and then the number to specify the sequence and my mask also has the same name and number as the corresponding image they should be applied to for the resize mode just resize will ensure that your generated image fits within the specified resolution even if that means stretching and skewing the image in ways which may look unnatural crop and resize will ensure that your generated image fits within the specified resolution even if it has to crop certain portions of the image resize and fill will ensure that your entire generated image fits within the specified resolution even if it has to fill in empty space just resize Laten upscale is the same as just resize but this doesn't use will of stable diffusions upscale models and instead uses encoders and latent space the rest of the featur fees I'll be covering from here on out were also covered in my text to image video so consider checking that out for for examples which may help your understanding sampling method is an algorithm used for Den noising an image or in plain terms stable diffusion correcting its Mistakes by generating new samples of an image in steps and giving you a cleaner image the more steps you have sampling steps are the number of times the sampling method will produce an improve image and the higher this value the cleaner the final result and the lower the value the more noisy the final result restore faces can improve faces in generated images that have been distorted or messed up by stable diffusion it's controversial because it can produce worse results depending on the type of checkpoint you're using being better suited for realistic ones tiling is used to ensure that images can be tiled or in plain English to ensure that generated images can be placed next to one another seamlessly resize 2 lets us choose a size we want our generated image to be by specifying a width and a height value and resize by lets us choose our resolution via a multiplier which multiplies the original size of the image by our specified number batch count determines how many sets of images you will generate by multiplying the batch count by the batch size while batch size determines how many images should be generated in a single batch so if I have a batch count of Two and a batch size of three I'm generating two sets of three images the CFG scale will determine how strongly the image generated should conform to your prompt and how much it should be completely unrelated to your prompt most checkpoints suggest the ideal CFG scale for their particular model but keeping it at a modest value between 5 to 9 tends to lead to the best results the seed is a random number which determines the variation of your generated image and using the same seed should give you the same image each time you generate an image providing you don't change any of the settings on the topic of seeds this dice icon will set your seed to minus one which is a special value that will select a random seed each time you generate an image this Green Recycle option will reuse the seed from your last generated image which means you won't have to copy and paste the seed from the image file's name variation seed allows you to select a second seed which can be combined with your main seed for the purposes of combining two separate variations of the same image into one you can see both our original seed and variation seed and then the combined image in this example the variation strength controls how strong the blending is between the two seeds with a low value giving you the image from your seed and a higher value mixing in more of that variation seed when you change the height or width for your generated image it will also change the variation you get even if you use the same seed to avoid this we can use the resize form width and height option so even if we change the resolution of our image we can use the variation from a different resolution stable diffusion also allows you to utilize custom scripts and comes with a few installed for you to use you then have two options that show up when you generate an image called interrupt and Skip interrupt will stop the generation of all images while skip will skip the generation of the current image then move on to the next one when generating multiple images through methods like batch this blue arrow icon will and apply the prompts and settings from your last generated image providing the prompt box is clear of any text you can also use this to take generation data copied from your prompt box and apply it to the appropriate Fields the bin icon will delete both your prompts and negative prompts from the promp box but your settings will not be deleted the Styles box has undergone a change where you can now create Styles in a much easier way by clicking this pencil icon and entering your prompts alongside the name for the style and pressing SA save you can also now delete or edit your Styles all within this window which is an amazing Improvement to the feature when the style is selected from the drop- down list it will be applied at the end of your prompt by default you can apply the contents of a style directly to your prompts by using the notepad icon within the style itself lastly you can use the name of the style in curly brackets to specify a position for that style within your prompts like how wild cards work this will give you a greater degree of control over your prompting when saving Styles you like and applying them to your work you then have the classic image box which is where your generated image will show up when it's completed you have a few additional options when you click through the image such as toggling the zoom view previewing the image through tiling saving the image to your computer and exiting the zoom view you have the folder icon which is used to open the directory where the image is located and you can access all image types here the save dis icon will save your image and an Excel spreadsheet of the generation data to the stable diffusion folder this is different to the output folder where generated images are automatically stored you then have this filing cabinet icon which is used to save your images into a zip folder with the generation data similar to the previous option but with a zip file the canvas icon will send your image to the imageo image Tab and you can also send the image to the in paint tab these options are useful for sending new images to these tabs without having to drag and drop them lastly the ruler is used to send the image and its generation data to the extras tab where you can access upscaling and other options you have installed which use that tab but I hope you found that useful and if you did like the video and subscribe for more upcoming content this is bite-sized genius and I hope you enjoyed

Channel: Bitesized Genius

Views: 15,523

Keywords: image to image stable diffusion, image to image stable diffusion tutorial, stable diffusion image to image controlnet, image to image stable diffusion anime, Image to image stable diffusion face, stable diffusion gui, stable diffusion browser, stable diffusion ui tutorial, stable diffusion options explained, stable diffusion tutorial, stable diffusion img2img, img2img stable diffusion, img2img stable diffusion tutorial, img2img tutorial, img2img automatic1111, img2img video

Id: PUwLT9JwCs8

Length: 12min 51sec (771 seconds)

Published: Wed Oct 04 2023