Stable Diffusion IMG2IMG: EVERYTHING you need to know IN ONE PLACE!

[00:00] in today's video I'm going to show you

[00:02] everything you need to know about the

[00:03] imaged image tool in automatic 1111

[00:06] image to image sketch in painting in

[00:09] paint sketch even in paint upload this

[00:11] video has it all ready let's get started

[00:14] the image to image tab is an essential

[00:16] tool in your air toolbox it lets you

[00:19] create a new image or new elements of an

[00:21] image from an existing picture that you

[00:23] provide there are so many powerful tools

[00:25] here but let's start with the image

[00:27] damage tab itself really you're going to

[00:29] use this tool if you have an image that

[00:31] you want to use as a starting point you

[00:34] can use that image to pull elements of

[00:35] composition and color into a brand new

[00:38] image so I've got an image here that

[00:41] I've generated earlier it's a portrait

[00:43] of a girl on a city street I use the

[00:45] sdxl base model to generate this image

[00:47] but you can use any image you want an

[00:50] existing photo a painting whatever so to

[00:53] start we'll drop that into the image to

[00:54] image box you can also send images here

[00:57] from the text image tab just click on

[01:00] send to image to image under your

[01:02] generation at the top just like we would

[01:04] for text image we can add in positive

[01:07] and negative prompts if you did send

[01:09] your image across from text to image the

[01:11] original prompt will appear just like

[01:13] you see here below the image box you'll

[01:15] see all the settings we can play with

[01:17] now a lot of these settings in this tab

[01:19] are shared with text image so I'll only

[01:21] touch on the image to image specific

[01:23] ones here for a more in-depth

[01:25] explanation of settings check out the

[01:27] video in the description the First new

[01:29] setting is resize mode use this if your

[01:32] new image will have a different size or

[01:34] aspect ratio than your original image

[01:36] for example if your original image is

[01:38] landscape and you want your new image to

[01:40] be portrait the default is just resize

[01:43] this will stretch or shrink your

[01:45] original image to fit the new image you

[01:46] create crop and resize will keep the

[01:49] aspect of your original image and crop

[01:51] it to fit the new canvas size resize and

[01:55] fill takes the original image fits it

[01:57] into the canvas and fills in the blanks

[01:59] with colors from the input image I don't

[02:01] see much use for this one but let me

[02:03] know in the comments if I'm missing

[02:04] something just resize latent upscale is

[02:07] just like just resize but also upscales

[02:10] the image if necessary note that this

[02:12] setting is a bit tricky to get right and

[02:14] there are easier ways to upscale your

[02:16] images below resize modes you'll find

[02:18] settings you're already familiar with

[02:20] including sampling method sampling steps

[02:23] your size and batch settings and the CFG

[02:26] scale the most important setting here is

[02:29] denoising strength stable diffusion

[02:31] makes images by taking random noise and

[02:33] converging it into an image based on

[02:35] your prompt this setting will control

[02:37] how much extra noise is added to the

[02:39] picture and in turn will determine how

[02:41] different the new image is from the

[02:43] original lower settings won't change the

[02:45] image that much and higher settings will

[02:47] make bigger changes I like to use 0.7 as

[02:50] a good starting point okay so now for

[02:52] the fun part once you've sent over an

[02:55] image you can start tweaking some

[02:56] settings to refine your picture we sent

[02:58] this image over from text to image so we

[03:00] have all of our settings and the

[03:02] original prompt now we could just start

[03:04] tweaking our prompt here but bear in

[03:06] mind that the entire prompt plus the

[03:08] image will be used to create a new image

[03:10] so if you aren't using the same model

[03:13] seed and sampler from the original image

[03:15] you may get some weird results so

[03:18] instead we're just going to use some

[03:20] simpler instructions to see how it will

[03:22] affect the image so let's get rid of the

[03:24] prompt and put in pink and and blue hair

[03:30] and let's keep the same sampling method

[03:33] and steps we'll tick restore faces and

[03:37] we'll keep the default on just resize

[03:39] and keep our Dimensions the same and

[03:42] we'll set our denoising strength to 0.7

[03:45] so we should see a bit of a variation in

[03:48] our output foreign

[03:53] seven that's had a pretty big effect so

[03:56] if we want something a little more

[03:57] subtle let's try lowering the denoiser

[04:00] strength to 0.5

[04:03] [Music]

[04:07] and yeah that's a much more subtle

[04:09] impact on the image

[04:11] so you can keep on adding more detail

[04:13] here and reworking things until you get

[04:15] an image that you're looking for

[04:17] when we use this method though we can

[04:19] see that the prompt has had an impact on

[04:22] the entire image the hair and eyes have

[04:24] changed but so has the face and the

[04:26] clothes let's say we just want to change

[04:28] a particular part of the image and keep

[04:31] the rest just as it is that's where in

[04:33] paint comes in in paint is a powerful

[04:35] tool that allows us to paint over the

[04:37] specific parts of images that we want to

[04:39] change this is especially useful when

[04:41] you like the overall composition of your

[04:43] image but you get one of those Cosmic

[04:46] nightmare faces from the hell Dimension

[04:48] just no to start you can click on the in

[04:51] paint tab or just click in paint under

[04:53] the image to copy the image over and all

[04:56] of its settings so now we can just paint

[04:58] over the particular part of the image we

[05:00] want to change in this case the hair we

[05:03] can use the settings in the top right of

[05:05] the box to change the brush size or

[05:07] clear our selection if we go wrong so

[05:10] let's just paint over the hair and then

[05:13] we can check our settings we've seen

[05:14] some of these already so we'll keep

[05:16] resize mode on default mask blur is the

[05:19] amount of blurring that happens around

[05:21] your painted mask lower numbers will

[05:23] give you harder edges while cranking

[05:25] this up will affect how many pixels

[05:27] beyond your mask will be affected by

[05:29] your prompt the default of four is

[05:31] usually fine but feel free to play with

[05:33] this if you're getting hard edges mask

[05:35] mode determines what is actually changed

[05:37] in the image in paint mask will change

[05:39] the parts you've painted while in paint

[05:42] not masked will change everything but

[05:44] the parts you painted this is great if

[05:46] you want to change the background or

[05:48] large amounts of your image in one go

[05:49] next you can choose the masked content

[05:52] setting these settings tell stable

[05:54] diffusion what method it should use as

[05:56] its basis for generating the new image

[05:58] part The Fill option will take the

[06:00] impainted area blur it then generate

[06:03] using that as its base the original

[06:05] setting will use the original image

[06:07] unaltered as its base to generate a new

[06:09] image from latent noise will fill the

[06:12] impainted area with new random noise

[06:14] based on the seed number this can be

[06:16] used if you want to result completely

[06:18] different from the original Late nothing

[06:20] will fill in the painted area with a

[06:22] blend of the colors from the surrounding

[06:24] pixels and base the generation of that

[06:25] these settings will have a dramatic

[06:27] effect on the output but if you're

[06:29] making Minor Adjustments the original

[06:31] setting will usually work great the

[06:33] in-paint area setting is easily confused

[06:35] with mask mode but they are different

[06:37] choosing whole Picture Tells stable

[06:39] diffusion to use the whole picture as

[06:41] inspiration for the in-paint generation

[06:43] while only masked will treat the Mast

[06:46] area in isolation and not draw from the

[06:48] rest of the image typically whole image

[06:50] is the best setting if you want the

[06:52] painted area to blend in but if you do

[06:54] choose only masked you can change the

[06:56] padding size right next to it which will

[06:58] tell stable diffusion how many

[06:59] neighboring pixels it should take into

[07:01] consideration as inspiration for the new

[07:03] generation and of course below that we

[07:05] have the CFG scale and the noising

[07:07] strength let's stick this on 5.5 and 0.6

[07:11] and see what happens

[07:17] now that looks pretty good but we can do

[07:19] even more within painting sketch so

[07:21] let's say we want to give our model a

[07:23] scarf we can send this image to in-paint

[07:25] sketch just like we did before we can

[07:27] paint over the image but we now have the

[07:30] option of choosing colors so let's go

[07:32] with red here and just paint in the area

[07:35] we want to add our scarf to

[07:38] and in the prompt we'll just type in red

[07:40] Woolen scarf

[07:42] for this you usually have to bump up the

[07:45] denoising strength so let's go to 0.8

[07:47] and we can keep all of the other

[07:49] settings the same as before

[07:51] and let's hit generate

[07:56] and yeah that looks pretty good for a

[07:58] first try now you can keep iterating and

[08:01] adding to this to really get the results

[08:02] that you want the last in painting tool

[08:05] is in paint upload now I'm not going to

[08:07] go into too much detail because it's

[08:09] more of an advanced topic but basically

[08:11] you can create a mask in another program

[08:13] like Photoshop using black for the parts

[08:16] that you want to keep and white for the

[08:18] parts that you want to change it's

[08:19] actually a pretty powerful tool and if

[08:21] you have the patience to get really

[08:22] detailed it can be really useful

[08:25] and finally that brings us to the sketch

[08:27] tab if you're struggling to get the

[08:29] image you have in your head onto the

[08:31] screen using image to image or other

[08:33] tools you can always flex your creative

[08:35] muscles and draw it out here

[08:37] simply drop in a white or black mask in

[08:39] the Box provided and use the paintbrush

[08:42] to sketch out your idea

[08:43] use color to highlight the details you

[08:46] want to show up and pair that with your

[08:48] prompt and watch your sketch turn into

[08:49] something incredible

[08:51] there is so much more image to images

[08:53] capable of and in my next video I'll be

[08:55] showing you even more ways to level up

[08:57] your art with the image to image tab if

[08:59] you found this video helpful why not

[09:01] leave a like And subscribe to see more

[09:03] videos like this one thank you for

[09:04] watching to the end I'll catch you later

[09:07] foreign