Inpainting in Stable diffusion for beginners.
HTML-код
- Опубликовано: 21 сен 2022
- Updated Advanced Inpainting tutorial here: • Inpainting Tutorial - ...
In this tutorial I'll show you how to add AI art to your image while using #inpainting in automatic1111's webui of #stablediffusion on your computer. Super easy, start making your own #ai images similar to #midjourney
Chat with me in our community discord: / discord
Support me on Patreon to get access to unique perks!
/ sebastiankamph
Ultimate Stable diffusion guide
• Stable diffusion tutor...
The Rise of AI Art: A Creative Revolution
• The Rise of AI Art - A...
Ultimate Animation guide in Stable diffusion
• Stable diffusion anima...
Dreambooth tutorial for Stable diffusion
• Dreambooth tutorial fo...
5 tricks you're not using
• Top 5 Stable diffusion...
Avoid these 7 mistakes
• Don't make these 7 mis...
How to fix live render preview:
• Stable diffusion gui m...
In order to get even better results, and a better integration of the new object into the picture (with an appropriate style, texture, lighting, etc.) it is best to keep the prompt as close as possible as the prompt that had been used for generating the original image. You can then just insert the description of the new object at the end of the main scenery/object description, separated with commas, just before the description of the style, lighting, artists, etc. If there are many other objects that are described in the prompt, you can delete some of the inessential ones, or just increase the weigh of the new object that you want to inpaint in order not to confuse the AI as to what you want to add. (If you merely put the bare description of the new object in the prompt, the result often looks a bit like a decal sticker had been stuck on the original image; it blends more with the immediate surrounding thanks to the mask blur, but it still clashes with the overall scenery and lighting conditions).
Hi, do you think you could please write an example?
@@azeemotron I am thinking something like this.
Original Prompt: A photo of black and white cat laying on the sand in a desert with cacti in the background, highly detailed, grainy texture, 35mm, art in the style of catman
Inpainting prompt when adding a bowl of milk: A photo of black and white cat laying on the sand in a desert with cacti in the background, a red bowl filled with milk, highly detailed, grainy texture, 35mm, art in the style of catman
@@time2slither307 no, delete everything an just keep artstyle of original prompt;
inpainting prompt when adding a bowl of milk:
a red bowl filled with milk, highly detailed, grainy texture, 35mm, art in the style of catman
@@Nick-vd7cg That makes way more sense than what Poker and Philosophy said.
@@fluffylittlebear I mean he basically said the same thing just more complicated
Finally some decent explaining done by someone!
Thanks, helped me out a lot on improving my images that just needed a final touch!
This is so much easier to understand. Thank you I've been fighting with inpainting since its introduction.
You are so welcome! Really happy that you feel it's easier now.
Thanks for the video, I just got hooked into this stable diffusion movement and needed to understand what the heck is inpainting. Subscribed and I will watch the rest of your videos.
you are the best. There are ton of people that just say "provide png with transparent mask, it works"
Thank you for this !
Very interesting …. Thanks for the explanation!
Thank you! Is there a workflow for using inpainting on the generated images and then use the Video/Animation option to interpolate?
Oh, also would be great if you explain the SCRIPTS feature of the automatic1111 UI, the "SD Upscale" is very interesting.
Be happy to provide more explainer videos, I'm making them as fast as I can and will get more and more up. If people feel they're helpful I'll keep at it :) Next up will be an explainer on audio syncing AI animations automatically.
Good tutorial
Great video :) Unfortunately sometimes SD don't respect inpact command and doesn't paint nothing. They should better this amazing function that already works amazinly on others. But they have limitations
Is there any domain specific languages to describe precisely what we want in a scene? If not we desperately need it!
Thanks for the video. Can you make one on the correct usage of the other masked content modes? Fill, Original, Latent Nothing. I have looked all over, including the official documentation, but there still seems to be a lot of confusion. Thanks.
Agrees, these functions are not covered at all
I cover them in my ultimate guide. See my channel.
Wow watching this in the future when used to modern Sebastian is trippy, you were just a wee little baby AI artist!
Haha, so true. Production quality was lousy 😅. You think a new inpainting video is needed?
@@sebastiankamph I would love one, just because I struggle with it even now. Latent Noise fill never wants to agree with me. You could also cover the Out Painting A111 extension too if you wanted to since that one seems super powerful!
Hey Sebastian,thanks for the great video
I had a question about the masks,
What I want to do is only to create the prompt I want without a background, while using a png image transparent in some parts.
For example, a transparent man is the image I have, and I want a crown on his head, I paint his head or make a mask for it, prompt " a golden crown" and the crown is looking good, but SD always fills the background with random stuff.
Always creates a background for me even if I upload my own mask, is there a workaround you know?
Try playing around with the mask blur settings. If that doesn't help, I'd take the new hat and manually mask out the background in Photoshop above your initial image. Let me know if you find a better solution.
how can you make it so it shows the iterations/steps where it creates the ship? When I use stable diffusion I only get to see the final result
Do you have a tutorial about inpaint face?
How do you subtract from the mask? I can only make more mask, not remove overspill.
Thanks! Btw, how are you viewing the rendering process as it happens? My version (latest Automatic1111 WebUI) shows the final result only.
In settings, look for "Show show image creation progress every N sampling steps. Set 0 to disable."
@@mattaaron79 this is such a neat feature. Thanks :)
Lol. Turns out I've been doing inpainting completely wrong. Thanks.
Happy to hear the guide helped! What was it that fixed it for you?
@@sebastiankamph I was doing the prompt for the inpainting wrong.
thanks as well. my problem was thinking fill was what I wanted to add something like "fill with your prompt"
Hey, whenever I try to generate a new image using inpainting it gives me the same thing back. It ignores the mask and my mask brush is also black instead of white. Does anyone know what the problem could be?
Same for me. Have you found a solution yet?
@@user-go5ri2yg5f Same, and I get the error "ValueError: Coordinate 'right' is less than 'left'" apparently, Inpainting broke right after this video was made
Hey in my version the inpainting UI keeps the image really small, impossible to see what Im doing. not the image scale itself, just in the UI, and no zoom option. Any ideas
There is a zoom option, sort of. It's a function of your browser. While your mouse pointer is within the UI area, press down the ctrl key and scroll up with the mouse wheel. It will magnify every element of the interface, including the image. You can then scroll back down when you're done. This of course works just the same with any web page.
How to change mark color ?
Hi, I am using the latest version of Automatic1111 but can't seem to find anywhere to load an external mask. Has this feature been removed?
Did everything pretty much like you did, but for whatever reason I can't get inpaint to work. Quite literally nothing happens to my image when I try to inpaint on it; it just generates the same exact picture. Any idea what might be borked on my end? SD still takes it usual amount of time "generating" the image and doesn't show any errors.
Try filling instead of original and only masked instead of whole picture, or variations of these. Maybe that helps
ships, yes, but not one of them looked like a Viking ship..
Still, it was a good explaining of how the program work. 🙂
Hey everyone. I got a big problem with Inpaint where I need help. I mean, the concept sounds simple enough. Create a mask, enter a prompt, let SD do the job.
Problem is: I don't getting it to work what so ever. In the last couple of hours, I read and watched half a dozen tutorials (including this video) on the topic. But no matter what picture I use, what prompt and worst of all independent of all the different settings, I always get the very same result: A blurred area where the mask was. And I don't know what's wrong. Or what I am doing wrong? tried to recreate and experiment with every setting I saw in those tutorials.
Example
1. I let SD create a TXT2IMG with a simple prompt: "empty room with a table". SD gave me exactly what I asked for, so far, so good.
2. I took this picture to the IMG2IMG / Inpaint tab. I even copied over the seed key
3. I want to let SD draw "a big book" onto the table. So I paint a mask of the area, where SD shall add "a big book". Again, I tried for hours experimenting with the differen settings.
4. The result is always the same: Just a blurred area where the mask was. No big book. Or whatever.
I am going nuts here, please help!
Hi ! do you know if can i get the same result in my mask, but changing my image reference? i tried to take the same seed, but doesn't work :/ I try to do this to inpaint a image suite, to convert in video. Thanks for your answer and for your videos !
Hey, not following your question. Feel free to hop on Discord and ask in #help with image references to explain further.
@@sebastiankamph Thanks for your answer, yes i'm going to send on discord, se if i can find some help !
In Masked Content, isn't better to always use "Original", instead of the other options?
Thanks for asking, it depends a little bit of what's there originally. My experience, and of those I've talked with, is that latent noise gives most consistent results, especially when there is no guide to work from (like in the case here, where there's just water). If you paint a crude sketch of a boat and img2img or inpaint from that, that's a different story. Would love to hear your experiences and why you prefer original! This is new to all of us and we're all learning as fast as we can.
Can someone link that tutorial on how to install this? It didn't popup when he said it
when i see "localhost" in the browser - do i understand correctly it's like u said "installed" so it work 100% locally not sending data to the internet?
I have a question. It seems my window has some issue for inpainting. The upper few inch are not targetable, and my mouse lose the target if i move down. I also cant paint the space i want it's always on the wrong hight. What could that be.
I does not go away after restart.
The paint space is is smaler as the actual window.
Do you use a size other than 512x512? Sometimes that can create errors.
@@sebastiankamph I use now an 1024x1024. And oddly the problem is gone. The 512 have the issue.🤔
@@sebastiankamph Found the Issue. Do not use the Microsoft Edge browser...
There is a way to inpaint your own image?
Updated Inpainting guide here: ruclips.net/video/No1_sq-i_5U/видео.html
The FREE Prompt styles I use here:
www.patreon.com/posts/sebs-hilis-79649068
When I try inpainting, it keeps giving me a blurred area that I masked out. My mask is black as well, so not sure thats why? I don't really know how to not make my mask not black.
Yeah, sorry, not sure what might be wrong ;(
my stable diffusion doesnt have the option for uploading mask, why?
Mine as well, seens the interface changed. I guess...
Does anybody know how to upload a mask? What format do I use with Krita or Gimp?
pretty sure it's a PNG
Can do jpg or png. Black/white/grey. No transparency.
@@sebastiankamph Thanks!
what gpu you use?
Rtx 3080
You look like a young Derren Brown 😏
Great tutorials. Do you know how to fix Inpainting? It seems to be broken, and I'm not finding a solution online.
Inpaint "Only masked" gives error "ValueError: Coordinate 'right' is less than 'left'"
Inpaint Whole picture renders an unchanged picture, or perhaps a black screen. It doesn't seem to matter which model of diffuser I use.
same for me...
@@joey4track Disable the DuckDuckgo browser extension, that was the cause for me :)
but how to remove objects?
Your interface looks different than mine.
"It's an older video, but it checks out."
Not trying to embarrass you, but what does "t and p" stand for? 😅
has inpaiting in SD changed at all? cause this totally does not work anymore..
I got a new video on inpainting going up today. Small differences, but no huge changes.
@@sebastiankamph thanks for the reply.. will check it out.. btw, are there any ways to do this as described in this video ruclips.net/video/t7b91-fr09A/видео.html is it possible to do this in auto1111? cause i would be very keen to draw my own shapes and then use that as reference for whatever i want to simulate but cant seem to find way to do that
@@sebastiankamph oh snap. Wait a moment.. I just figured it out! mind blown..
.☮️.✌️.
Argh! My eyes hurt with all that bright-backgroundness!
fuck yeah. Each and every app installed on my Windows (and there are tons of'em) has a dark theme. Can't understand how some people can till stand those unbearable pure white backgrounds.
The way you said automatics. I think it would sound less Swedish if you said Automadics and didn't emphasize anything. Your Swedish dialect is very obvious
Yes, I love being Swedish.
@@sebastiankamph It's nice, but not the language. Couldn't we have just used English instead?
Bro are you really complaining that a Swedish guy sounds Swedish?
@@gmalivuk Apparently