STOP making BORING AI art
HTML-код
- Опубликовано: 9 июн 2024
- Full guide for Stable Diffusion inpainting, outpainting, image to image, details and masking. Explaining what everything does and how to use it best!
Create great images with storytelling and no limitations!!
------------ Links used in the VIDEO ---------
Unlimited AI images, models and more!!!: civitai.com/
Best upscaler (4x-UltraSharp): upscale.wiki/wiki/Model_Database
------------ Social Media ---------
-Instagram: / not4talent_ai
-Twitter: / not4talent
This is a really cool way to mix your images and put your desired characters on them!
Learn how here:
• INSTANTLY improve your...
Make sure to subscribe if you want to learn about AI and grow with the community as we surf the AI wave :3
#aiairt #digitalart #automatic1111#stablediffusion #ai #free #tutorial #betterart #goodimages #sd #digitalart #artificialintelligence #inpainting #outpainting #img2img
0:00 intro
0:13 Img2img Tab
0:26 Sending the image to i2i
0:47 What and how
1:13 Denoising Strength
1:27 Using img2img
1:46 Finding the right Denoising Strength
1:58 Aspect Ratio and Resize Modes
3:35 Adjusting the composition
3:56 Img2Img for inspiration and variations
4:15 Denoising Strength Curve
4:27 Interrogate Clip/DeepBooru
4:45 Inpainting intro
5:15 Prompting accordingly
5:33 Masking
6:18 What is inpainting?
6:29 Masked content
7:46 Inpaint Area
8:33 Step By Step face fix
9:28 Controling the result
9:43 Inpaint Sketch
10:16 Not mask transparency
10:27 Adjusting
10:33 A quick upscale
10:48 You need the next video :3
![Natti Natasha - Quiéreme Menos [Official Video]](http://i.ytimg.com/vi/R57RrTjRGxk/mqdefault.jpg)
This is what many, many people overlook. The *TRUE* AI art comes from inpainting, because txt2img is just boring, even when doing plain img2img with ControlNet. With Inpainting, you can create a really professional looking art and remove all AI-generated artifacts that immediately give away where the low quality details came from :)
Also, keep in mind, that when inpainting with Only Masked, it is important to use the correct upscaler for img2img (available in Upscaling page in the settings).
Side note: inpaint already upscaled image (adjust the mask size too), because upscaling it as a last step defeats the whole purpose of detailed inpainting ;)
Yes!! I agree, this puts the "art" in AI art. As it is you who control the final result. And nice side notes! I'll pin the comment for people to see! Thank you
@@Not4Talent_AI Thanks :D I also forgot to mention something important about the mask size: if you *really* care about the details, but you don't care about going above 1080p images, e.g. creating standard wallpapers (depends on the size of the areas you want to mask, so some big images might still work well), it's best to set your image size to 1024x1024 (will be slow on weaker GPUs if you use high denoise + tile_resample ControlNet combo) and try to mask areas, that are close to 512x512 or smaller with higher padding.
The reason for this is to let SD regenerate the masked content as upscaled with twice the size, then scale it back down and this step will reduce the amount of low quality details, so this is why choosing the img2img upscaler is important. Another part of the story is deep diving into the Upscale Model Database - you will need to download the right upscaler for masked area, there are models specialized in different tasks, so that's a must-have if you *really* care :D - you can add img2img upscaler to the quicksettings with "upscaler_for_img2img"
In general you will always want your masked content to be lower than target image size to let the upscaler do the detailing work, but this is a bit of an off-topic, because this mostly matters if you inpaint the entire image piece by piece, which is a very long process, that can take even up to 3 days.
To visualize what I mean, and if youtube doesn't hide the comment because of links, here is a before-and-after example:
- base Image - i.imgur.com/JnJx1xw.png
- upscaled, *then* fully inpainted - i.imgur.com/bV0L6b1.jpg (some parts were left behind)
Happy Inpainting! :)
@@DarkStoorM_ damn man, you sound like you know a ton!! Would you mind if I contact you to talk for a bit? (Might not do it today cuz Im editing and have a tight time limit, but eventually)
@@Not4Talent_AI Sent you an email :)
@@DarkStoorM_ this is amazing! thanks for sharing. So if you do need to upscale from 1080p to 4k after inpainting all your details, how would you go about it?
A little fast, but a very straight to the point tutorial! Like very much! 🤩 some things I didn't know and asked for weeks how they work. Thanks, man!👍🏻
glad it helped!! thanks for the feedback too :3 (I'm working on pacing and stuff, still learning hahahah)
i just touch stable diffusion for 2 days because of deforum, it's really information overload for me lol
before i have used midjourney and usually playgroundai, buat whoaa this stable diffusion thingy is a whole new eye opener and perspective. gonna need to watch this video several times to learn this massive tools. thank you for the video!
btw i tried the "interrogate clip" but it seems takes forever to actually see the result (and i cancel it tho). is it always this long?
Yess, SD has so much stuff its overwhelming xD And sorry, the pacing of the video is too fast, trying to work on that!
Yeah, it usually takes a while and the prompt it gives is pretty meh xD I never actually use it, but its something to keep in mind for some specific cases. It can give some guidance on what the ai is "seeing" on the image.
Thank you btw!
Underrated channel, really nice approach explaining your process and tools
Thank you so much! hope it was helpful
From there on, stuff gets real
(P.D srry for the outpainting typo, didnt realize until now😂)
I can't wait for that man! you're 4 videos in and already the g.o.a.t. to me
@@S4SA93 hahaha thank you very much!!
Hmm already had everything recorded. I planned on making a whole video for stuff like that. Like how to use controlnet + photoshop to basically have full power over the final result.
Wanted it to be a new video as puttinh it in this one maybe is a little too much stuff.
If you want I can make a short on it so you dont have to wait until a full video tho!
@@Not4Talent_AI No stress man, do your videos as you see fit :) What you wanna do sounds great already. (I heard tho, shorts gain you a lot of traction on youtube :P)
@@S4SA93 hahaha we'll see then XD At least is a good way to give little information without having to make a whole video
Thanks! covered some things of in painting I didn't know about! Gonna help a lot!
Glat it did!!
Great explanation! I've been trying to learn how to create more elaborate illustrations in SD, and this helped me a lot.
Super happy to hear that! Thanks!
Wow! I've been using SD for a while now but a lot of the options I just 'had a feel' for and never quite understood what they did. Thank you for the video! It was very informative, I just subscribed, can't wait to learn more, keep up the good work! ^^
Glad you liked it!! I'll always try to explain deeply what things do if I can find it or, at least, experiment until I know what to use them for😂
Thanks for the kind words!
great quality content. fast and well organized. good energy too ! you are going to skyrocket in popularity like this !
Thanks!!! I'll try to improve as we go! Glad you liked it😄
Really energetic 👍
@@armanhasan-pl4oi ty! Hope that is a good thing 😂
I had never touched Inpaint Sketch, but that actually looks pretty cool, I was wondering if SD had something like NovelAI's painting feature so its nice that it exists here too.
Yep it is really fun to play arround with too hahha
Hiiii what is Inpaint sketch ?
@@gerryjunpyo2462 in the video I go a little bit over it. It is like usual inpaint but you can create a mask that can acta as a sketch for the result you are looking for
This answered a lot of my questions, one being "is there a way to draw in what I want for the AI to work off of?" Thanks!
Glad it helped!!
Thanks, you are awesome for sharing all tips and strategy for create great image with stable diffusion
Glad it was useful!! Thanks!
image to image is the only tab i use XD i actually just used the text 2 image tab to just check out how a model looked, but once i came upon a style i liked i would create my own base compositions, feed it to image 2 image and control net and then take out my shots, AI is good, but it can quickly become generic. Having some sort of art background or even an idea of what you want give the best outputs
Agree, img2img lets you overcome AI s shortcommings 😂
Super useful content! Gracias amigo!
Thank youu!!! De nada!
Thank you, very clear and interesting. Great examples too 🙂👍👍👍
Glad to hear that!!
One of the best videos I've found!! +++
Glad you liked it!!
Thank you for the video it's very helpful
thank you for watching!! glad it helped
Bro you just gained a subscriber. You are so entertaining
Thank you so much!! Glad you liked it!
so helpful, thanks for this video
Thank you!!
You deserve more subscribers! Could you do some Lora&checkpoint&negatives recommendations for different situations? 🎉 thank u for all your works
Currently I have a large list of videos that I want to do. But that sounds like a good in-between.
When I come back from vacations Ill start the testing to make the video.
Ty for the suggestion and kind comment!!!
@@Not4Talent_AI Love ur content!
This was really entertaining and informative. Thank you. New subber here
Thank you so much!! Glad you enjoyed!
Hey, great tutorials. I like the cropping tool you used in the inpainting section, but my version of A1111 doesn't seem to have it. I have the pre-installed canvas extension, and the popular canvas zoom. Am I missing something here?
nope, they just took it out cuz no one knew how to use it xD
(Im mad about it too ngl)
This videos are so damn good!!!! Good Job!!!!!!
Thanks!!! Super glad you like them!
Like I said. Pretty nice :D
And there were no pacing issues haha c:
I think I spotted my mistakes thanks to you :>
I just need to play around a bit to be sure
So thanks again for your lovely tutorial.
I can imagine the time you put into this
thanks!!! Happy to hear it helped :3 ty again for the super kind words :3
Inpaint addiction is really a problem for me 😂😂
Hshahhahaha same, too hard to stop
So good. Keep going
Will do!! thanks!
Okay this was too fast explained but i completed my knowledge with your video thank you !
hahaha sorry! Still tying to get the pacing right XD
Inpaint is good, but for some reason every iteration of inpaint that i do significantly reduces the image quality until im left with an extremely noisy and faded looking final result and no upscaling can fix it. It happens anytime I try to edit the image whether using inpaint, photopea, paint. Any idea what might be wrong?
Are you using whole image? And if so, what width/height?
Does it happen when using only masked as well?
If it happens regarthless of the optiom you choose then maybe it has to do with the model or a broken instalation.
If it only happens when using whole picture, try to match the w and h of you original image.
Now, if it happens using photopea too... thats kind of f up. Wouldnt really know. If you mean that the colors look muted. Then maybe it is becouse of a VAE issue. But dont think that should affect photopea in any way....
Maybe re-set the setting of SD, could be that it is saving the images without any real color data or something
Hope it works!
I img2img your img 2 make more img, however thank you,I now have a better idea of inpainting after watching this video, especially with photoshop masking and bluring
Glad it helped! Thank you!
Could you make a video about Photoshop tutorial for stable diffusion for basic things sir Please ❤
Could be an idea to have in queue, sure. Any particular thing you would like? (not sure if or when I'll make it. But maybe if it's easy I could do a short on it :3 )
@@Not4Talent_AI Yeah it supposed to be so short thanks for reply 🥰😇
@@Axherion thank you for the suggestion!!
🙏 Tnx
How much Vram does ur cpu have I have a 4gb vram cpu and want to create pictures like the one you showed as an example is it possible?
I think it is, but with a lot of patience. Since I didn't use controlnet or a super detailed upscaling. I have 11Gb vram though, can't tell you for sure if you can or not. I would recommend working on a smaller with and height, and then inpainting everything before upscaling.
9:33 I do this one and it's the most reliable for me personally. it take less time than rendering to get the best output, and combine it with Adetailer for hands and Ctrlnet inpaint, I got some good result with it.
and this video is good reminder for people who already know how it works and good resource for people who starting to use the tools so, ty and have a nice day sir. :)
and don't use inpaint sketch guys its so finicky I cant take it. lul
Hahahahhaha thanks!!! And inpaint sketch is actually top tier 😡 😂😂
Your videos deserve 1 million views each ! I know inpaint is popular to 'perfect' an image e.g. hands look horrifying, but I never really got how it worked just from looking at the buttons and trying to get it to work for myself, I think this is the first video i've seen that actually explains what to do ( but its still a bit hard for me to immediately understand what should go in the prompt fields for each specific use case and options selected ). I wonder if the way you use SD would work on a GTX 1080 or similar or if it would just be too slow.
I have a 1080 TI xD
And thank you so much btw!!
The way to prompt is pretty much just prompting for what you want inside the masked area. If its a hand then "hands" if its a face then "face".
The only issue with that is that maybe when using "only masked" it doesnt know in what position it should create stuff. That's where mask padding pixels come in. Putting those higher should help a lot
@@Not4Talent_AI Oh older card than I expected, I had 2070 super with waterblock in the past and also briefly tried a 3070 but its high power draw caused issues with my PSU and even caused an expensive RAM module to fail so I gave it to my brother for now and i'm still with GTX 1080. Even the RTX 3070 only had 8GB VRAM so less than your card.
@@ash0787 I think once you have more than 8gb you can do pretty much whatever at a decent pace. But obviously the more the better for stull like this xD
😢😢😢 TV TV fe1 23rd eg
@@skullseason1 are you ok ?
Tried to get 2k*2k via image2image, "not enough memory". Running 7900xtx 24GB on Linux. AMD is bit difficult to handle.
wow, hope thats not the norm honestly
Seem very interesting, but I just can't find any of these functions on Civitai.
I can't find the img2img and all the images generated are very small.
Are these functions only available if you become a supporter?
Hmmm I dont know about civit ai tbh. Im running this locally on a1111.
Maybe civitai only has txt2img.
Platforms with img to img can be rundiffusion, running a google colab and such.
No idea what the latest free online SD versions are tbh. There are some out there that have it tho!
@@Not4Talent_AI Ah, thanks! I saw the link to CivitAI on the description and assumed the video was about CivitAI.
I want to run it locally too, some time soon.
@@lamarckleland5731 oh, my bad then hahahha I always do everything locally for now
Great video, very helpful. Where can one find the denoising script at 1:46 ?
Thanks!! It comes already installed with Automatic1111. You have to go to scripts > x/y/z matrix, and then choose either x, y or z to act as the variables. (by default, x comes with "seed", you can just change that one. Clicking on it will open that list of scripts where you will find the "denoising" one.
@@Not4Talent_AI Awesome, thanks!
Hi. I'm having trouble on the aspect ratio of my paint on img2img
When I select any of the resize mode nothing happens.
I don't see any button either on the top corner of my image where I can crop it.
Please help🙏
hey! I think they took it out on the newer versions cuz noone used it XD
☹️Ohh
I wonder if you could use textual inversion on a single image to have an image that was even better described than the prompt that is interrogated
not exactly sure what you mean. training an embedding with a single image to then re-use it as a base?
@@Not4Talent_AI Yup, so basically it behaves more like a text to image while still being the image you want to modify. It looks like there is a script that behaves similarly to this already in img2img. I might be spouting crazy talk
@@ayrengreber5738 tbh I have no idea XD Experiment and see I guess hahaahha could lead to interesting results
@@Not4Talent_AI Shrugs, a casual idea... that I thought you might find makes an interesting video. I'm too lazy to experiment. :)
@@ayrengreber5738 If I have time for it I will thwn, ty!
1:47 How did you do the denoising script thing? You went wayyy too fast there.
Sorry! I explained it in another video so rushed it a little there.
You go to scritps- x/y/z - then you change one of the parameters, in this case the X parameter. To "denoising strenght" as shown in the video.
You can type all the denoising strengths you want to test separating them with commas. ALWAYS starting with the one you have at the moment.
So if I had denoising strength at 1 and wanted to test a range from 1 to 0.5, I would type this: 1, 0.9, 0.8, 0.7, 0.6, 0.5
Then just generate. It will generate with same seed and prompt. Once with denoising strength at 1, then with 0.9 then 0.8.... Until it finishes and gives you that matrix I show in the video.
Hope it helps!
@@Not4Talent_AI Wow, thank you!!! btw I followed everything else perfectly. Great vid =D
@@NickEnlowe ty!!!
Thank you for making such amazing videos. Can you plz make a video on training LOra?
Thanks!!! I want to touch the controlNet install first, but maybe then I could do that! Even though Olivio Sarikas just made a really good video on that!
@@Not4Talent_AI thank you for your reply. Currently, I am watching that video 😅
@@saqlainrazee4607 hahahhahaa hope it helps then!😂😂
ok, I like the way you explain things... new subscriber I am now...
Thank you so much!!
when i use inpaint, the brush lags so bad, i cant even move the cursor properly. I use mac m2, is this the reason?
I would love to help but I have no idea. Does it happen always or only with hi-res images?
@@Not4Talent_AI is it because the uploaded image had too much hires? let me try with lower res. the image was like 8 Mb in size
@@RoboMagician All I could find is that if you have a very big image in img2img it can cause some lag. Also keep in mind that the output will usually follow the input size you give it. Mainly if using "whole picture" or "just resize". So a 2048x2048 can turn into a 512 by 512. and reverse
Just generate it, and upgrade it with inpaint
Yeah, basically xD
I have to re watch it several times because the video is fast. I think you could change video's speed
Yeah, Im trying to adjust that on my newer vids, my bad😅
Where do i get the Denoising script ?
It's installed by default. If you mean the one I use to test which denosing strength is the best one for my case. then you need to use the "x,y,z plot" script.
Change one of the axis and you will hava a LARGE list of options. Yhere you can find the Denoising Strength one
@@Not4Talent_AI Is there a differance in the image depending on if I use Y over X or Z?
@@WatchingWithDinner nope
Hello. Is tiene a good and cheap automatic 1111 online service? My PC only has 6vram
I think this UI should work github.com/ddPn08/automatic1111-colab
(if it doesnt let me know! )
@@Not4Talent_AI Well i don't know how to use that, if you have a video on how to use it, it would be good. i already he it installed locally, but my pc is not good enogh to generate good images
@@mateodiaz5757 What do you mean with that? It cant generate good images cuz it crashes or...?
Btw, it should be just "clicking on open colab" and there click the "play" button on every step. After that it should work
@@Not4Talent_AI Thanks, i had never used google collab. What i mean is that i always get out of memory error when generating images.
@@Not4Talent_AI Excuse me for the question, is there a way to upload my own models and loras to the UI?
i feel like i've been using everything wrong hahahaha
hahahahahahah Happens, that's why I made the video tbh, to learn what everything really does xD.
have to say though, if it works you ain't doing it wrong xD
Why i have not the pencil on i2i section? :(
are you in the "inpaint" part?
@@Not4Talent_AI yes, i have the "undo", "clear", "remove image" and "use brush" tools
Sorry, i should have been clearer: i miss the tool to increase the size of the picture to generate something on the side of my image
ohhhhh I see. I think they changed that part in the later updates. Not entirely sure how it works atm. You can try to upscale using controlnet inpaint tho@@calvinmachnich2504
I have watched some other tutorial, even older than yours, and i see nobody have that feature. Are you sure it's not an extension?
A1111's inpaint sketch lags like FUUUUUUUUUUUUUUUUUUCCCCCCCCCC
Really? I havent had any problems with it in my exp. Dont know if it is an xformes, gpu or version thing
3:00 The resize and fill is so useless no matter what. you can just edit the image outside SD and then ported again to do the inpaint, in the beginning I was so happy that it had resize and fill for outpainting but nope it doesn't works that way and so stupid really while we need other extension to do the outpainting, so what I did was just resize the image outside of SD and ported again for polish my shitty rough painting from CSP. xd
Hahaha yeah, it isnt great. It does the job if you just need a quick adjustment though
@@Not4Talent_AI i never use it that way it's a bit scuffed, i think people mention that it's not behave that way before (ie cropping stuff without confirmation). before it was similar to Outpainting and you can use it as "mask" and then use it like inpaint mask.
what I did from time to time is something like this (ruclips.net/video/V1aaB7UgP7M/видео.html) or this (ruclips.net/video/ocDWBDnDKt0/видео.html) which is basically the same thing. but it works really well ngnl.
@@MrSongib UUUU those are pretty cool tutorials! Thanks!
Nice rhythm but too fast for detailed learning
Agreed, my bad! Trying to correct that little by little :3
Great job you did ! So much info .. Thnak you so much. Next time ... If you can peak just a bit ... .... ... slower. would help to follow your words and the film at the same time.
Hahahah thanks!!! And yess, I agree, srry. Need to practice a little bit more on the pacing of the videos. Hope to do it better next time😅😅 thanks for the feedback!!
@@Not4Talent_AI No problemo Bro. You give really great amount of info. Everyone can also adjust the speed of the video to ... ... ... 0,25 😅😇
@@gilles49280 haahahaha thanks !!! Glad it helped😄
What am I doing here with my potato PC that do 20 steps 512 x 512 for 1,5 minutes? xd
😂😂 that's hard ngl
@@Not4Talent_AI but still thank you for explanation for inpaint. Now I don't have to waste my time on glichy result.
Also, do you know what to do about wrong skin color? It is good while generating, but after it applied in full image it is... More grey than rest of image(I use "only masked" Mostly)
@@SuHISKgeroto It's likely that it is because of the "VAE" either not Appling correctly or not being used at all. Try going into, Settings > Stable diffusion> (and change the VAE to one you like. Forcing it to apply)
@@Not4Talent_AI thanks. I had VAE for Anything v3. I think I could just use it for all models, cause it looks nice ¯\_(ツ)_/¯
@@SuHISKgeroto i pretty much do that yeah 😂😂 only change it for very specific stuff
Like the content but the attention grabbing text popups are quite annoying
Which ones would you be referring to? (There are quite a lot, which I guess could be annoying xD) If you could give an example of which ones are the worse ones I'd appreciate that. There is some I think are pretty ok, like in the intro or the ones that I use to explain important parts. But maybe there are some specific times where I use the text popups and they aren't really helping, but rather distracting.
Ofc no obligation to respond, I'll figure it out with time and practice eventually, its just that it would be really helpful to know.
Thanks!!
10:30 too lazy to get AI to work? Simple, just hop on photoshop and start drawing. Not only you get a live preview, you also have control over EVERY SINGLE PIXEL of your image
agree
great, educational, but soo fast, nervous and jumpy to actually be useful... even if one knows exactly what is going on... imagine new to a1111, img2img......
yup, I'm trying to make the vids a bit slower now. I agree that this one is way too fast hahahaha
@@Not4Talent_AI apologies if i sounded rude or disrespectful, but it's a shame to waste all of that knowledge and effort just because of speed....
No worries! I agree, making tutorials is also a learning curve. which is why feedback like this is super useful to not make the same mistake later on. So thanks!! @@damird9635
You get a thumbs up just for saying 'photopea' properly. :)
hahahahahahah thanks! xD
Haha, great video, and you guys are great but wow I'm soo suck of every damn thing someone generates being an anime chick.
Hahahhahahha it is what it is 😂😂😂 Next video I'll generare mostly other stuff. Tried making various examples and with different models and characters
would be nice to redo this video, with slower step by step and explanations...too fast and all over the place. nice video though
Thought about this. Maybe I'll make a patreon to post vids like that As I want the chanel to mantain a consistency on the types of videos I post. But rn I dont have time for apatreon so maybe I could post them and keep them up until I make one
@@Not4Talent_AI just explain the steps a little more in details and show everything you do, and it will be great! Dont do patreon, its a waste of time and most ppl wont pay as there are plenty of explanations on yt. thanks for sharing your skills
@@abetuna2707 The next video will go a little more in depth on all this steps + controlnet included in the process. I'll try to go slower than in this video, as I am trying to make my vids a little more easy to follow.
The patreon stuff would be just extras as well as other things I still have to think about. Wouldn't paywall actual tutorials.
ty for your feedback! appreciate it!
@@Not4Talent_AI thank you!
"Art"
AI art, that's kind of the point 😂