STOP making BORING AI art

there is some AI art that you just can't replicate and that no matter how good your prompts are you will never get to that level of detail but today we're going to learn how to make your own incomparable art pieces that the image to image tab is a very underrated part of automatic 1111 but when used correctly it can lead to the best images that no one can replicate and to do that we need to really understand how it works when you have created a picture that you really like learn here you have this sent to image to image option clicking it will transport all the settings you had into this page as well as adding the picture into the image to image UI if you don't have a previously generated image you can also import any photo that you have on your PC by dragging it in and now that we have inputted our image you will see that the UI still has most of what you had in text to image so what's the deal here well we're going to use this picture as a base to get the perfect result we're looking for and there are so many techniques so let's get started first we need to know that stable division will take whatever it sees in here and try to create a new image based on that and the prompt that you give it you can use the exact same problem seed if you just want to upscale the image but for now I'm interested in having a random seat and changing the prompt this way we can start to really change the original picture how much the result differs from the input will depend on the denoising strength the lower the number the closest to the original it will be and the higher the number the more it will change it is important to know that even with the noising strength at 1 the result will still take some of the original image as reference here we have the image I just made that completely by chance it looks like Katara from the good Avatar would make a really good match with it so let's change the prompt at the Katara Lora that I have and then we'll hit Jared this time with the default denoting strength at 0.75 not a bad result but I feel like it loses a little too much of the original image so I'm just gonna use the script to change the noising strength and find the result that I like the most it still needs work but I like the result at 0.6 strength so I'm gonna drag it in and keep working from there the composition is fine but I would like a little more space to the right this way I can add floating water later on here the aspect ratio comes into play we are not bound to the original images aspect ratio but if we want to modify it there's some things we need to keep in mind if we want to upscale the image for example double it in size or if you are a psychopath and you just want to stretch it we need to use the just resize mode to crop the image to a new aspect ratio we will use crop and resize if you click this icon right here a grid will appear playing with it you will see that it isn't bound to the input aspect ratio but don't worry the grid only marks what you want the initial picture to be it will follow your aspect ratio starting from the center of the grid so here you can adjust what you want the new focus of the image 3. be really careful though always maintain the grid inside the picture or it will create pure black Parts on the result it will be the same as if you inputted an image like this also once it crops I don't think you can Digraph which is annoying you will have to drag your initial image again if you grab too much the same applies to resize and fill this is the out painting option it will take your original input and rematch in the surroundings using AI until it matches the new size it might look intuitive to use the grid to guide stable diffusion on what parts you want it to fill but it will just make everything black remember that the Great only marks what you want the input to be so keep it inside the initial image and just use it to pick a new center if you have to and finally just resize blade and upscale this will reimagine everything but following the new image size I tried finding a use for it but I don't know just worthless if you know how to use it please leave it in the comments for us thank you to get the new composition I need let's just use the third option we will probably have to do this twice to get the amount of space I'm looking for but that's fine I'll keep the denoising strength at less than half of this as I don't want it to change the initial image but I also don't want this to happen you can keep iterating all you want and adjusting the prompt as you go once you've found an image you like you can click on send to image to image and keep going you can always use image to image just for inspiration generating similar images but with new Concepts it would be like normal iterating but with a head start for example here I could have proceeded with this other image that was also pretty cool and followed the same concept you can even change the model that you're using to get a completely different result with different styles and capabilities but playing with the noising strength as you see fit notice that from 0.3 to 0.5 the image stays relatively close to the original and going over 0.6 starts changing the image quite a lot then between 0 to 9 and 1 there's a massive difference what's this what oh it's the video is getting boring arm oh my god get this quick if you have an image and don't have the prompt to recreate it you can just drag it into image to image and then click either interrogate clip or interrogate The Brew what whatever and in a few seconds it will give you a prompt that will try to create a similar image okay prices are burden now on to this in bending once you start with this face it will be almost impossible for people to recreate your images the hardest part of this process is deciding when to stop it is so much fun you can click this button here or just drag the image in we will use this to change the image one part at a time be it by adding details or correcting a certain part like the face changing a part to make it completely different or adding new fun stuff to add to the images story to do this we need to mask the part that we want to change and then prompt accordingly to what we want up until now we prompted what we wanted the whole image to look like but from now on we will prompt based only on what we want inside the mask it is fine to give some context though I also like keeping the enhancers in to better match the mood of the overall image to mask we can paint over the input adjusting the paint brushes size up here on the viewers up just click this button to undo this is the most common way to do it but if you need more Precision you can actually import a mask by going into input upload for example let's take this image to photopia or your editing software of choice and create a quick mask make what you want to be masked White and the rest can be full black then just add the original image up here and the new exported mask down here if you want you can invert the mask inside the UI by clicking the in paint not masked option so now we could change the main character or if we wanted we could change everything but the main character careful using two precise of a mask as then the image can look like a poor Photoshop job you can use the mask blur to expand the borders of the mask with a slight fade in painting is basically an image to image changing only the content inside the masked areas in order to get the best results we need to understand these options though so let's take a look at what they do first masked content refers to what the content will be inside the mask when starting the image to image process you can understand this better by putting the noise strength at zero and seeing what each model is filling your mask with the first option will fill the mask areas using the original images colors surrounding them this is a pretty good option to erase parts of the image that you don't like or to create new parts using the original color scheme the anointing strength will vary depending on the situation the original option will keep the insides of the Mask as they were before and use that as a base if you want something similar to what you have then use a low denoising strength else you can play around with it this will probably be the option that you use the most but it allows for minor changes or complete reworks also there are some cool tricks that we will see later on latent noise is pretty weird it will use a new noise that looks like this I would suggest using denoising strength at more than 0.75 if you don't want our third window in the final result lowering it can get some cool results though I'd use it if I wanted something really different from the original image or if I wanted this effect for some reason like there's nothing like the name implies it's good for pretty much nothing it's like a bad feel version I don't know I guess it can work if you need something really different from the input but other options aren't working but at that point you should just use the tab that we are about to see before doing that though we need to understand the most important of options in paint area here we have either whole picture or only mask and this will be what you play around with the most so I'll try to explain the best that I can whole picture it resizes the image to the input resolution while making the changes it will basically do an image to image of the whole input but only applying the changes to the masked area it also takes into account all the image as context when deciding what to create it is really good to erase things from the image or create stuff that needs the overall context like in this case to avoid having weird perspective changes only must generates the new part of the image with the input resolution so instead of resizing the whole image it creates a new generation with their input size and then pastes it back into the original image this makes it really easy to add detail as you can manually adjust the resolution of each part of the image hey to explain only mask only fixing the face step by step I'll do it this way so we can chill for a second and really understand what's going on first thing to do is to mask the face out and then type a new prompt I'm gonna take pretty much everything out except the Lauren Perfect face and the enhancers and Etc I want her to look focused so I'm gonna add frowning as well we will obviously use only Mast option and that's because faces need a lot of detail the face that we originally had isn't that bad so I'm gonna use the original as the masked content this mode uses only masked pixels for context by the way this only works in this mode so the more pixels you add the more context of the original image will be used we can see this in the preview where you can actually see the panning pixels that are being taken into consideration make the padding larger if you think the result needs a lot of context and smaller if you want to focus more on the details and you don't really care that much about the rest of the image for the face I want detail but I'll need at least the whole head for context there is two options to really control what you want in the image one is the tedious job of creating a composed picture in Photoshop and then using that as the background with the original mode on and to be honest that is the most reliable one but I like efficiency so let's go to infant sketch I haven't seen many people use this but I find it to be such a powerful tool here we can draw over the image with any color you like this color will not only act as the mask but you can also use it as part of the input image at the same time with some super basic drawing skills you can give a base for stable diffusion to create pretty much anything you need don't expect it to create exactly what you had in mind but depending on your drawing skills it might be better or worse so I paint the water shapes I use the original as the masked content mode and bearing the denoising strength depending on your drawing we can get what we wanted I would love to explain what mask transparency does but I haven't found a useful way to use it yet using what we learn we can repeat these steps over and over again until we get the result we were envisioning I'll pop into Photoshop to adjust what I'm too lazy to do with AI and finally we have our image now on its left is to upscale it I recommend downloading this upscaler that I'll leave in the description just put in the models yes or gun folder and the easiest way to do this is to send the image to extras and upscale it right here choosing the new upscaler to unlock the best upscaling option and start using stable diffusion like a pro you need controller so watch this video as I teach you how to install the most important extent ever made so yeah

Channel: Not4Talent

Views: 34,300

Keywords: stable diffusion, sd, stable difusion, promting, prompt, tutorial, guide, stable diffusion prompt guide, stable diffusion tutorial, ai art, ai, crate ai art, beautifull ai art, create good prompts, talk to ai, talk to stable diffusion, inpaint, img2img, i2i, outpaint, inpainting, outpainting

Id: hDfJajYxOc4

Length: 10min 59sec (659 seconds)

Published: Thu May 11 2023