How to do Outpainting without size limits in A1111 Img2Img with ControlNet [Generative Fill w SD]!
ฝัง
- เผยแพร่เมื่อ 30 มิ.ย. 2024
- #aiart, #stablediffusiontutorial, #automatic1111
This tutorial walks you through how to Outpaint any image by expanding its borders and filling in details in the extra space outside of your original image, similar to the generative fill functionality of photoshop. We will also walk through how to unlock the 2048 x 2048 image size limits of Automatic 1111 by using the super-secret-ultimate "Limit Break" technique. =P
NOTE: In the video, I showed how to outpaint in the vertical direction first, then the horizontal direction. But you can just as well outpaint in the horizontal direction first, then vertical, if it makes sense for your image.
Here are a few of my other videos that I reference in this video which might be helpful as reference:
Inpaint Anything - Part 1, • How to change ANYTHING...
Inpaint Anything - Part 2, • How to change ANYTHING...
Chapters
00:00 Intro
00:44 Outpainting with Control Net
00:57 Overview of Video Topics
01:23 Need to install Control Net
02:24 Detailed walk-through of all parameters and settings
03:02 Resize Width and Height, expand image vertically first
03:20 Discussion of the effect of different Height changes
06:21 Batch count vs. Batch size
09:08 Detailed walk-through of ControlNet settings
10:14 How End Control Steps affect the output
11:50 Increasing the Height again
13:00 Outpaint in the horizontal direction
14:15 Explaining the “Limit Break” beyond 2048 x 2048
15:16 How to change maximum Width and Height values
17:25 Inpainting step to fix the image
18:10 How to fix CUDA out of memory error
18:31 Final Image!
Useful links
Control Net Installation Instructions:
github.com/Mikubill/sd-webui-...
Hugging Face Control Net models for v1.1.
huggingface.co/lllyasviel/Con...
Control Net Inpainting:
github.com/lllyasviel/Control...
**If you enjoy my videos, consider supporting me on Ko-fi**
ko-fi.com/keyboardalchemist
I like how you explain everything in details and shows how changes in values changes the results with examples.
Thanks for the videos you upload.
They are really helpful.
Hope to see more videos from you.
You're welcome! I'm glad it was helpful.
your tests are so thorough it makes the learning really great. i hope to see more vids from you. thank you
Thank you for your support! Stay tuned for more videos.
This was super clear to follow. I'll definitely be taking advantage of this function when creating desktop wallpapers. Thanks!!
I'm glad to hear it was helpful for you. Thanks for tuning in!
thank you very clear ! Looking forward for more tutorials from you
Glad it was helpful! More to come!
Thank you very much. I really like your teaching method.
You are welcome!
another solid tutorial, the reason you were getting random girls appear in your outpainting was because the kept the 1girl in the positive prompt btw
Thank you, I'm glad you like the video! Yeah, I agree, the positive prompt plays a large part in getting those artifacts, but I noticed that it works in conjunction with the spacing (i.e., if you do not give it enough space, it won't show up).
Nice! Thank you!
You're welcome!
I have found that when we set Ending Control Step to 0.5 we give AI more liberty so that it can be creative after 0.5 value.
It is a good idea when we need a specific pose of a character at the same time some changes to it too. More from a creative angle.
That's good to know. Thank you for sharing!
super workflow, easy to make this by my self, after i watch this Video, Abo your chanell.👍👍
I'm glad you liked the video. Thanks for watching!
Excellent tutorial and well explained. Subbed!! Looking forward for more tutorials.
Can Loras, be used on a specific part of a generated image with the inpaint/inpaint sketch, for ex: I want to change only the arms like a cyborg or having a prosthetic arm kind
Thanks for the sub! Yes you can. The LORAs that you put into your prompt will be applied when you inpaint. I have two videos on how to do inpainting, check those out to get some ideas. Cheers!
@@KeyboardAlchemist Thank you very much. I have checked out the videos and working on some images to get the hang of it.
Just incredible! Q: How do you have the Lora tab in the interface of img2img?
I think it was there by default. If it's not, you can just create a LORA folder in your 'models' subfolder that is under your main stable diffusion install folder.
Just found your vids today, needless to say a new subscriber now - great work and clean info.
Quick question, is it possible to outpaint in only one direction, ie add some to just the left, that would be great for improving composition.
Check out the poor mans outpainting script in img2img :)
Thanks for your support! I'm glad you liked my videos. This script looks like a helpful tool. Thank you for the tip!
I hear Bill Evan’s and I fucking love it
love it! thanks! But how can I expand an image that wasnt created with SD
Great video! But what about the palm tree behind her head? Where does it end? :D
Honestly, no idea. =P
also idk if this is 100% right but I think you can avoid the CUDA memory issue by using tiled diffusion when rendering
I'm glad you brought this up! You are correct, tiled diffusion is a great way to do img2img latent upscaling. I plan on covering it in a future video. Cheers!
hi I just found this tutorial which I found very useful. Qns though: when I use controlnet to outpaint, it changes the original image also.. not just outpainting.. what am I doing wrongly? Do I need to adjust the denoise strength?
SAME issue HALP
I have the same problem, did you find the issue ?
i ve try what you explain, but was too long on my pc. More easy to use inpainting, do the inpaint brush (sorry for my bad english) on each border you want he create. Up the level of image. And it do outpainting^^.
so creamy
Thanks, but I get this error 'TypeError: 'NoneType' object is not iterable' if I use Resize and Fill' in controlnet!!
why i follow all your steps but it is not outpainting, but keep scaling up my photo and crop out?
Make sure you are setting Resize Mode to 'Resize and Fill' and only try to extend one dimension at a time. Thanks for watching!
ISSUE, did the same steps as you BUT original image keeps changing as i generate!? why
Until it didn't.. i'm confused
@@Knox420 I have the same problem, any idea how you did resolve ?
Did you change the seeds? You should use the same seeds to prevents image for changing
What gpu do you use for tutorial?
Nvidia 3060Ti 8GB. It's not fast, but it gets the job done.
Doesn't work. Just inpaints a very blurry noticable bar at the edges.
how do you inpaint stuff out of the picture?
Check out my Inpaint video here: th-cam.com/video/k8FfCicu5G8/w-d-xo.html. I explain it in detail in the video.
I take this error and cant solve it, because I have no info coding, python etc. TypeError: Cannot convert a MPS Tensor to float64 dtype as the MPS framework doesn't support float64. Please use float32 instead.
1:18 how did you open that window?
I believe that is just a screenshot opened and put in the video for the viewers to understand.
Yeah, this. It's just my video editing. It would be cool if A1111 had some kind of split screen function though.
@@KeyboardAlchemist but how do I get there?
@@poraomaneiro from either txt2img or img2img, you can scroll down until you see ControlNet and click on the triangle on the right.
Or just use openoutpaint, no limit, way better quality