UPDATE! So I found it much easier to create the blank portion of the Canvas through Paint3D which is a standard windows App. So check out this video for a faster and improved way to extend your canvas and make it blank, its a 50 sec short. ua-cam.com/users/shortsL-q_9Cz2ogU?feature=share. The 0 denoise trick sometimes has issues for reasons I can't explain. 0 denoise should be no change but sometimes due to custom models or custom settings it can still change a bit but sometimes it doesn't. UPDATE! It seems that the zoom function is possible because I have a extension called canvas zoom such as the color palete and zoom functionality. I put a video together on how to install that as well as a photo editor called photopea. Check it out here - ua-cam.com/video/-ySxWKKEjAQ/v-deo.html This tool will also allow you to scroll out on the picture to shrink it and fit it in a bigger box so you can outpaint above and blow the image or on all sides. Make sure you click on the image 1 more time after resizing so it will auto adjust the aspect ratio and prevent warping and stretching. Outpainting is a great way to extend your picture and add content that didn't exist as if it was always there. This is a great way to make youtube thumbnails or just extend a picture without stretching everything in the photo or zooming in super close. No downloads for this one but check out my video description for more tutorials including one to install stable diffusion if you haven't done so already. 😁
You can right click on your image to see Canva's shortcuts, in particular "R" to refresh the position of the image, and "F" to drag the image when ultra zoomed (shift scroll).
I'm pretty sure that was from Canvas Zoom. I'm not sure how well that tool is on 1.6 since it just came out but it will give you the pencil, some color mask, and the ability to auto generate the Height and Width dimensions of the image your in painting on.
@@ArtificallyIntelligentAi I installed the canvas zoom extension but the pencil button still isn't there. Could it have been moved or changed since then?
Ya, a high denoise like 0.7 is typically better for this method or you just get blurry colors. I got lucky with 0.4 because of the model I was using and other factors.
The reason the OpenOutpaint isn't working for you is because you didn't use an inpainting model. Once it works, that application is a lot easier to use.
Hi i tried the v1.5 inpainting model with openoutpainting and got the same results. I got it to work in the past but its always been a struggle. Outpainting has always been mixed results for me so I just decided to work around it.
Hey my bad I didn’t know it was from an extension called canvas zoom I made a full video on it and put it in the description and my pinned comment but it’s a regular extension install inside stable diffusion, you don’t need a url
@@ArtificallyIntelligentAi Oh right, I actually had it installed a while back but it was giving me problems so I unchecked it. I'll try it again, thank you!
Hi the pencil came from a ext called canvas zoom. sorry I had to put that it my pinned comment but i recommend just using windows photo viewer. Here is a 50sec video on that. ua-cam.com/users/shortsL-q_9Cz2ogU
at 0:59 in your video you show that you click on the picture and you're able to drag to the left to extend your image. When I do that in my Stable Diffusion, the picture is completely zoomed in and I'm not able to drag a box to expand the image. What am I doing wrong here?
Are you in Img2img? It doesn't work in Inpaint. It's also a bit hard to grab and you have to make sure to click on the square boxes within the lines which are in the middle. When you click on the picture you have the square box outline around the image and in the center of each side of the square there is a solid square box that you need to click on the drag it.
Actually I think you brought up a good point and I forget what extensions I have. I think this is possible with an extension called Canvas Zoom. I'll prob pin a comment about it at the top.
To do that zooming in and out on the input image, you need to put the cursor over the centre of the image - wherever the cursor is , is the point around which the zoom happens.
I also had the extension Canvas Zoom which is a simple extension install in stable diffusion. I forgot it wasn't a standard feature. Or is it? Let me know,.
I been struggling big time with outpainting so I gave up and I just do this now. I have to do it everytime I make a youtube thumbnail to that that 16:9 aspect ratio. It's hard to create and image that just stays on one side so I make the 1x1 and then outpaint to the left.
Hey it’s an extension called canvas zoom. I made a tutorial on it 2 days after this one after realizing it wasn’t a standard feature. It’s a simple install through the extensions tab
@@ArtificallyIntelligentAi That did the trick. Thx. However I had to realize it's not working very well on completing bodies. Like adding a head. Also extending images I had mixed results.
@@sixsix8 I had that issue with wings and I just inpainted the the crooked parts with no prompt. A face is a bit more tricky , usually u need set it to “only masked” when in painting the face to fix a face because u don’t want it to get inspiration from trees and dirt and just the parts inside the mask. Out painting and in painting using “whole image” which this trick uses doesn’t do so great with body parts or faces because it gathers data from the the entire photo. Aiemtrapranuer has a 50 second short that does great job of showing how to inpaint faces or heads and it works really good.
OpenOutPaint works really really well and its super fast, but out of the box none of the default settings really make much sense. You do have to locate the long version of the documentation and read the entire thing, yes really, then it will start working for you.
Thanks! At some point I got it to work but then I forgot how I did it and just couldn't figure it out again so I just use this method. It works pretty good for me and I do it everytime I make a youtube thumbnail.
Interesting method. Have you tried also using the outpainting mk2 in the scripts ? It works well too. When choosing whole picture or only masked it anyways modifies only what is inside the mask, the difference is that with whole picture, if you have for example a 512x512 image , your mask will get just a fraction of that resolution but if you choose only masked, your mask will get the whole 512x512 resolution which could be good to have more detail
And another, probably more important difference is, tha when you choose "Whole picture" your prompt still describes the whole Image while if you choose "Only masked", your prompt describes only the masked part! This is so important because with changing your prompt completely with the "Only masked" option enabled, you are now able to describe a whole new scene which will then be rendered into the masked area! - So everything you describe in your prompt will then make it into your masked area. A wrong understanding of this beahavior is often the reason why people "seem" to get strange inpainting results when they enable the "Only masked" option. So this option is for example extremely useful for repairing crippled hands put on a leg: Mask the crippled hand, turn on "Only masked" and then switch the subject of your prompt to "beautiful detail rich female hand put on a leg". Keep the prompt parts for camera, lighting etc. and choose Masked content "original". Or if you like to render in a whole miniature castle to the feet of someone, also select this option and mask a part at the feet of your model and change the prompt to something like "a beautiful castle with a beautiful woman on the balcony" and voila... ;-)
I haven't, i got to try that. I actually use a combination of this method and control net now so it inpaints on the missing portion using resize and fill with a diff canvas size on controlnet. There is an amazing video by sebastian kamph on it. ua-cam.com/video/Z2d4tFan7cE/v-deo.html . Definitely a must watch.
It was canvas zoom. It will give you the pencil and the auto width and height feature as well as the option to use diff color mask. I would just extend the canvas using microsoft paint 3d i found it easier to do it that way after all the a1111 updates.
Is there a way to extend vertically? I've Canvas Zoom but it only works with Inpaint. Also my resize function is either acting weird or difficult to use lol! BTW If I use inpaint FILL, I just get a blurry mix of colours... (edit: since I couldn't resize the image, I didnt that part in Photoshop, filling the canvas with black, so maybe that's why it doesn't work)
U need to scroll down on the image while editing in img2img and it will shrink but warp the img then click on the img again to auto adjust the ratios. The denoise has to be super high for this to work, my model rpgzovya can get results with .70 but other models required around .80 to .90. Just crank it up till it shows something and keep it on fill.
Hi, I think that was from an extension from canvas zoom but I actually throw it in MS Paint now to extend the canvas which I made a short about. With all the new changes w/ A1111 extending the canvas in the program isn't reliable but the outpainting technique works even better now with SDXL.
Thanks for this insightful tutorial on outpainting. I've been attempting to implement these techniques using the "poor man's outpainting" script, which has been providing satisfactory results, but I'm always looking to improve. I noticed a couple of points in your video that sparked my curiosity. At around the 03:00 mark, the appearance of the woman seems to be influenced by your prompt "a woman in armor...". I've found that sometimes altering the prompt can yield different results. Can this be? At the 07:45 mark, you seem to use a high denoising strength for a small change. In my experience, a lower strength tends to maintain more of the original detail, but perhaps this model requires a higher setting? I'm keen to understand your approach better. Thanks again for the informative video. Looking forward to your response.
The updates to A1111 really messed up that feature. I also made a tutorial 50sec long and put it on the pinned comment at the top on how to edit it in windows which is faster and free.
ok i hAVE been watching all your videos but for some reason when things happen in your videos, in my pc they dont! i use sdxl models, i guess thats the problem but like,,,, isnt there a way to outpaint for sdxl models??? i tried almost everything... no avail can you please help me if you know a way to do this for sdxl
Kinda and yes. stable diffusion works on latent diffusion so a latent upscale is upscaling while reintroducing noise to recreate the finer details. If you started with a pic that was just a blur it does wonders but will slightly change the image. There are actual latent upscalers that do the same thing but choose the settings for you. High steps increase the quality and are necessary for some of the sampling models like karras. It is technically not needed but it will “typically” give u better results.
Deng it is from an extension called canvas zoom which u can easily install inside the stable diffusion extension tab. I actually was going to upload a video tomorrow on it as well as adding a photo editor directly inside stable diffusion
Hey sorry about that it’s hard to keep up with my extensions but I made a video about the extension that does that canvas zoom ua-cam.com/video/QD32dTMemkI/v-deo.html it’s a quick install
@@ArtificallyIntelligentAi you linked the wrong video but it’s okay i just saw your pinned message 😂. i also found a way around not having the extension. i just masked the picture i wanted to outpaint and selected ‘Inpaint not masked’. Anyway thank you this is a good method that i’ll be adding into my workflow when it calls for it
PLEASE GUIDE - What might have gone wrong in my part, even after extending the canvas size, changing the dimensions, and setting denoise to 0...... I pressed generate button. But after some silence nothing happened but there was a short duration of silence with the text "Waiting...." in the generation box left top. Please Photopea result was worse, the Img2img window went on opening for 800 seconds until I forcibly closed. ////Please note, this is the third time I am reinstalling A1111, and in the process of installing the extensions. Totally clueless about why I could not do what you have done.//// (I have a gaming mother board, i9 processer, 64GB Ram and 8GB Nvidea Vram)
HI there! Do you have cuda drivers installed by any chance? and the --xformers argument in your webui-user.bat file? That significantly increases the speed and the decreases the vram needed. Also what dimensions are you using? and does the console (cmd prompt) say Cuda out of memory? I also edit the image in Paint3D for the blank canvas now and it takes me 3 seconds to do, I made a short on it ua-cam.com/users/shortsL-q_9Cz2ogU?feature=share Let me know if you have Cuda drivers and what size pic you were trying to inpaint on and we can troubleshoot from there. Here is a really good site that has an article on the diff of vram usage and speed times for diff optimizers vs. none. stable-diffusion-art.com/speed-up-stable-diffusion/ benchmark results are at the bottom.
@@ArtificallyIntelligentAi Thank You for the response. I have been fiddling the A111 without seeing your reply. I will get back ASAP. ITs 12.30 midnight. Will reply you tomorrow. Thank You again
Happy dear..... Paint 3D is working perfectly with controlnet.(well why photopea doesn't do so?) ??? ///// I have done clean install with a launcher running in advance everytime that has options to load xformers and lowvram for every start ///// I guess that I have perfect Cuda drivers, installed with my basic knowledge of tweaking with the guidance videos and blogs from people like you, but I am not sure, if I have done it properly. I have 2060 Nvidia 8 GB - Can you please help me to make it properly. /// I want to do the outpainting as you did in your video - please guide - I feel ashamed of failing in that because inpainting also then fails automatically ///// MORE than any thing, how to composite an image in A111 using rule of thirds in photography (Eg: Along view of beach, a child in the left third walking on the edge of waters towards right during sunset, the reflections of the bay is visible on the grounds, sun and sea in the background, sun is postitioned in the rith third of the frame in the right top) OPTIONAL REQUESTS:- 1. Why prompting is like a magic wand? CIVITAI samples are not giving a complete idea to create a new one. of my own. I want to edit the styles. csv with camera tpes, lenses, depth of field, frame compositions. Clueless to proceed further 2. Please post a video about wild cards usage with SD prompts in details. 3. Please guide to create images like MIdjourney 4. Do I require Invoke and Prodia to train loras? 5. Is there a complete tutorial about A11 describing all its extensions in detail? 6. I have Adobe Photshop Beta with creative fill, by any chance may I able to use A1111 plugin with that ALL THE ABOVE QUESTIONS ARE OUT OF FRUSTRATIONS AS I AM NOT ABLE TO CREATE WHAT I WANT. INSTEAD I AM CREATING COMETHING SIMILAR TO WHAT OTHERS HAVE DONE. THE KEY FACTORS LIKE PROPER SEEDING, THE SLOT OR BOX FOR MENTIONING THE MODEL HASH. IN ONE WORD, I WANT FREEDOM TO CREATE WHAT I WANT FROM STABLE DIFFUSION A1111. THank You for your patience in reading all my cries.
@@tmpda Hey there. Hopefully I understand it everything but the rule thirds you can accomplish by using control net and you can do quite a few things with it like sketch the subjects outline like a terrible stick figure and it will create the images based on your sketches. You can also use segmentation maps with control net and photobash which is just grabbing the control net preprocessed map the outlines, depth maps, segmentation maps and then photoshop/bash them together and ask stable diffusion to color within those lines or create what you want. You can also use control net to do depth of field or use inpainting to create that effect. They got specific models for DOF as well on civitai. Also if you read a bit on the civitai models there is usually some recommendations on settings. You'll have to create a lora or embedding to get a particular style of your own or if its just a face you can use roop to copy the likeness of the image. You don't need Invoke to train lora or embeddings files infact you don't even need a111 you can do it through google colabs or LeonardoAI for the embeddings/textural inversion. Maybe someday I'll do a video on the wildcards but stable diffusion is about to change bigtime with the SDXL release so I'm going to hold off until I see how that changes things because it's going to be drastic. There is a lot of models that claim to be like midjourney like deliberate but it's subjective and I think midjourney is way better. Deliberate has instructions how to make it better in civitai you have to change your pression setting and add a custom seed. Still not midjourney. The New SDXL thats coming out is catching up to midjourney. I think when people create models for it, it might compare with midjourney. As for the extensions there are way to many for someone to do an all extensions video. By the time they completed it, 2 more would release. This site is a gold mine for everything stable diffusion and has a lot of extension reviews stable-diffusion-art.com/ As for the Photoshop Plugin check out Olivio Sarikas, his Channel is amazing and it's the best ai art channel and he goes over the Photoshop plugin. If your doing AI art you have to check out his channel and Sebastian Kamph. 2 awesome channels.
@@tmpda i wrote all that and I forgot to tell you to check out a tutorial by sebastian kamph that uses control net to do outpainting which is essentially the same thing I did with my outpainting, just a bit more complicated since you need control net installed. It might solve your issues since the canvas is just extended using control net in the background.
Thanks for sharing this! It isn't working for me though. :( I guess I'm doing something wrong. I'll just keep trying. I still haven't figured out how to outpaint. :( . . . update, after watching another one of your videos and after uploading Photopea, I can outpaint in Photopea . . . or inpaint . . . with the Interrogate Clip (had no idea what this was) button and the other instructions! Thank you!! If you come out with a online class, and if I can afford it, I'd totally take it!! You're the bestest estest!! 😊
Awe sorry about that. Ya this one is tricky and the inpaint has to be on "Fill" and the denoise has to be all the way up to 0.80 or higher. I started making the blank part of the canvas on Paint3D (default windows app) because it's more consistent. The zero denoise doesn't work for some images. Fill will tell the system there is nothing to interpret and it's a blank spot so grab from the rest of the image. I clicked on Interrogate clip as a lazy way to fill the prompt with similar details but it's best to make sure you don't add details of anything you don't want into the photo. If it's blurry, crank the denoise up. No online class but if your curious about something just let me know and I'll try and put it a new video.
Was going to do a video on that actually but after I do a video about negative prompts in stable diffusion. I was going to use control net to recreate popular video game scenes. I got the idea becuase someone did whiterun from skyrim using the HED model (soft).
I actually noticed I had an extension called canvas zoom and you can click on the picture and zoom out to make the image smaller than the box so there is room at the top but then you have click on the image one more time and make sure it auto adjusts or you will get a stretched image. It should automatically resize to a locked aspect ratio and prevent warping.
@@ArtificallyIntelligentAi Hey that worked! Biggest issue is how to control where the image goes. Using the cursor before zoomin in helps but its kindof glitchy. I settled for using paint for the time being.
@@ga5518 there is actually a photophea extension that has all the paint features and u use and install it inside stable diffusion. It has ads but it can do things like paint, masking, image layering, and even remove background.
@@ga5518 Here is a video how to install photopea into stable diffusion to use photoshop like functions inside stable diffusion ua-cam.com/video/SpuYEB_nZW0/v-deo.html
Right click on the generated image and select rotate left or right, whichever you wish. Then add the image to img2img and extend left or right. Does that make sense? You don’t have to install a bunch of other extensions or applications to extend vertically.
Just masked and entire picture have nothing to do with where it fills. They determine what's considered by the model. So in this case you want it to look at the entire picture to understand what to place in the black area.
Does it work for you with original instead of fill in the blank part of the canvas? My understanding is a1111 is telling it to not get inspiration from the blank portion which is why original doesn’t work great. If it works for you then awesome. I’ll do more testing
Ya every update made it worse, just throw it in ms paint 3d which is a default windows app and then click edit and pull the canvas and reupload the image. I found it was easier and I made a 50 second UA-cam short on it
@@ArtificallyIntelligentAi Yeah, I was just resizing in img2img with a denoising of 5 or so, but I've found it easier to toss it in Gimp and change the canvas size, because most of the time I don't want it evenly expanded, I usually want just one side expanded, or one side more than the other to center the subject.
@@Jaradis Ya, unfortunately it got harder to do with each update. I put out a short doing this whole process in 60 seconds starting with Paint 3d since all windows users should have that but GIMP works great.
Ya it's prob great but you might violate their terms of service if your not outpainting on a image made in fire fly. I don't know how they regulate that but I'm always a bit scared with that stuff.
Among the stable diffusion features, the least effective one is outpainting. It's a feature that I've never been able to execute correctly, even when I've attempted it myself. This outpainting is simply pointless, so I would like it to be eliminated from stable diffusion.
Playground Ai also has a canvas now, and allows for 1000 generations per day. I would say the quality is very close to leonardo now that they have custom models.
Cool to know. I actually like playground ai better than Leonardo but I knew leonardo had that AI cavas thing. I got to check out playground AI's tool. I did a video on them a couple months back.
"At 00:54, I couldn't find the pen when I put the image into img2img, how can I get that pen? Please answer me, thank you very much."
@@Knox420 What's the extension called?
0:55 I don't have a pen to click for the canvas
Same here. I think the program changed
What extension do you use for the paintbrush option to extend the picture?
I don't have that brush in my webui either
Nor do I.
Canvas Zoom extension
this is incorrect@@vb6code
@@vb6code i added it but still no brush (
On the img2img tab I don't see the pencil for editing the canvas. Is that an extension?
Me neither!!
how did you automatically choose the resolution I don't have that button?
UPDATE! So I found it much easier to create the blank portion of the Canvas through Paint3D which is a standard windows App. So check out this video for a faster and improved way to extend your canvas and make it blank, its a 50 sec short. ua-cam.com/users/shortsL-q_9Cz2ogU?feature=share. The 0 denoise trick sometimes has issues for reasons I can't explain. 0 denoise should be no change but sometimes due to custom models or custom settings it can still change a bit but sometimes it doesn't.
UPDATE! It seems that the zoom function is possible because I have a extension called canvas zoom such as the color palete and zoom functionality. I put a video together on how to install that as well as a photo editor called photopea. Check it out here - ua-cam.com/video/-ySxWKKEjAQ/v-deo.html
This tool will also allow you to scroll out on the picture to shrink it and fit it in a bigger box so you can outpaint above and blow the image or on all sides. Make sure you click on the image 1 more time after resizing so it will auto adjust the aspect ratio and prevent warping and stretching.
Outpainting is a great way to extend your picture and add content that didn't exist as if it was always there. This is a great way to make youtube thumbnails or just extend a picture without stretching everything in the photo or zooming in super close. No downloads for this one but check out my video description for more tutorials including one to install stable diffusion if you haven't done so already. 😁
You can right click on your image to see Canva's shortcuts, in particular "R" to refresh the position of the image, and "F" to drag the image when ultra zoomed (shift scroll).
whats the name of the "get width and height button" please ?
@@高茂林-d1y I got that from the canvas zoom extension
that pencil thingy on the upper right of img2img windows doesn't exist anymore innit? is there an extension for it?
I'm pretty sure that was from Canvas Zoom. I'm not sure how well that tool is on 1.6 since it just came out but it will give you the pencil, some color mask, and the ability to auto generate the Height and Width dimensions of the image your in painting on.
@@ArtificallyIntelligentAi I installed the canvas zoom extension but the pencil button still isn't there. Could it have been moved or changed since then?
Hmm mine didn't work with the denoising strength at 0.4, I had to change it back to 0.7, thanks for the tut
Ya, a high denoise like 0.7 is typically better for this method or you just get blurry colors. I got lucky with 0.4 because of the model I was using and other factors.
Hey is that "normal" that I don't have the "painting" icon for changing color at 2:04 by any chance ?
Yes and my bad it’s from an app called canvas zoom. I put the video on how to install it I. The description and it literally takes 30 seconds.
@@ArtificallyIntelligentAiAh check I was missing this too, thanks!
hey Mate great video, whats the name of the "get width and height button" please ?
Anyone know where I can find the extension, addon?
Sorry for the late reply but it was canvas zoom that gave me that get height and width button. It looks a bit diff now but it does the same thing.
Now I finally know what outpainting is LOL - thx! Theres just too many things in this software to discover and master
The reason the OpenOutpaint isn't working for you is because you didn't use an inpainting model. Once it works, that application is a lot easier to use.
Yeah, unfortunately OpenOutpaint has a bad rep because it isn't a great out of the box experience, you need to work a little bit
Hi i tried the v1.5 inpainting model with openoutpainting and got the same results. I got it to work in the past but its always been a struggle. Outpainting has always been mixed results for me so I just decided to work around it.
how did you get the colour picker? I don't have it in mine.
Hey my bad I didn’t know it was from an extension called canvas zoom I made a full video on it and put it in the description and my pinned comment but it’s a regular extension install inside stable diffusion, you don’t need a url
@@ArtificallyIntelligentAi Oh right, I actually had it installed a while back but it was giving me problems so I unchecked it. I'll try it again, thank you!
Cannot extend the canvas in img2img tab, the pencil button isn't there.
Hi the pencil came from a ext called canvas zoom. sorry I had to put that it my pinned comment but i recommend just using windows photo viewer. Here is a 50sec video on that. ua-cam.com/users/shortsL-q_9Cz2ogU
Ah I see I see, well, I am just getting error messages in SD whenever I try to use the feature with the model I want :(@@ArtificallyIntelligentAi
at 0:59 in your video you show that you click on the picture and you're able to drag to the left to extend your image. When I do that in my Stable Diffusion, the picture is completely zoomed in and I'm not able to drag a box to expand the image. What am I doing wrong here?
Are you in Img2img? It doesn't work in Inpaint. It's also a bit hard to grab and you have to make sure to click on the square boxes within the lines which are in the middle. When you click on the picture you have the square box outline around the image and in the center of each side of the square there is a solid square box that you need to click on the drag it.
Actually I think you brought up a good point and I forget what extensions I have. I think this is possible with an extension called Canvas Zoom. I'll prob pin a comment about it at the top.
I just updated Auto1111 yesterday and I don't have that color wheel button in the inpainting tab :/
Hi, sorry for the late reply, it was cavans zoom that gives the color options.
Outpainting MK2 is what I usually use. It is significantly easier than this and works pretty well. Check it out, I think you would like it.
I actually had another person recommend that. I got to check it out.
To do that zooming in and out on the input image, you need to put the cursor over the centre of the image - wherever the cursor is , is the point around which the zoom happens.
I also had the extension Canvas Zoom which is a simple extension install in stable diffusion. I forgot it wasn't a standard feature. Or is it? Let me know,.
Cool stuff. Been struggling with other forms of out painting. Excited to try this method out!
I been struggling big time with outpainting so I gave up and I just do this now. I have to do it everytime I make a youtube thumbnail to that that 16:9 aspect ratio. It's hard to create and image that just stays on one side so I make the 1x1 and then outpaint to the left.
HELP! I don't have the option to change brush color 1:52. I can only change the brush size. How do I get this?
Hey it’s an extension called canvas zoom. I made a tutorial on it 2 days after this one after realizing it wasn’t a standard feature. It’s a simple install through the extensions tab
@@ArtificallyIntelligentAi Thank you so much. I was struggling with this. Liked and subbed on ya.
I am just getting blurry colors where I outpaint using this technique. Any ideas what I am doing wrong?
If the denoising strength is to low is the main reason , crank it up to 0.9 and see if that’ works. Also make sure in painting is set to fill.
@@ArtificallyIntelligentAi That did the trick. Thx. However I had to realize it's not working very well on completing bodies. Like adding a head. Also extending images I had mixed results.
@@sixsix8 I had that issue with wings and I just inpainted the the crooked parts with no prompt. A face is a bit more tricky , usually u need set it to “only masked” when in painting the face to fix a face because u don’t want it to get inspiration from trees and dirt and just the parts inside the mask. Out painting and in painting using “whole image” which this trick uses doesn’t do so great with body parts or faces because it gathers data from the the entire photo. Aiemtrapranuer has a 50 second short that does great job of showing how to inpaint faces or heads and it works really good.
I love those meme gifs you put in this
❤Thanks!
Thank you again! Love the way you always go full in detail. Good luck to your future!
Also appreciate the kind words and thank you very much!
OpenOutPaint works really really well and its super fast, but out of the box none of the default settings really make much sense. You do have to locate the long version of the documentation and read the entire thing, yes really, then it will start working for you.
Thanks! At some point I got it to work but then I forgot how I did it and just couldn't figure it out again so I just use this method. It works pretty good for me and I do it everytime I make a youtube thumbnail.
no pencil button ?cant see thatg
Good work! Clear instruction with no fluff or ego.
Fantastic video! You explain things in an intuitive way and easy to understand. Great job!
Thanks!
Interesting method. Have you tried also using the outpainting mk2 in the scripts ? It works well too.
When choosing whole picture or only masked it anyways modifies only what is inside the mask, the difference is that with whole picture, if you have for example a 512x512 image , your mask will get just a fraction of that resolution but if you choose only masked, your mask will get the whole 512x512 resolution which could be good to have more detail
And another, probably more important difference is, tha when you choose "Whole picture" your prompt still describes the whole Image while if you choose "Only masked", your prompt describes only the masked part!
This is so important because with changing your prompt completely with the "Only masked" option enabled, you are now able to describe a whole new scene which will then be rendered into the masked area! - So everything you describe in your prompt will then make it into your masked area.
A wrong understanding of this beahavior is often the reason why people "seem" to get strange inpainting results when they enable the "Only masked" option.
So this option is for example extremely useful for repairing crippled hands put on a leg:
Mask the crippled hand, turn on "Only masked" and then switch the subject of your prompt to "beautiful detail rich female hand put on a leg". Keep the prompt parts for camera, lighting etc. and choose Masked content "original".
Or if you like to render in a whole miniature castle to the feet of someone, also select this option and mask a part at the feet of your model and change the prompt to something like "a beautiful castle with a beautiful woman on the balcony" and voila... ;-)
I haven't, i got to try that. I actually use a combination of this method and control net now so it inpaints on the missing portion using resize and fill with a diff canvas size on controlnet. There is an amazing video by sebastian kamph on it. ua-cam.com/video/Z2d4tFan7cE/v-deo.html . Definitely a must watch.
I'm very new to SD and will certainly be trying this, thanks :)
ummm i don't have the pencil icon latest version, what add-on is giving you that feature
It was canvas zoom. It will give you the pencil and the auto width and height feature as well as the option to use diff color mask. I would just extend the canvas using microsoft paint 3d i found it easier to do it that way after all the a1111 updates.
@@ArtificallyIntelligentAi canvas zoon i have but no pencil, wonder if version issues, thank you still
How to get button from 1:31? Thank you!
It’s from an extension called canvas zoom and it’s the auto height width button. Here is a tutorial on it ua-cam.com/video/-ySxWKKEjAQ/v-deo.html
Is there a way to extend vertically? I've Canvas Zoom but it only works with Inpaint. Also my resize function is either acting weird or difficult to use lol!
BTW If I use inpaint FILL, I just get a blurry mix of colours... (edit: since I couldn't resize the image, I didnt that part in Photoshop, filling the canvas with black, so maybe that's why it doesn't work)
U need to scroll down on the image while editing in img2img and it will shrink but warp the img then click on the img again to auto adjust the ratios. The denoise has to be super high for this to work, my model rpgzovya can get results with .70 but other models required around .80 to .90. Just crank it up till it shows something and keep it on fill.
@@ArtificallyIntelligentAi Got it, tyvm!! I hope they change the resize tool lol
I have not the pencil to extend the size of the picture in i2i... Do you know how i can do it?
Hi, I think that was from an extension from canvas zoom but I actually throw it in MS Paint now to extend the canvas which I made a short about. With all the new changes w/ A1111 extending the canvas in the program isn't reliable but the outpainting technique works even better now with SDXL.
Thanks for this insightful tutorial on outpainting. I've been attempting to implement these techniques using the "poor man's outpainting" script, which has been providing satisfactory results, but I'm always looking to improve.
I noticed a couple of points in your video that sparked my curiosity. At around the 03:00 mark, the appearance of the woman seems to be influenced by your prompt "a woman in armor...". I've found that sometimes altering the prompt can yield different results. Can this be?
At the 07:45 mark, you seem to use a high denoising strength for a small change. In my experience, a lower strength tends to maintain more of the original detail, but perhaps this model requires a higher setting? I'm keen to understand your approach better.
Thanks again for the informative video. Looking forward to your response.
Sadly in my version I don’t have this pencil button. How can I do it with the actual version (dl one week ago)?
The updates to A1111 really messed up that feature. I also made a tutorial 50sec long and put it on the pinned comment at the top on how to edit it in windows which is faster and free.
@@ArtificallyIntelligentAi the updates have messed up many other features too like controlnet and dreambooth
ok i hAVE been watching all your videos but for some reason when things happen in your videos, in my pc they dont! i use sdxl models, i guess thats the problem but like,,,, isnt there a way to outpaint for sdxl models??? i tried almost everything... no avail
can you please help me if you know a way to do this for sdxl
In the new version of A1111 there is no pencil in IMG2IMG. That function is removed. How would you do this without it?
I put a short tutorial in the description and in my tagged comment at the top it’s 30 sec
Video needs a million views
Hopefully it gets it !
So latent upscaling just means img2img with very very low denoise, and very high steps?
Kinda and yes. stable diffusion works on latent diffusion so a latent upscale is upscaling while reintroducing noise to recreate the finer details. If you started with a pic that was just a blur it does wonders but will slightly change the image. There are actual latent upscalers that do the same thing but choose the settings for you. High steps increase the quality and are necessary for some of the sampling models like karras. It is technically not needed but it will “typically” give u better results.
Did not work for me. I ended up with blurriness on the spots where i painted.
I dont have that color palette icon in my inpaint, how did you get that? Im using A1111 v1.3 (the latest)
Deng it is from an extension called canvas zoom which u can easily install inside the stable diffusion extension tab. I actually was going to upload a video tomorrow on it as well as adding a photo editor directly inside stable diffusion
I lost it at 2:42 the John Cena meme was on point!
Haha that thing always makes me laugh out loud. 150% acting
I don't have the color palette thing?
Hey sorry about that it’s hard to keep up with my extensions but I made a video about the extension that does that canvas zoom ua-cam.com/video/QD32dTMemkI/v-deo.html it’s a quick install
@@ArtificallyIntelligentAi you linked the wrong video but it’s okay i just saw your pinned message 😂. i also found a way around not having the extension. i just masked the picture i wanted to outpaint and selected ‘Inpaint not masked’. Anyway thank you this is a good method that i’ll be adding into my workflow when it calls for it
PLEASE GUIDE - What might have gone wrong in my part, even after extending the canvas size, changing the dimensions, and setting denoise to 0...... I pressed generate button. But after some silence nothing happened but there was a short duration of silence with the text "Waiting...." in the generation box left top. Please Photopea result was worse, the Img2img window went on opening for 800 seconds until I forcibly closed. ////Please note, this is the third time I am reinstalling A1111, and in the process of installing the extensions. Totally clueless about why I could not do what you have done.//// (I have a gaming mother board, i9 processer, 64GB Ram and 8GB Nvidea Vram)
HI there! Do you have cuda drivers installed by any chance? and the --xformers argument in your webui-user.bat file? That significantly increases the speed and the decreases the vram needed. Also what dimensions are you using? and does the console (cmd prompt) say Cuda out of memory? I also edit the image in Paint3D for the blank canvas now and it takes me 3 seconds to do, I made a short on it ua-cam.com/users/shortsL-q_9Cz2ogU?feature=share Let me know if you have Cuda drivers and what size pic you were trying to inpaint on and we can troubleshoot from there. Here is a really good site that has an article on the diff of vram usage and speed times for diff optimizers vs. none. stable-diffusion-art.com/speed-up-stable-diffusion/ benchmark results are at the bottom.
@@ArtificallyIntelligentAi Thank You for the response. I have been fiddling the A111 without seeing your reply. I will get back ASAP. ITs 12.30 midnight. Will reply you tomorrow. Thank You again
Happy dear..... Paint 3D is working perfectly with controlnet.(well why photopea doesn't do so?) ??? ///// I have done clean install with a launcher running in advance everytime that has options to load xformers and lowvram for every start
///// I guess that I have perfect Cuda drivers, installed with my basic knowledge of tweaking with the guidance videos and blogs from people like you, but I am not sure, if I have done it properly. I have 2060 Nvidia 8 GB - Can you please help me to make it properly.
/// I want to do the outpainting as you did in your video - please guide - I feel ashamed of failing in that because inpainting also then fails automatically
///// MORE than any thing, how to composite an image in A111 using rule of thirds in photography (Eg: Along view of beach, a child in the left third walking on the edge of waters towards right during sunset, the reflections of the bay is visible on the grounds, sun and sea in the background, sun is postitioned in the rith third of the frame in the right top)
OPTIONAL REQUESTS:-
1. Why prompting is like a magic wand? CIVITAI samples are not giving a complete idea to create a new one. of my own. I want to edit the styles. csv with camera tpes, lenses, depth of field, frame compositions. Clueless to proceed further
2. Please post a video about wild cards usage with SD prompts in details.
3. Please guide to create images like MIdjourney
4. Do I require Invoke and Prodia to train loras?
5. Is there a complete tutorial about A11 describing all its extensions in detail?
6. I have Adobe Photshop Beta with creative fill, by any chance may I able to use A1111 plugin with that
ALL THE ABOVE QUESTIONS ARE OUT OF FRUSTRATIONS AS I AM NOT ABLE TO CREATE WHAT I WANT. INSTEAD I AM CREATING COMETHING SIMILAR TO WHAT OTHERS HAVE DONE. THE KEY FACTORS LIKE PROPER SEEDING, THE SLOT OR BOX FOR MENTIONING THE MODEL HASH. IN ONE WORD, I WANT FREEDOM TO CREATE WHAT I WANT FROM STABLE DIFFUSION A1111.
THank You for your patience in reading all my cries.
@@tmpda Hey there. Hopefully I understand it everything but the rule thirds you can accomplish by using control net and you can do quite a few things with it like sketch the subjects outline like a terrible stick figure and it will create the images based on your sketches. You can also use segmentation maps with control net and photobash which is just grabbing the control net preprocessed map the outlines, depth maps, segmentation maps and then photoshop/bash them together and ask stable diffusion to color within those lines or create what you want. You can also use control net to do depth of field or use inpainting to create that effect. They got specific models for DOF as well on civitai. Also if you read a bit on the civitai models there is usually some recommendations on settings. You'll have to create a lora or embedding to get a particular style of your own or if its just a face you can use roop to copy the likeness of the image.
You don't need Invoke to train lora or embeddings files infact you don't even need a111 you can do it through google colabs or LeonardoAI for the embeddings/textural inversion.
Maybe someday I'll do a video on the wildcards but stable diffusion is about to change bigtime with the SDXL release so I'm going to hold off until I see how that changes things because it's going to be drastic.
There is a lot of models that claim to be like midjourney like deliberate but it's subjective and I think midjourney is way better. Deliberate has instructions how to make it better in civitai you have to change your pression setting and add a custom seed. Still not midjourney. The New SDXL thats coming out is catching up to midjourney. I think when people create models for it, it might compare with midjourney.
As for the extensions there are way to many for someone to do an all extensions video. By the time they completed it, 2 more would release. This site is a gold mine for everything stable diffusion and has a lot of extension reviews stable-diffusion-art.com/
As for the Photoshop Plugin check out Olivio Sarikas, his Channel is amazing and it's the best ai art channel and he goes over the Photoshop plugin. If your doing AI art you have to check out his channel and Sebastian Kamph. 2 awesome channels.
@@tmpda i wrote all that and I forgot to tell you to check out a tutorial by sebastian kamph that uses control net to do outpainting which is essentially the same thing I did with my outpainting, just a bit more complicated since you need control net installed. It might solve your issues since the canvas is just extended using control net in the background.
Thanks for sharing this! It isn't working for me though. :( I guess I'm doing something wrong. I'll just keep trying. I still haven't figured out how to outpaint. :( . . . update, after watching another one of your videos and after uploading Photopea, I can outpaint in Photopea . . . or inpaint . . . with the Interrogate Clip (had no idea what this was) button and the other instructions! Thank you!! If you come out with a online class, and if I can afford it, I'd totally take it!! You're the bestest estest!! 😊
Awe sorry about that. Ya this one is tricky and the inpaint has to be on "Fill" and the denoise has to be all the way up to 0.80 or higher. I started making the blank part of the canvas on Paint3D (default windows app) because it's more consistent. The zero denoise doesn't work for some images. Fill will tell the system there is nothing to interpret and it's a blank spot so grab from the rest of the image. I clicked on Interrogate clip as a lazy way to fill the prompt with similar details but it's best to make sure you don't add details of anything you don't want into the photo. If it's blurry, crank the denoise up. No online class but if your curious about something just let me know and I'll try and put it a new video.
Can you do videos on multi-control net uses? Or interactions with other features/extensions from it?
Was going to do a video on that actually but after I do a video about negative prompts in stable diffusion. I was going to use control net to recreate popular video game scenes. I got the idea becuase someone did whiterun from skyrim using the HED model (soft).
Great video thanks for sharing!
Thanks for the appreciation!
Best way for me, is Hua, and mask parts that you don't want changed
How do you extend vertically?
I actually noticed I had an extension called canvas zoom and you can click on the picture and zoom out to make the image smaller than the box so there is room at the top but then you have click on the image one more time and make sure it auto adjusts or you will get a stretched image. It should automatically resize to a locked aspect ratio and prevent warping.
@@ArtificallyIntelligentAi Hey that worked! Biggest issue is how to control where the image goes. Using the cursor before zoomin in helps but its kindof glitchy. I settled for using paint for the time being.
@@ga5518 there is actually a photophea extension that has all the paint features and u use and install it inside stable diffusion. It has ads but it can do things like paint, masking, image layering, and even remove background.
@@ga5518 Here is a video how to install photopea into stable diffusion to use photoshop like functions inside stable diffusion ua-cam.com/video/SpuYEB_nZW0/v-deo.html
Right click on the generated image and select rotate left or right, whichever you wish. Then add the image to img2img and extend left or right. Does that make sense? You don’t have to install a bunch of other extensions or applications to extend vertically.
The reason outpainting doesn't work for you is because you have to use an inpainting model. With base models the results are usually wonky.
Just masked and entire picture have nothing to do with where it fills. They determine what's considered by the model. So in this case you want it to look at the entire picture to understand what to place in the black area.
Does it work for you with original instead of fill in the blank part of the canvas? My understanding is a1111 is telling it to not get inspiration from the blank portion which is why original doesn’t work great. If it works for you then awesome. I’ll do more testing
Cool hack!
Thanks!
Does not work anymore, there is no option to edit the canvas with the pencil
i think outpaint mk2 script is pretty good, have used it a couple times
hate how its limited to 256 pixels on each side.
Got to try that out.
@@trippybruh1592 yeah but you can do it in incremements
Expanding the canvas doesn't work for me the way you show. The program just hangs forever never producing the image.
Ya every update made it worse, just throw it in ms paint 3d which is a default windows app and then click edit and pull the canvas and reupload the image. I found it was easier and I made a 50 second UA-cam short on it
@@ArtificallyIntelligentAi Yeah, I was just resizing in img2img with a denoising of 5 or so, but I've found it easier to toss it in Gimp and change the canvas size, because most of the time I don't want it evenly expanded, I usually want just one side expanded, or one side more than the other to center the subject.
@@Jaradis Ya, unfortunately it got harder to do with each update. I put out a short doing this whole process in 60 seconds starting with Paint 3d since all windows users should have that but GIMP works great.
"click outside twice" that's man who has suffer too much in the past
sooo its an inpainting video then ?
Inpainting for your outpainting lol. Yes, it is but it gets you an outpainting result which is extending your image.
This feature is now removed.
(Little pencil box)
You mean The """"Easy""""" Way to Outpaint in Stable Diffusion
How hard is it to find a proper inpainting tutorial for beginners... smh..
Pitcher
they added in and out painting prompts in photoshop beta so.....
Ya it's prob great but you might violate their terms of service if your not outpainting on a image made in fire fly. I don't know how they regulate that but I'm always a bit scared with that stuff.
@@ArtificallyIntelligentAi true already got one
Among the stable diffusion features, the least effective one is outpainting. It's a feature that I've never been able to execute correctly, even when I've attempted it myself. This outpainting is simply pointless, so I would like it to be eliminated from stable diffusion.
Playground Ai also has a canvas now, and allows for 1000 generations per day. I would say the quality is very close to leonardo now that they have custom models.
Cool to know. I actually like playground ai better than Leonardo but I knew leonardo had that AI cavas thing. I got to check out playground AI's tool. I did a video on them a couple months back.