21:22 by the way, the reason why you can use more steps in img2img is, that Steps are multiplied by Denoise, so you are not actually performing 100% steps, but for example, 50% on 0.5 denoise (this also applies to hires-fix), that's why it's faster than txt2img. If you put 100 steps with 0.2 denoise, your SD will only go as high as 20 steps or 21 (close approximation). It will work like that as long as you don't enable this setting: "With img2img, do exactly the amount of steps the slider specifies". Micro-tip: in photoshop, with brush tool and for standard brushes, you can lower the spacing to 1%: F5 -> Brush Tip Shape -> Spacing. Disabling it will lock your strokes to refresh rate and disables shift+click for line strokes. This only applies to the brushes you use for masking to smooth out the strokes - not suitable for painting as some brushes have fixed spacing for a reason. I also wanted to mention the Tile Resample model in CNet. Some experimenting is needed, but when inpainting for higher details, you are safe to use a very high denoise, like 0.7 or even 0.8. The Tile model will try to generate something very close to your source image, so you don't have to worry about fiddling with the denoise. You can just enable CNet with tile, pick control mode and you are done. If you are changing source images a lot, e.g. dragging the result over, just leave CNet blank and let tile preprocessor do the job (it will be a bit slower, but totally worth it). Also, thanks for the mention ❤ P.S. Raw filter got an update recently with some cool presets!
thank you for providing in depth explanations of stuff, helps a lot!! btw had no idea about the Cnet tile model, all the experiments I've done have always ended up in 45023 faces all over the image xDD will have to experiment more hahahaha tyty for everything!
@@Not4Talent_AI just a thought, FWIW... I haven't tried it yet myself, but saw some discussions where somebody suggested using full-sized models as opposed to pruned ones for upscaling. The person they were speaking to tried this and there is no doubt the results they posted afterwards were remarkably better!..
Do note that this is just the default behaviour of img2img, and in the settings you can disable this and make img2img do the exact amount of steps you requested.
Bro respect your dedication to this, not for me but i see the value in it. Im learning art the old fashioned way and dont understand the people butthurt over this even seeing your process
hahahah thanks!! I was learning art the old fashioned way, got a file with 7months of progress corrupted and said "fck it, amma learn AI" xD. Not that it will stop me from learning how to draw normally either. People now a days have to choose a side and blast the other one, which is sad. Appreciate your take on it :3
@@Not4Talent_AI huge respect to ya since you're not using it for malicious purposes. You're contributing to ai being a good tool for all artists so yea huge respect and also, keep your normal art journey, with knowledge of both ai and normal art you'll have basically infinite possibilities of artistic expression
People are generally upset because people are going into a prompt editor, writing "Draw me a character with big tits and ass and do it in style of x" and the thing can do that because x's images are scraped off the internet and x isn't getting paid for the amount of effort the AI snatched from them.. Or they're writing "Make character and don't specify who you're referencing" and when it does they say THEY did it, as if they didn't just press a "Make good (usually lifeless) art" button.
Well, most people using image gen AI just use online tools, generate 10 or 30 images and take the least crappy one for publishing on Twitter or wherever. And then I get why people think, AI art is not art. Classic drawing is a multi-step process of rapid sketching, settling on one design, multi-stage refinement and finishing touches. And "real" AI art is exactly that. It's a tool that does nothing exceptional on it's own. It's just the bristles on the brush. The controversy about AI training aside, this is just as amazig to watch as watching someone draw on a tablet through all the stages.
I've been a storyboard artist in animation for 26 years. I've been wondering if A.I. will eventually take over being used to make storyboards. It seems I really don't have to worry yet. BUT I shouldn't take for granted that A.I. is rapidly evolving.
Agree with that AI is really unpredictable in what it will take over. But I think storyboarding is such a "thinkable" thing that AI would have a very hard time doing. Not only it needs a super great understanding of the story you want to transmit. But also the FEELING every shot has to make you feel. Aside from other stuff like budget and what is and isnt possible. I just think AI could make it easier for people like storyboarders to maybe do an animatic or a rough moodboard faster. But I dont really think it will take over. At least not in a VERY long time
i think a intermediate AI user can probably do insane levels of storyboarding that surpass full on storyboard artists. But with that being said, a storyboard artist who learns ai will have a helpful tool that will only improve their work.
@@DimiArt I've done a bit of storyboarding myself and I think ai art has no chance in beating a trained storyboard artist. Theirs too much factors that play into the process of storyboarding you need full control over every aspect something I don't really think can ever really be achieved with ai art,
@trialshdultimategame I'd disagree, I do storyboards as well, they way I've been using ai is that I'd do a very rough storyboard and then use image to image to make those sketches almost fully detailed completed images in the style you want.
This is truly a well explained workflow i think its one of the first videos focusing on a stable diffusion work flow on TH-cam you are really underrated
@@reycaribe controlet, latent couple, photoshop, the extensions to create the open pose... etc? Like try to devide the video in parts so I can refere people to those videos if they missed sometjing? I think it could be a good idea, Ill see what I can do. Maybe Ill make a new image and post the proces in a slower and better explained pace
It would be nice to see the process with one image that would make it easier to follow. Heck you can make a video challenge. BTW inspiring stuff. Keep up the awesomeness. BTW, what happen to the umbrella ⛱️😂
I have a better idea you can learn to draw :) "that sounded a lot meaner than intended but it would be far less complicated to just draw it yourself or commotion someone to draw for you, this is all something people can already doo."
I already can draw. Not to a pro level, but I would probably be able to draw the anime piece myself. But, for example, the volleyball one, I couldt. As I cant draw realistic at all. And learning to draw like that is a process that takes years. Ive been learning how to draw for about 5 years (stopping a lot, so maybe on real study time would be about 2-3). I think it is really fun and I actually want to continue learning. I just think I need to learn from a different prespective. Learn how to draw what AI cant or what can help AI the most. I dont think AI will do stuff people cant do already. I just think it will make it faster. But yeah, I think the idea of learning how to draw is really good. And I agree. Just not as I was learning before. (Even though you could as well. Drawing is fun af) Ty!
A lot of great tips in here! And I figured out the latent couple thing, which turned out to be extremely useful. Thanks so much for taking the time to explain all these things. I know some were confused, but I don't think it's really that confusing, it's just a lot of information at once. AI tools are fantastic to use for an art flow. I've worked professionally as a creator for over 20 years, this is a godsend to efficiency and quality consistency. You're doing great things and I know artists rarely are appreciated for being avant garde, but it's what real artists do best!
Many artists hates AI arts, even though they can be use as a tool to generate the perfect reference for the beginner artists to practice their skills and learn the composition.
Thank you very much, it is a lot of information crammed into little time but it is made efficiently and enjoyable, sure I will have to watch the video several times but it is way better than those tutorials where people need 1 hour to describe just the first 5 minutes of your vid. I would love a Photoshop tutorial with your style. I am just an amateur that has used GIMP and that's another can of worms.
Thanks!!! I need to learn how to pace stuff better tbh xD I've been requested that some time. I'm starting to plan a full roadmap towards making comics. (so bringing this process to an actual hard level of use, where you need to fix everything and keep consistency in a lot of stuff). So I'll probably be using Photoshop a lot in that. Could consider making a video only on that if I see it gets complex. Ty for the idea btw!
hahahahaha could be. Depends on your drawing skill. I cant render realistic at all so I'd disagree in my case. I'm sure there are super skilled artists that can do this way better and faster for sure
Lately I’ve been photobashing a lot more in photoshop with the generative fill. Then use those rough AI bashes, back in SD and control net with uncanny and also reference_adain_attn. Like someone else mentioned the tile with ultimate SD upscale with a higher denoise is super powerful in changing the content or even adding new details on upscale.
Still completely new to this, but I've notice "BREAK" in prompts helps if one thing is bleeding into another. Like if you have a blue dress and it keeps making the eyes or hair blue because of it, adding BREAK will sometimes shake it lose it's fixation, but it adds a LOT to the prompt counter.
I've been learning how to draw for 6 years. (2-3 of actual practice). Had a photoshop file with 6months of work that I kept in order to see my progress /drawing every day/ and it got corrupted.. (my bad for not having a backup xd). I decided to learn AI and drawing, focusing mainly on the aspects of drawing that will help me use AI for the best
As an artist I feel impressed by these people's programming skills to teach a dumb computer about perspective in a way that a human would be capable of doing it. We artists in the art community are so affraid that "AI will take our places 😨😨" but damn look at the work this guy had to take to make one single image like I would have done it in a few hours because I know how to draw it by hand but a computer is like a baby it doesn't know anything until you teach it lmao We commonly forget that these are different areas (one targets hand-drawing and painting skills while the other targets AI programming skills for image generations) but both activities can be useful in some way. Artists used AI with Clip Studio's auto-coloring, didn't they? And you feeling frustated because an AI can draw "better than you" is like... wierd? Because you can improve your skills too just the way AI is improving in image generation every single hour of the day. Let's keep practicing our skills!
Could be yeah, there are people that have amazing skills. In an hour they could do what I did in 3. And way better The other part is that Im non of those people😂😂 so its faster/posible with AI
You packed more info and knowledge in the making of this one video than i did in 8 months on SD. However i'm quite confident on my knowledge about model and Lora training.
Good ideas in here, but you mixed everything in one big chaos :) If you did one by one tutorial instead of mixim all of them in separate parts, it would be much better. Imagine, that you want something from one particular tutorial, so you have to jump forward and backward looking which was which instead of simply watching without jumping. While it's not bad idea for ducementary, for tu torial it's worst nightmare :)
I agree, got a little lost in the making. Tried to taclke too much and didnt know how to do it. Mainly wanted to show the whole process, but ended up making a big mix of stuff. It was a good learning experience though, ty for your feedback and the comment! Ill try to make it better next time, still a lot to learn😅
@@Not4Talent_AIstill a good video on it's own. It show that the whole AI diffusion workflow is not as easy as people think it is. Just input text ? AH ! No thanks. I have 35 other slider and ratio to combine first !
@@lefourbe5596 hahhahha I think it is an "uploadable" video, but I do want to learn how to explain stuff like this in a more organized and simpler way. Thank you btw!!!
First of all, love the videos, have been watching your back catalogue and always in awe of talented and imaginative people, not to mention those who take the time and effort to spread their skills to the rest of us who would love to do something similar. As software can now create characters for people like me who can't draw to save their life, we can realise a dream to create something like a graphic novel (a small one to start with) as I'm full of ideas but no drawing talent! I'd love to know, to see a how to video on recreating, preserving an AI generated character across different situations, poses, etc as they will always change each time you ask for something new. Maybe using dedicated manga software in conjunction is the key, but to have the same backgrounds and faces to use again and again is something I haven't yet found in a video. Looking forward to more videos from you, cheers. ✌
Ty so much for the super kind comment! Im also looking for that stuff, so I experiment with pretty mich everything I have. If I find something that can be of help, Ill share it without doubts. For now I havent found anything super great, mainly for characters that are non-lora trained. Mantaining poses you could do with openpose 3D, which would alllow you to have the same pose from different prespectives. And for the same background everytime, you could do that with Blockade labs, gaugan360 or these tools. Even with A1111 extensions. That would also allow for the same background from different prespectives, even if not perfect at the moment. (You would most likely have to compose the 2 toguether, char and bg, but thats simple enough)
@@Not4Talent_AI artists copy concepts, not full blown designs and layer them on other drawings. If you want to make a point please say the full thought and not just make a short statement.
Sure, even though the process has now been simplified. But if you can model, rig, pose, light, texture, and render those scenes in less than 1/2 hours, then go ahead. It will 100% look better as it will fully depend on the skills of the producer. I think it might be easier to paint it too, rather than use 3D to create the scenes. But If you dont have those skills, then this is an option that opens the door for storytelling
Thanks for sharing this. As fast and hard to follow as it is, it gives a good overview of your process and to show the complexities of creating good AI art. There's a misconception out there that AI art generation is just a one click solution and there are WAY too many youtubers out there who lead their audience to believe that this is the case just so they can get more followers and likes. But too many people don't understand that it doesn't stop after hitting the Generate button, there is so much work that goes into refining and perfecting the image that takes hours and even days to complete. Muchas gracias, y espero a mirar mas videos de ti.
Thank you so much for the kind comment! Planning on making another video like this but WAAAAY easier to follow and trying to keep it simpler. I got too much stuff in one video for this one xD Espero que el contenido sea útil! jajaja
Awesome, thank you! This was really helpful to see your beginning to end workflow. One part I didn't quite understand though. At At 20:33 you said "mix all the results together, with the parts I like the most". What process do you mean to mix them together?
Thanks!! some times I just generate different images and then mix them in photoshop with the best parts of each. Like the background on one might be better than another, but have a worse foreground. That way I mix a good foreground and a nice background to get an overall better image
@@Not4Talent_AI Thanks for making this content, I like to use Stabble Diffusion, but I never thought I could come up with this innovation to create this kind of images. It is really very difficult for me, I will take advantage of the free time to try all of this.
@@Not4Talent_AI Por cierto 💡💡 viendo que eres español creo que tienes una puerta abierta a un montón de gente de habla hispanoamericana y que no entienden ni papa de inglés, pero igualmente quieren aprender SD. Entiendo que el habla inglesa llegue a más público, pero hay ya bastante competencia también. Igualmente pienso que tus vídeos tienen gran calidad y pueden competir ;) No obstante, como los vídeos los tienes ya en inglés, la parte más difícil la tienes hecha!... lo único que faltaría sería traducirlos. Te animo porque seguramente no te llevaría mucho y estarías en un nicho sin apenas competencia. Hay youtubers que se dan cuenta de esto muy tarde, como Veritasium por ejemplo. Sin rodeos: podrías abrir un canal en español y multiplicar tu crecimiento desde el principio con poco esfuerzo... No sé cómo lo ves
@@TheRmva2 Gracias! Creo que no es mala idea en absoluto, el problema es que a duras penas llego a hacer videos para este canal. Tengo un curro de 8 a 18. Podría intentar pillar los videos y doblarlos al español, pero no sería lo mismo, y regrabarlos es demasiado tiempo. Toy ahí ahí hahahhaha. Podría probar a hacer un "not4talent ESP". Pero dudo que la gente quiera ver un doblaje pocho. No te voy a mentir que, era una idea en mi cabeza pero has hecho que me la plantee más enserio hahahhaha tyty! Voy a mirarlo más a fondo jajaja
to get very specific and consistent things, ai art is not easier than regular art. basically its art for people who want to do it with mostly menus and prompts. i want to use ai for image2image post processing but i prefer drawing and 3d rendering for most of my work as that gives most intuitive control for me
Yeah, I would mostly agree, even though not fully. I think AI art is mostly wanting to skip some steps that you arent really ready to do yet. Be it for a lack of skills, or for a lack of motivation. In my case I rarely render stuff, I mostly enjoy sketching, so AI art is kind of a jump from start to endline directly. Ofc the results arent as good as if someone that actually knows how to render stuff did it manually, but it's way faster than learning all of that. (Even though I still try to learn it, cuz its fun af). I'd say AI art is sacrificing extra time on clean up and retouching to skip or postpone the years of learning a specific skill. Or at least speed up the process if you already have that skill. Really depends on what you do with it, cuz AI aint beating manual in terms of quality any time soon. Just in speed and accessibility
i think the area im not seeing ai being realistically implemented in a large extent is large projects. that is what i mean by specific and consistent. like in movies, series, games, comics. there you have so much work that goes into fitting all these vast concepts together(ranging from story elements to art). i am very pro ai to help individuals work more efficitently. im constantly dealing with traditional artists who want to ban it and i think theyre stupid for wanting to ban technology that could make them work more efficiently(and health issues are very common, i would have so much pain if i didnt use 3d). also im glad ai art is devaluing this "perfect" generic art style, im hoping there will be similar counter wave like when photography became mainstream, of more expressive and unique art styles @@Not4Talent_AI
so what im saying is i dont get the fear around ai cus i dont see how the human guidance will become obsolete any time soon. so i think artists should adapt to it@@Not4Talent_AI
Thank you for this, as an artist, I want to use things like midjourney more like a tool rather than a replacement for actual work. I'm also an indie game developer and I was wanting to make a visual novel sometime but I wanted to make a visual novel with more dynamic poses and character art instead of just reusing the same character pngs over and over.
happy to hear that! I'm planning to do a video to create characters and then make them consistently with AI. I've been experimenting on it today, with some really nice help. Hope to have it ready soon! Ty for the nice comment!
I am a programmer by profession. Even in programming, I find that getting the AI to do 80% and then doing the remaining 20% by hand is the most time efficient thing to do. Even here I see that you need a real artist in the driver's seat to make good images no matter how good the AI becomes.
Should matte painters physically draw in every texture they use in their work instead of photographing? What about vfx and 3d artists who, in mass, don’t use their own sculptures/meshes but instead the asset store to finish their work. Ya see where I’m getting at?
Goddamn... I see I'm still behind the curve for these more complex processes, holy shit. This feels pretty hyper advanced by my standards with SD, I will probably need to back up a few steps before watching this again.
Hahahah step by step, I'll probably get runned over by AI if I cant use it for a week 😂 this stuff goes too fast Hope it helps though! Ty for the comment🧡
Kind of late to the party here, but if you are applying the git command in command prompt you can right-click in the stable diffusion folder and choose to launch stable diffusion in that location
@@Not4Talent_AIIn windows, if you want to launch command prompt in a specific folder, if you open that folder in windows explorer you can right-click and there is an option to launch command prompt in that folder. It can make it easier to navigate there. I am not sure it works for git bash though
oh my god thanks so much for this!!! I draw professionally, but I've always wanted that beautiful painting A.I. does in MUCH less than half the time it takes me to do myself. A.I. also helps me with my realism, which I can draw but am god awful at coloring with
So happy to hear that!! There are some steps that I over complicate cuz of a lack of practice here. (You dont need so many controlney references most of the time. And inpaining can help more that how it does in the video) but I hope it helps :3
i allready tryed to use latent couple.. Now you told there are two versions? this was to fast for me in the film.. which one exactly is the one to install? maybe this was my problem.. Could you explain this a bit more?
yes! Srry There is the (two shot) latent couple, which only allows for square shapes that you specify by typing inside the extension. Then there is miZyind's Latent couple, the one that I installed, and that allows for both the mask method and the squares method. You will also need to activate "composable lora" though, which can be easily installed by searching it on the extensions tab, or with this github.com/opparco/stable-diffusion-webui-composable-lora.git
I like the idea of artists using AI like any other tool, just not the practice. This is more complex than just learning art fundamentals! I feel like I'm falling behind...
I think it is how it should be used tbh. And dont worry, Its simpler than I make it seem in the video (i overcomplicated stuff bc of a lack of practice). Also, as time goes on the tools themselves advance and become easier to use
ok, so here is my question as someone who is studying traditional way of art making. In Art school there many talented art students to the point that if you're a talent artist yourself, doesn't really matter cuz everyone is both fast and good roughly. Although it takes multiple months or even years to improve depending on your commitment to your art. when I compare the usage of these tools to a traditional artist who has all the knowledge and the talent to execute what they want or been asked to do within 15 to 30 minutes. At what point all of those A.I. tools that we don't actually have much control of, so we keep on adding additional tools to help us gain more control over the output, does the process become even longer? To a point where simply knowing the traditional way of art making and being excellent at it, is much faster especially at the beginning of the process?
yup. You are 100% right. Or at least for now. Idk how aI will evolve. But atm a good and experimented artists is 20 times better than an AI, 100 times better. Better quality, faster, understands what is wrong and how to change it, extreme control, etc. To me, the point of AI at the moment, is to make it help you in parts that you dont really enjoy about drawing or art. Maybe even if you enjoy them, it can be a fast way to iterate and test stuff. It can also help improve your own drawings (planning to make a video on that soon enough). Even though you'd still need the artistic eye to know what's missing and what you should correct. As well as a some level of skill to change it. There are more points to this but thats a basic rundown of how I see it for now. Hope it helps!
@@Not4Talent_AI Thank you for your response, I agree that AI can be another tool for artist to use, and maybe help in exploring ideas. The only thing I fear is when people lose the knowledge of why does something look good as it does? When you look at an amateur artist versus one who has better understanding of the art making, you can clearly see that the better one has good understanding of shapes, light, and values awhile the amateur artist haven't gain good understanding and knowledge of those yet, so their work looks flat. It can take quite amount of time and practice to finally understand those elements. With A.I. you're not really practicing art, you're just using a tool, as far as I can tell for now at least. I think the ones who'll be good with those tools are going to be like a good photographer who has understanding of shapes, light, and values with the additional understanding of the tool itself. But like you said we really don't have a clear idea of where all of this is going.
@@MustafaO.Artisan totally agree yeah. that's why I've never though artist would be obsolete. I do think its something that takes out a major part of beginner or amateur artists. But knowing about all the things you mention is super important and will continue to be
How do I set my own character as the person who is going to be on the control net. So basically I set poses with my designed then tell diffusion to put my character (that i have uploaded) . Thank you
Thx man for a very detailed dive in to the process, if you don't mind, how much time does it take to produce single image like this all together from start to finish involving all the nuances and hiccups that occur on the way?
Depends on the image and the result you want. The hardest here probably was the volleyball one, that took about 2-3 hours including planning and everything. Most of the time is just waiting for the AI to generate though. So I watch some series or exercice while I wait xD
Hi bro, thanks for the help. can you please put in the description the text for latent couple installing , you use at minute 4.46... I tried to write it from your screen but I can't see it completely... Please put this in the description so we can follow your steps.. Thanks again
very good, I knew someone would figure out all these things, it just takes a lot of time, I have experimented a bit with trying to turn regular drawings into AI pictures and also img2img stuff, i've also used controlnet a good few times by now but the quality of the results I get vary a lot.
@@Not4Talent_AI tried a few but one I thought might work but didn't was a video with a limited number of similar frames and mosaic censor pattern in the mouth area, I think that aspect really confused the program.
@@Not4Talent_AI the industry if you can call it that yet, moves so fast ! if you only do a couple hours per week its hard to keep up, your programs UI looks totally different to mine for example. I rarely tried to get a perfect image and did high effort, but rather just try to get a feel for the limits of the technology at any given time, and enjoy the variety of things it outputs, it can be an aid for the imagination. I thought perhaps I could easily make a 'really cool' picture by adding several Lora files including a character, pose / object and a background, but this didn't exactly pan out and like you say I had to do some tricks and extra steps to try and make it work, but the result was imperfect.
I absolutely cannot get latent couple to work. All forks of it are 6 months old and the mentioned git patch "does not apply", because the stuff it wants to alter is no longer in the place it used to be. (Some "two-shot" script thing) Is there some updated fork of latent couple that works with current A1111 V1.6.0 or later?
I'd delete the extension and install it from A111 directly again. (since some changes have been made and now it is easier to install, as it installs the same as every other extension) If it still doesnt work, I'd go for Regional Prompter which acts pretty much the same way
Huh, I haven't even heard about that, do you know where I can read more? Sounds really interesting. The contradictory thing about making videos is that I have less time to be up to date xD
Still not sure why some people are so sensitive whenever AI is used, even when the person is still drawing the image. Apparently, all AI usage is bad, even if you pick up a pencil and draw, because "AI bAd!1!".
Ty!! Depends on which one. The longes was the firstm it took about 3 hours but there was a lot of experimenting with different stuff. Its usually about 2 hours for the others. (Most of the time is spent waiting for the AI to generate, mainly on the upscale)
Hmmm not sure, havent tired. Good question though. I would do pretty much the same vut with a pixelart focused model and a pixelart lora. And then see how that does
Loading new extensions from the extensions tab doesnt work for me. Literally nothing comes up when I search for openpose. Ive always had to go to hugging face to manually download them and drop them into my folders. Without this tool I cant follow the video. Anyone know what Im missing
wait, really? Does the CMD give you any errors when clicking "load available" ? All extensions have the GIT link in the description bellow just in case btw!
@@Not4Talent_AI Got past that hurdle for some reason after leaving and coming back it was working fine.... story of my life. my current error is the Latent couple text for cmd: copy paste from the description is not workin. getting an error -error: can't open patch 'extensions/sd-webui-latent-couple/0001-Adding-after_ui_callback-for-scripts.patch': No such file or directory-
@@Not4Talent_AI yeah always good to step away for a while and come back. Right now Im going to see if I can find more info on the 3d openpose tool. Im a Maya guy so this is probably more my speed, and hopefully if used properly it can deal with the issue of the AI not making back facing characters that you ran into with the volleyball spiking character. Also should be better for keeping hands consistent
@@scetchmonkey007 Yeah hahhaa Sure! you can also use this if you want, probably easier. I have a video on it and other tools: app.posemy.art Really useful and can export OpenPose, canny, depth and normal maps. Also a maya guy here btw xD
what do I do once I used open pose editor? Hitting send to txttoimg or imgtoimg does nothing controlnet doesn't recognize it as a picture its just sees a black square.
@@sestep09 wtf, then Id re-install the extension or use an alternative. Cuz it should work if your downloaded image has the pose in it. If it does but cn is the one that turns it into black. Try chacking if you have an active preprocessor. If you do, change it to "none" and try again
Thank you changing the preprocessor to none did the trick. None of openposes preprocessors worked and sure enough none did thanks again.@@Not4Talent_AI
10:40 Best MMO EVER ^^ btw Photoshop beta lets you use SD to inpaint an image as a layer (so you keep the base image). This could reduce the number of steps in your workflow
@@Not4Talent_AI Since 2004 I spend like 25000 hours on Dofus (no I'm not an addict). Normally, if you have a paid Photoshop subscription, you should have access to the beta. I haven't had any particular problems using it.
@@losing_interest_in_everything I used to play that so much, come back to it every once in a wile. Even met my gf on dofus touch 😂😂 I should try it, ngl. Im just that type of person that wont update a program they are used to using just because there is a minor change in something. (Would probably get used to it in 3 days). 😂
I use AI to helping with the planning stage of the art, then I use that as inspiration to draw from, seems like everyone wants to 100% use AI start to finish which I think isnt the point of AI art.
nice!!! I dont think using AI for a 100% of the process is the best way either, but it is a nice way to create and compensate for your shortcomings as an artist, while you learn how to overcome them
this is next level!! newb here but would you consider making a tutorial about making a scene? one place. multiple. same clothes.color. with different angles. consistency is what i struggle with. faces with loras is not bad but a scene. 3d background (sketchfab) for the photo angles and control net then impaint/seg like you went over. would love an expert to walkthrough that specifically.
thanks!!! Now I'm doing an expressions video that will come out as soon as I'm feeling well and can record voice over. Next I'll try to make a tutorial on lora sliders, but it shouldnt take too long. And after that I had planned a video on how to get spacial consistency and all that stuff. Tbh not fully sure how to do it for now. But I'll be experimenting a lot to bring ya'll the best methods I can
@@Not4Talent_AI it can overbake the image afaik if you lock the seed and all that, you can try it out just in paint one area than use that as base image then repeat.
The problem for me is that, composition stuff and can be annoying af, since it's random. ill try some of your method sometimes. i did try to get my sketch working long time ago and I end up giving up at the time I just messing around with inpaint and SD base model so kinda hard to control stuff. xd
It is yeah, thats why I create a bunch of different controlnet models, to see if I can controll that better. Another option is photobashing and latent coulple with contolnet. At that point you are using AI to mix stuff for you and that is it, but it may work😂
@@Not4Talent_AI latent couple like gambling aswell but can be good, last time I tried it's not working anymore on my 5700xt. maybe I should try your command at some point, maybe it can fix my problem.
@Not4Talent hmm wouldn't there be a way to utilize unreal engine to create the scene? you could use cheap assets or free assets to create the beach and use the basic character models for the positioning of the characters?
Yeah, ofc. Or just painting directly xD But that wasnt really the point. To be honest creating that scene in unreal would be very hard due to the fact that you need rigged 3D models for all characters, then the time of modeling and stuff for the beach. The result would be 1000 times better and in video format, but the time invested in the models would be veeeeeery high ino
@@Not4Talent_AI hm probably so. i was thinking of just basic unreal models that come with the software for proper positioning of the characters so you could frame a scene. i think in the future that might be the more efficient way to do it but we aren't there yet. fully customizable themes with trained models i think would be the most top notch. will be very interesting to see how this technology improves in the near future
Ohhhh I see, I think I misunderstood the initial comment. But yeah, I hope to see this tech evolve to make stuff like that easier and faster @@Atticus_Moore
Yeah, I didn't mean it to follow along but it is a common complaint I had. Next one I'll try to keep it one at a time, or at least easier to make ones. Sorry
Dude, your tutorials are good, but from a certain point you really go off on a tangent. From the middle of the video things become confusing and difficult to follow.
Do you feel this happens in every tut. Or just this one? I can see how in this one I go off. But not sure about the others. I tend to go on tangents irl too, so ut wouldnt surprise me if people feel like thats the case. I need to understand it to fix it tho. Thanks for the feedback!!
@@Not4Talent_AI Don't get me in the wrong way: your tutorials are gold, but from the middle of the video you give too much things for granted as if this was an advanced guide for expert people. Maybe, this is stuff too advanced for me. I will check for tutorials about Controlnet before rewatching this.
Oh, that could be. I tend to make tutorials for people that have some experience with the program already. As explaining everything everytime would be a pain. Maybe I could create a series of shorts or more noob friendly videos so I can refere people to those instead of explaining it in the video? Ill think about it. Thank you so much!
Yes! You absolutely could. I talk about that in another video a little more slightly. I'll make a new video with more workflows eventually. Though it is true that for some you could just photobash, for the volleyball image, for example, it is easier to make the controlent references and then phot bash from there. In my opinion ofc. There is soooo much testing to be done still. I think I need to optimize more what can be done with controlnet, what with photobashing, what with inpainting... etc
ty! its on the the suggested content. (could put the actual video on cuz I hadnt recorded it at the time xD mb) th-cam.com/video/SvGIkWFh-eo/w-d-xo.html
Not really. you would get better results for sure. But learning how to draw takes way more time. I'd like for this to act as a gate way for people to learn how to draw and improve upon AI's creations
hahahhahahaha yeah, if you already know how to paint you'd get better results and faster just drawing xD Seeing how someone like samdoesarts can paint masterpieces in 1hour, this is pretty lame I'm still not anywhere near that drawing level tho xD
well explained and very entertaining video, i was just wondring if you can share with me the resault images you got in higher quality? the instagram ones are low quality (512x282), In any case, thanks for your help ^^
Really helpful and informative videos, thanks for making them. keep up the good work pls :D Maybe you can make a video explaining what AfterDetailer is? thanks again.
Agree! I have a bit of a new process, but I'll be posting 2 vids soon enough about it. 1 will be on this style, faster in nature. But then I'll also post a, howeverlong video of an image start to finish, while explaining a bit of my though process while making it. It might have some speed-ups on stuff like generating images, or iterating. But the rest will be slow and chill
21:22 by the way, the reason why you can use more steps in img2img is, that Steps are multiplied by Denoise, so you are not actually performing 100% steps, but for example, 50% on 0.5 denoise (this also applies to hires-fix), that's why it's faster than txt2img.
If you put 100 steps with 0.2 denoise, your SD will only go as high as 20 steps or 21 (close approximation). It will work like that as long as you don't enable this setting: "With img2img, do exactly the amount of steps the slider specifies".
Micro-tip: in photoshop, with brush tool and for standard brushes, you can lower the spacing to 1%: F5 -> Brush Tip Shape -> Spacing. Disabling it will lock your strokes to refresh rate and disables shift+click for line strokes. This only applies to the brushes you use for masking to smooth out the strokes - not suitable for painting as some brushes have fixed spacing for a reason.
I also wanted to mention the Tile Resample model in CNet. Some experimenting is needed, but when inpainting for higher details, you are safe to use a very high denoise, like 0.7 or even 0.8. The Tile model will try to generate something very close to your source image, so you don't have to worry about fiddling with the denoise. You can just enable CNet with tile, pick control mode and you are done. If you are changing source images a lot, e.g. dragging the result over, just leave CNet blank and let tile preprocessor do the job (it will be a bit slower, but totally worth it).
Also, thanks for the mention ❤
P.S. Raw filter got an update recently with some cool presets!
thank you for providing in depth explanations of stuff, helps a lot!!
btw had no idea about the Cnet tile model, all the experiments I've done have always ended up in 45023 faces all over the image xDD will have to experiment more hahahaha
tyty for everything!
@@Not4Talent_AI just a thought, FWIW... I haven't tried it yet myself, but saw some discussions where somebody suggested using full-sized models as opposed to pruned ones for upscaling. The person they were speaking to tried this and there is no doubt the results they posted afterwards were remarkably better!..
@@ArielTavori Oh, thats really interesting. Thank you! Ill have to try
I love the comment section X)
Do note that this is just the default behaviour of img2img, and in the settings you can disable this and make img2img do the exact amount of steps you requested.
this is how I want to see ai art tools used. in a way that feels more like you're not just taking the worst first thing and calling it a day.
Totaly!
Bro respect your dedication to this, not for me but i see the value in it. Im learning art the old fashioned way and dont understand the people butthurt over this even seeing your process
hahahah thanks!! I was learning art the old fashioned way, got a file with 7months of progress corrupted and said "fck it, amma learn AI" xD. Not that it will stop me from learning how to draw normally either.
People now a days have to choose a side and blast the other one, which is sad. Appreciate your take on it :3
@@Not4Talent_AI huge respect to ya since you're not using it for malicious purposes. You're contributing to ai being a good tool for all artists so yea huge respect and also, keep your normal art journey, with knowledge of both ai and normal art you'll have basically infinite possibilities of artistic expression
@@hypergu Thats the goal hahha
Thanks!! and good luck with your artistic journey, tough but really fun!
People are generally upset because people are going into a prompt editor, writing "Draw me a character with big tits and ass and do it in style of x" and the thing can do that because x's images are scraped off the internet and x isn't getting paid for the amount of effort the AI snatched from them..
Or they're writing "Make character and don't specify who you're referencing" and when it does they say THEY did it, as if they didn't just press a "Make good (usually lifeless) art" button.
Well, most people using image gen AI just use online tools, generate 10 or 30 images and take the least crappy one
for publishing on Twitter or wherever. And then I get why people think, AI art is not art.
Classic drawing is a multi-step process of rapid sketching, settling on one design, multi-stage refinement and finishing touches.
And "real" AI art is exactly that. It's a tool that does nothing exceptional on it's own. It's just the bristles on the brush.
The controversy about AI training aside, this is just as amazig to watch as watching someone draw on a tablet through all the stages.
I've been a storyboard artist in animation for 26 years. I've been wondering if A.I. will eventually take over being used to make storyboards. It seems I really don't have to worry yet. BUT I shouldn't take for granted that A.I. is rapidly evolving.
Agree with that AI is really unpredictable in what it will take over. But I think storyboarding is such a "thinkable" thing that AI would have a very hard time doing.
Not only it needs a super great understanding of the story you want to transmit. But also the FEELING every shot has to make you feel. Aside from other stuff like budget and what is and isnt possible.
I just think AI could make it easier for people like storyboarders to maybe do an animatic or a rough moodboard faster. But I dont really think it will take over. At least not in a VERY long time
i think a intermediate AI user can probably do insane levels of storyboarding that surpass full on storyboard artists. But with that being said, a storyboard artist who learns ai will have a helpful tool that will only improve their work.
@@DimiArt I've done a bit of storyboarding myself and I think ai art has no chance in beating a trained storyboard artist. Theirs too much factors that play into the process of storyboarding you need full control over every aspect something I don't really think can ever really be achieved with ai art,
@trialshdultimategame I'd disagree, I do storyboards as well, they way I've been using ai is that I'd do a very rough storyboard and then use image to image to make those sketches almost fully detailed completed images in the style you want.
This is truly a well explained workflow i think its one of the first videos focusing on a stable diffusion work flow on TH-cam you are really underrated
thank you so much! Happy to hear it was useful
@@Not4Talent_AI It's too overwhelming. Can you put it in a playlist of all the elements we should know for this process please?
@@reycaribe controlet, latent couple, photoshop, the extensions to create the open pose... etc? Like try to devide the video in parts so I can refere people to those videos if they missed sometjing?
I think it could be a good idea, Ill see what I can do. Maybe Ill make a new image and post the proces in a slower and better explained pace
It would be nice to see the process with one image that would make it easier to follow. Heck you can make a video challenge. BTW inspiring stuff. Keep up the awesomeness.
BTW, what happen to the umbrella ⛱️😂
@@Eloyo agree xDDd And yeah I forgot the umbrella XDDDDDDDDD
Thanks btw!!
I have a better idea you can learn to draw :) "that sounded a lot meaner than intended but it would be far less complicated to just draw it yourself or commotion someone to draw for you, this is all something people can already doo."
I already can draw. Not to a pro level, but I would probably be able to draw the anime piece myself.
But, for example, the volleyball one, I couldt. As I cant draw realistic at all. And learning to draw like that is a process that takes years.
Ive been learning how to draw for about 5 years (stopping a lot, so maybe on real study time would be about 2-3). I think it is really fun and I actually want to continue learning. I just think I need to learn from a different prespective. Learn how to draw what AI cant or what can help AI the most.
I dont think AI will do stuff people cant do already. I just think it will make it faster.
But yeah, I think the idea of learning how to draw is really good. And I agree. Just not as I was learning before. (Even though you could as well. Drawing is fun af)
Ty!
A lot of great tips in here! And I figured out the latent couple thing, which turned out to be extremely useful. Thanks so much for taking the time to explain all these things. I know some were confused, but I don't think it's really that confusing, it's just a lot of information at once.
AI tools are fantastic to use for an art flow. I've worked professionally as a creator for over 20 years, this is a godsend to efficiency and quality consistency. You're doing great things and I know artists rarely are appreciated for being avant garde, but it's what real artists do best!
Happy to hear that, thank you so much for the kind comment!!!
Many artists hates AI arts, even though they can be use as a tool to generate the perfect reference for the beginner artists to practice their skills and learn the composition.
Yep, and many other stuff like colors and just inspiration
Thank you very much, it is a lot of information crammed into little time but it is made efficiently and enjoyable, sure I will have to watch the video several times but it is way better than those tutorials where people need 1 hour to describe just the first 5 minutes of your vid.
I would love a Photoshop tutorial with your style. I am just an amateur that has used GIMP and that's another can of worms.
Thanks!!! I need to learn how to pace stuff better tbh xD
I've been requested that some time. I'm starting to plan a full roadmap towards making comics. (so bringing this process to an actual hard level of use, where you need to fix everything and keep consistency in a lot of stuff). So I'll probably be using Photoshop a lot in that. Could consider making a video only on that if I see it gets complex.
Ty for the idea btw!
This is way more work than just drawing
hahahahaha could be. Depends on your drawing skill. I cant render realistic at all so I'd disagree in my case. I'm sure there are super skilled artists that can do this way better and faster for sure
Different people have different interests and skill sets. Hard to believe, I know.
I'm learning a lot from your videos, thank you so much for sharing.
happy to hear!! Tysm for watching and the positive comment!
gotta say it's a lot more work than just straight up drawing
hahahaha agree, but only if you know how to draw already
8:46 Then, in the middle of the night, you are watching, and you can't sleep anymore, lol.
😂😂😂😂😂😂
amazing workflow is incredible how fast prompting became obsolete when you start making more complex things!
Yeah, knowing prompting can be really useful, but wont ever give you the control this stuff gives 😂
Lately I’ve been photobashing a lot more in photoshop with the generative fill. Then use those rough AI bashes, back in SD and control net with uncanny and also reference_adain_attn. Like someone else mentioned the tile with ultimate SD upscale with a higher denoise is super powerful in changing the content or even adding new details on upscale.
Cool! I'll see if I can try it then. Thank you¡
Still completely new to this, but I've notice "BREAK" in prompts helps if one thing is bleeding into another. Like if you have a blue dress and it keeps making the eyes or hair blue because of it, adding BREAK will sometimes shake it lose it's fixation, but it adds a LOT to the prompt counter.
Yess!! It acts like a prompt "restart". really nice for colors and cohesiveness
As an artist who had been learning art for a decade now, seeing people without any background in art make images better than me is quite depressing.
I've been learning how to draw for 6 years. (2-3 of actual practice). Had a photoshop file with 6months of work that I kept in order to see my progress /drawing every day/ and it got corrupted.. (my bad for not having a backup xd). I decided to learn AI and drawing, focusing mainly on the aspects of drawing that will help me use AI for the best
As an artist I feel impressed by these people's programming skills to teach a dumb computer about perspective in a way that a human would be capable of doing it. We artists in the art community are so affraid that "AI will take our places 😨😨" but damn look at the work this guy had to take to make one single image like I would have done it in a few hours because I know how to draw it by hand but a computer is like a baby it doesn't know anything until you teach it lmao
We commonly forget that these are different areas (one targets hand-drawing and painting skills while the other targets AI programming skills for image generations) but both activities can be useful in some way. Artists used AI with Clip Studio's auto-coloring, didn't they?
And you feeling frustated because an AI can draw "better than you" is like... wierd? Because you can improve your skills too just the way AI is improving in image generation every single hour of the day. Let's keep practicing our skills!
@@sleepylady2723 i mean,programming is easy.
@@Memethief621so does drawing backgrounds
I think professional artist could paint it faster, without AI.
Could be yeah, there are people that have amazing skills. In an hour they could do what I did in 3. And way better
The other part is that Im non of those people😂😂 so its faster/posible with AI
The funny thing is that if professional artist decides to use AI they will have better results, no offense intended
Fantastic workflow!
Thanks!!!
Bro you're a beast
hahahaha ty so much!!
You packed more info and knowledge in the making of this one video than i did in 8 months on SD.
However i'm quite confident on my knowledge about model and Lora training.
That's precisely what I'm 0% confident on xD Even if I've been playing with it for a week ahahaahahaha
Still dont know so much about it
Thanks so much for the video!
Ty for watching!
Excellent tutorial. Will take me so long time to only digest the load of information 😛. Bravo & ... thank you 👍👌
Hahahhaha thank you so much! I'll try to improve the tutorials so they are a little more digestable too😂 dont have enough practice yet ahhaha
Good ideas in here, but you mixed everything in one big chaos :) If you did one by one tutorial instead of mixim all of them in separate parts, it would be much better. Imagine, that you want something from one particular tutorial, so you have to jump forward and backward looking which was which instead of simply watching without jumping. While it's not bad idea for ducementary, for tu torial it's worst nightmare :)
I agree, got a little lost in the making. Tried to taclke too much and didnt know how to do it.
Mainly wanted to show the whole process, but ended up making a big mix of stuff.
It was a good learning experience though, ty for your feedback and the comment!
Ill try to make it better next time, still a lot to learn😅
@@Not4Talent_AIstill a good video on it's own.
It show that the whole AI diffusion workflow is not as easy as people think it is.
Just input text ? AH ! No thanks.
I have 35 other slider and ratio to combine first !
@@lefourbe5596 hahhahha I think it is an "uploadable" video, but I do want to learn how to explain stuff like this in a more organized and simpler way.
Thank you btw!!!
First of all, love the videos, have been watching your back catalogue and always in awe of talented and imaginative people, not to mention those who take the time and effort to spread their skills to the rest of us who would love to do something similar.
As software can now create characters for people like me who can't draw to save their life, we can realise a dream to create something like a graphic novel (a small one to start with) as I'm full of ideas but no drawing talent!
I'd love to know, to see a how to video on recreating, preserving an AI generated character across different situations, poses, etc as they will always change each time you ask for something new. Maybe using dedicated manga software in conjunction is the key, but to have the same backgrounds and faces to use again and again is something I haven't yet found in a video.
Looking forward to more videos from you, cheers. ✌
Ty so much for the super kind comment!
Im also looking for that stuff, so I experiment with pretty mich everything I have. If I find something that can be of help, Ill share it without doubts.
For now I havent found anything super great, mainly for characters that are non-lora trained.
Mantaining poses you could do with openpose 3D, which would alllow you to have the same pose from different prespectives.
And for the same background everytime, you could do that with Blockade labs, gaugan360 or these tools. Even with A1111 extensions.
That would also allow for the same background from different prespectives, even if not perfect at the moment. (You would most likely have to compose the 2 toguether, char and bg, but thats simple enough)
If don't have talent, don't steal from other
@@ggadams639 hahahaha bro hasn't seen an artist in his entire life
@@ggadams639 What an ignorant comment.
@@Not4Talent_AI artists copy concepts, not full blown designs and layer them on other drawings. If you want to make a point please say the full thought and not just make a short statement.
w so many steps at some point it seems like it would be faster and simpler to model, rig and pose
Sure, even though the process has now been simplified. But if you can model, rig, pose, light, texture, and render those scenes in less than 1/2 hours, then go ahead. It will 100% look better as it will fully depend on the skills of the producer.
I think it might be easier to paint it too, rather than use 3D to create the scenes.
But If you dont have those skills, then this is an option that opens the door for storytelling
💎Amazing, TOP!
ty!!
This video is Amazing ! Very good ideas ! Do you have a way to use controlnet segmentation with SDXL ?
I dont know if SDXL has support for segmentation yet.
And thank you so much! I'm about to do some other ones more updated on this too soon enough. :3
I am starting my youtube channel (@papai.raconte) and you are, with a couple a of other youtuber a big help to this achiement.
Many Thanks !
@@roxmay nice!! wish you the best of lucks!
He just hit us with rickroll and low budget moistcritical less than minute 💀💀
😂😂😂😂😂😂
I Love How Video And Tittle Made For Each Other
Thanks for sharing this. As fast and hard to follow as it is, it gives a good overview of your process and to show the complexities of creating good AI art. There's a misconception out there that AI art generation is just a one click solution and there are WAY too many youtubers out there who lead their audience to believe that this is the case just so they can get more followers and likes. But too many people don't understand that it doesn't stop after hitting the Generate button, there is so much work that goes into refining and perfecting the image that takes hours and even days to complete. Muchas gracias, y espero a mirar mas videos de ti.
Thank you so much for the kind comment! Planning on making another video like this but WAAAAY easier to follow and trying to keep it simpler. I got too much stuff in one video for this one xD
Espero que el contenido sea útil! jajaja
still art theft, and less work then it actually takes to make real art
Awesome, thank you! This was really helpful to see your beginning to end workflow. One part I didn't quite understand though. At At 20:33 you said "mix all the results together, with the parts I like the most". What process do you mean to mix them together?
Thanks!! some times I just generate different images and then mix them in photoshop with the best parts of each. Like the background on one might be better than another, but have a worse foreground. That way I mix a good foreground and a nice background to get an overall better image
Konosuba fan !? instant sub
hahahhahahhah tyty!
8:47 Nightmare fuel holy hahah.
Amazing video yet again, this helps so so much, can't thank you enough!
😂😂
I dont at the moment, no time😅
But thank you so much! Comments like this are more than enough🧡
Goated channel
ty so much !!!
This is really amazing !
Ty!
@@Not4Talent_AI Thanks for making this content, I like to use Stabble Diffusion, but I never thought I could come up with this innovation to create this kind of images. It is really very difficult for me, I will take advantage of the free time to try all of this.
@@TataAIOfficial thank you for the kind comment! I hope you can create what you wanted :3
You're awesome! Advanced tutorilas just to the point. Keep going dude, you're gonna get far ;) Thank you btw
thank you so much for the suppport!!
@@Not4Talent_AI Por cierto 💡💡 viendo que eres español creo que tienes una puerta abierta a un montón de gente de habla hispanoamericana y que no entienden ni papa de inglés, pero igualmente quieren aprender SD. Entiendo que el habla inglesa llegue a más público, pero hay ya bastante competencia también. Igualmente pienso que tus vídeos tienen gran calidad y pueden competir ;) No obstante, como los vídeos los tienes ya en inglés, la parte más difícil la tienes hecha!... lo único que faltaría sería traducirlos. Te animo porque seguramente no te llevaría mucho y estarías en un nicho sin apenas competencia. Hay youtubers que se dan cuenta de esto muy tarde, como Veritasium por ejemplo. Sin rodeos: podrías abrir un canal en español y multiplicar tu crecimiento desde el principio con poco esfuerzo... No sé cómo lo ves
@@TheRmva2 Gracias! Creo que no es mala idea en absoluto, el problema es que a duras penas llego a hacer videos para este canal. Tengo un curro de 8 a 18. Podría intentar pillar los videos y doblarlos al español, pero no sería lo mismo, y regrabarlos es demasiado tiempo.
Toy ahí ahí hahahhaha. Podría probar a hacer un "not4talent ESP". Pero dudo que la gente quiera ver un doblaje pocho.
No te voy a mentir que, era una idea en mi cabeza pero has hecho que me la plantee más enserio hahahhaha tyty! Voy a mirarlo más a fondo jajaja
Although I’m already adept at this, thank you for doing this even though it’s late as per ai timeline (which is so fast you can’t keep up anyways).
Yeah, AI goes so fng fast😂😂 tyty!!
If this is what shadman did he would have gotten way less backlash
I have no idea what you are referring to ngl xD
@@Not4Talent_AI sorry i meant shadversity
to get very specific and consistent things, ai art is not easier than regular art. basically its art for people who want to do it with mostly menus and prompts. i want to use ai for image2image post processing but i prefer drawing and 3d rendering for most of my work as that gives most intuitive control for me
Yeah, I would mostly agree, even though not fully. I think AI art is mostly wanting to skip some steps that you arent really ready to do yet. Be it for a lack of skills, or for a lack of motivation. In my case I rarely render stuff, I mostly enjoy sketching, so AI art is kind of a jump from start to endline directly. Ofc the results arent as good as if someone that actually knows how to render stuff did it manually, but it's way faster than learning all of that. (Even though I still try to learn it, cuz its fun af).
I'd say AI art is sacrificing extra time on clean up and retouching to skip or postpone the years of learning a specific skill. Or at least speed up the process if you already have that skill. Really depends on what you do with it, cuz AI aint beating manual in terms of quality any time soon. Just in speed and accessibility
i think the area im not seeing ai being realistically implemented in a large extent is large projects. that is what i mean by specific and consistent. like in movies, series, games, comics. there you have so much work that goes into fitting all these vast concepts together(ranging from story elements to art). i am very pro ai to help individuals work more efficitently. im constantly dealing with traditional artists who want to ban it and i think theyre stupid for wanting to ban technology that could make them work more efficiently(and health issues are very common, i would have so much pain if i didnt use 3d). also im glad ai art is devaluing this "perfect" generic art style, im hoping there will be similar counter wave like when photography became mainstream, of more expressive and unique art styles @@Not4Talent_AI
so what im saying is i dont get the fear around ai cus i dont see how the human guidance will become obsolete any time soon. so i think artists should adapt to it@@Not4Talent_AI
Thank you for this, as an artist, I want to use things like midjourney more like a tool rather than a replacement for actual work. I'm also an indie game developer and I was wanting to make a visual novel sometime but I wanted to make a visual novel with more dynamic poses and character art instead of just reusing the same character pngs over and over.
happy to hear that! I'm planning to do a video to create characters and then make them consistently with AI. I've been experimenting on it today, with some really nice help. Hope to have it ready soon!
Ty for the nice comment!
@@Not4Talent_AI Are you ever gonna make a tutorial for making our own ai models for our own artstyle etc...?
@@malsolo293 good idea, I'll get it on my list!
I am a programmer by profession. Even in programming, I find that getting the AI to do 80% and then doing the remaining 20% by hand is the most time efficient thing to do.
Even here I see that you need a real artist in the driver's seat to make good images no matter how good the AI becomes.
Yup 100% agree
After this video no one will think, ai art is easy, ever again!
😂😂😂 I overcomplicated some stuff too ngl XD Right now the workflow is a little more optimized ahhahaha
A.I. image generation isn't an art
@@abanana2859 for you
just grab a pencil
I did xD
@uh3610 I never said not to do so. I encourage people to learn how to draw. But I like lreaning AI too. Its not mutually exclusive
@uh3610 I get it, dont worry hahahah
Pretty bad way of looking at it
Should matte painters physically draw in every texture they use in their work instead of photographing? What about vfx and 3d artists who, in mass, don’t use their own sculptures/meshes but instead the asset store to finish their work. Ya see where I’m getting at?
Love that you always go in depth
Goddamn... I see I'm still behind the curve for these more complex processes, holy shit. This feels pretty hyper advanced by my standards with SD, I will probably need to back up a few steps before watching this again.
Hahahah step by step, I'll probably get runned over by AI if I cant use it for a week 😂 this stuff goes too fast
Hope it helps though! Ty for the comment🧡
Very creative and well exposed workflow!
thank you!!
Awesome video, made it easy to follow the whole process 👌
Thank you! Happy to hear that :3
Kind of late to the party here, but if you are applying the git command in command prompt you can right-click in the stable diffusion folder and choose to launch stable diffusion in that location
I did not really understand what you mean. The git bash here thing?
Or smt else?
@@Not4Talent_AIIn windows, if you want to launch command prompt in a specific folder, if you open that folder in windows explorer you can right-click and there is an option to launch command prompt in that folder. It can make it easier to navigate there.
I am not sure it works for git bash though
oh, had no idea. I used to just type cmd on the path tbh. but thats nice to know. tyty!! @@stephenclement3349
oh my god thanks so much for this!!! I draw professionally, but I've always wanted that beautiful painting A.I. does in MUCH less than half the time it takes me to do myself. A.I. also helps me with my realism, which I can draw but am god awful at coloring with
So happy to hear that!! There are some steps that I over complicate cuz of a lack of practice here. (You dont need so many controlney references most of the time. And inpaining can help more that how it does in the video) but I hope it helps :3
i allready tryed to use latent couple.. Now you told there are two versions? this was to fast for me in the film.. which one exactly is the one to install? maybe this was my problem..
Could you explain this a bit more?
yes! Srry
There is the (two shot) latent couple, which only allows for square shapes that you specify by typing inside the extension. Then there is miZyind's Latent couple, the one that I installed, and that allows for both the mask method and the squares method.
You will also need to activate "composable lora" though, which can be easily installed by searching it on the extensions tab, or with this github.com/opparco/stable-diffusion-webui-composable-lora.git
@@Not4Talent_AI thank you 😃
@@tomschuelke7955 no problem! hope it helps :3
Really enjoyed this thorough video.
Thank you!!!
You did a amazing job on this, so much knowledge and tools used to created these images, The "Detail Tweaker LoRA" by CyberAIchemist
is really good.
thank you so much!!
I'll try it out too!
I like the idea of artists using AI like any other tool, just not the practice. This is more complex than just learning art fundamentals! I feel like I'm falling behind...
I think it is how it should be used tbh. And dont worry, Its simpler than I make it seem in the video (i overcomplicated stuff bc of a lack of practice). Also, as time goes on the tools themselves advance and become easier to use
Thank you for this in depth tutorial!
Tyy for watching!
ok, so here is my question as someone who is studying traditional way of art making. In Art school there many talented art students to the point that if you're a talent artist yourself, doesn't really matter cuz everyone is both fast and good roughly. Although it takes multiple months or even years to improve depending on your commitment to your art. when I compare the usage of these tools to a traditional artist who has all the knowledge and the talent to execute what they want or been asked to do within 15 to 30 minutes. At what point all of those A.I. tools that we don't actually have much control of, so we keep on adding additional tools to help us gain more control over the output, does the process become even longer? To a point where simply knowing the traditional way of art making and being excellent at it, is much faster especially at the beginning of the process?
yup. You are 100% right. Or at least for now. Idk how aI will evolve. But atm a good and experimented artists is 20 times better than an AI, 100 times better.
Better quality, faster, understands what is wrong and how to change it, extreme control, etc.
To me, the point of AI at the moment, is to make it help you in parts that you dont really enjoy about drawing or art. Maybe even if you enjoy them, it can be a fast way to iterate and test stuff.
It can also help improve your own drawings (planning to make a video on that soon enough). Even though you'd still need the artistic eye to know what's missing and what you should correct. As well as a some level of skill to change it.
There are more points to this but thats a basic rundown of how I see it for now. Hope it helps!
@@Not4Talent_AI Thank you for your response, I agree that AI can be another tool for artist to use, and maybe help in exploring ideas. The only thing I fear is when people lose the knowledge of why does something look good as it does? When you look at an amateur artist versus one who has better understanding of the art making, you can clearly see that the better one has good understanding of shapes, light, and values awhile the amateur artist haven't gain good understanding and knowledge of those yet, so their work looks flat. It can take quite amount of time and practice to finally understand those elements.
With A.I. you're not really practicing art, you're just using a tool, as far as I can tell for now at least. I think the ones who'll be good with those tools are going to be like a good photographer who has understanding of shapes, light, and values with the additional understanding of the tool itself. But like you said we really don't have a clear idea of where all of this is going.
@@MustafaO.Artisan totally agree yeah. that's why I've never though artist would be obsolete.
I do think its something that takes out a major part of beginner or amateur artists. But knowing about all the things you mention is super important and will continue to be
How do I set my own character as the person who is going to be on the control net.
So basically I set poses with my designed then tell diffusion to put my character (that i have uploaded) .
Thank you
Use the regular approach with prompting, but using a LORA for your character
Is Latent Couple can use in google colab? I downloaded it from extension but it is not working when I clicked the "I've finished my sketch"
Not sure, i hear it is giving problems in the latest version of A1111. Maybe if you are on 1.7, try using 1.6
10:40 DOFUS!?
haahhahaha yep
Thx man for a very detailed dive in to the process, if you don't mind, how much time does it take to produce single image like this all together from start to finish involving all the nuances and hiccups that occur on the way?
Depends on the image and the result you want. The hardest here probably was the volleyball one, that took about 2-3 hours including planning and everything. Most of the time is just waiting for the AI to generate though. So I watch some series or exercice while I wait xD
Hi bro, thanks for the help. can you please put in the description the text for latent couple installing , you use at minute 4.46... I tried to write it from your screen but I can't see it completely... Please put this in the description so we can follow your steps.. Thanks again
oh, yes, my bad XD (done)
@@Not4Talent_AI Thanks so much bro.. a subscriber and a fan here XD
@@proyectodigital7915 Ty so much! Hope it helped :3
very good, I knew someone would figure out all these things, it just takes a lot of time, I have experimented a bit with trying to turn regular drawings into AI pictures and also img2img stuff, i've also used controlnet a good few times by now but the quality of the results I get vary a lot.
Lots of manual work required for now, but mostly depends on the image you want hahahaha
@@Not4Talent_AI tried a few but one I thought might work but didn't was a video with a limited number of similar frames and mosaic censor pattern in the mouth area, I think that aspect really confused the program.
@@ash0787 sounds interesting tbh. But yeah SD is very easily confused XD
@@Not4Talent_AI the industry if you can call it that yet, moves so fast ! if you only do a couple hours per week its hard to keep up, your programs UI looks totally different to mine for example. I rarely tried to get a perfect image and did high effort, but rather just try to get a feel for the limits of the technology at any given time, and enjoy the variety of things it outputs, it can be an aid for the imagination. I thought perhaps I could easily make a 'really cool' picture by adding several Lora files including a character, pose / object and a background, but this didn't exactly pan out and like you say I had to do some tricks and extra steps to try and make it work, but the result was imperfect.
@@ash0787 yep, will see how it changes over time. As you say, this sht moves so fast XD
Amazing vid, I was looking for a channel from someone doing some serious SD
Thank you so much!!! Hope to keep proving the value you are looking for !
AI will never be considered art and those who use it will never be considered artist period!
Dont care what it is considered tbh. Its fun anyway
I absolutely cannot get latent couple to work.
All forks of it are 6 months old and the mentioned git patch "does not apply", because the stuff it wants to alter is no longer in the place it used to be.
(Some "two-shot" script thing)
Is there some updated fork of latent couple that works with current A1111 V1.6.0 or later?
I'd delete the extension and install it from A111 directly again. (since some changes have been made and now it is easier to install, as it installs the same as every other extension)
If it still doesnt work, I'd go for Regional Prompter which acts pretty much the same way
I somehow got the "Two-shot" fork to work now. I think it got updated at some point.
Is latent couple still relevant with tiled diffusion's region prompt control?
Huh, I haven't even heard about that, do you know where I can read more? Sounds really interesting.
The contradictory thing about making videos is that I have less time to be up to date xD
8:48 Aaaaand now I have nightmares
Hahahahahha
Most people say that they use ai bcus they dont have time to learn to draw, but have the time time to learn all this shit xD
😂😂😂 to be fair its still easier than learning how to draw😂 but true hahhaa
Still not sure why some people are so sensitive whenever AI is used, even when the person is still drawing the image.
Apparently, all AI usage is bad, even if you pick up a pencil and draw, because "AI bAd!1!".
Pretty much 😅
Because AI learns from other people arts without their permission. You basically can copy a style of famous artist by one click. And that is NOT good.
@@olyabutorina6869 you can, (which I agree with it not being good). But it doesnt mean you always do
can't wait for next year to be easier!
and better hahahaha
one of my new fav YT channels 🙌
ty so much🧡
Awesome. Thanks a lot!
Ty for watching!!
GREAT! I am already getting sick of boring ai art on instagram. How long do you take for this piece ? (How much hour ?)
Ty!!
Depends on which one. The longes was the firstm it took about 3 hours but there was a lot of experimenting with different stuff.
Its usually about 2 hours for the others. (Most of the time is spent waiting for the AI to generate, mainly on the upscale)
@@Not4Talent_AI what gpu did you use ?
@@akbarramadhan2691 I have a 1080 TI
What was thge 3d software you used to pose the figures for the sword and gun?
Maya, but I have it bc of my work. If you want a free 3d software I recommend blender
Any ideas how we can get this thing to pixel art without the result of nightmares?
Hmmm not sure, havent tired. Good question though.
I would do pretty much the same vut with a pixelart focused model and a pixelart lora. And then see how that does
0:49 I'm just gonna leave this here...
not sure if you are referring to the🤨at the beginning of the second, or just the bs sketch xD
@@Not4Talent_AI "this image represents multiple characters in a b**ch
@@CubsYT oh, ahhahahaaha My pronunciation ain't great, that's why I had to put an image to make sure xD
@@Not4Talent_AI That's completely fine! I have an accent too, just thought the sentence sounded fun
@@CubsYT 😂😂
can't believe wolverine's cousin is such a good AI teacher
ahhahahahhahaha ty! I guess XD
You left Lalatina Dustiness out…
she probably likes feeling ignored
are loras not working even if you use composable lora?
Yeah😅 at least for me. It creates abominations
Loading new extensions from the extensions tab doesnt work for me. Literally nothing comes up when I search for openpose. Ive always had to go to hugging face to manually download them and drop them into my folders. Without this tool I cant follow the video. Anyone know what Im missing
wait, really? Does the CMD give you any errors when clicking "load available" ? All extensions have the GIT link in the description bellow just in case btw!
@@Not4Talent_AI Got past that hurdle for some reason after leaving and coming back it was working fine.... story of my life. my current error is the Latent couple text for cmd: copy paste from the description is not workin. getting an error -error: can't open patch 'extensions/sd-webui-latent-couple/0001-Adding-after_ui_callback-for-scripts.patch': No such file or directory-
@@scetchmonkey007 Try ignoring that for now and see if it has installed. I got a similar error and blind eyes fixed it xD
if not, I'll look into it
@@Not4Talent_AI yeah always good to step away for a while and come back. Right now Im going to see if I can find more info on the 3d openpose tool. Im a Maya guy so this is probably more my speed, and hopefully if used properly it can deal with the issue of the AI not making back facing characters that you ran into with the volleyball spiking character. Also should be better for keeping hands consistent
@@scetchmonkey007 Yeah hahhaa
Sure! you can also use this if you want, probably easier. I have a video on it and other tools: app.posemy.art
Really useful and can export OpenPose, canny, depth and normal maps.
Also a maya guy here btw xD
what do I do once I used open pose editor? Hitting send to txttoimg or imgtoimg does nothing controlnet doesn't recognize it as a picture its just sees a black square.
You can download the image and impor it manually. Its usually what I do
That doesn't work when i save the image and do it manually its the same controlnet just sees empty black image no pose@@Not4Talent_AI
@@sestep09 wtf, then Id re-install the extension or use an alternative. Cuz it should work if your downloaded image has the pose in it.
If it does but cn is the one that turns it into black. Try chacking if you have an active preprocessor. If you do, change it to "none" and try again
Thank you changing the preprocessor to none did the trick. None of openposes preprocessors worked and sure enough none did thanks again.@@Not4Talent_AI
@@sestep09 glad to hear that!! npnp
10:40 Best MMO EVER ^^ btw Photoshop beta lets you use SD to inpaint an image as a layer (so you keep the base image). This could reduce the number of steps in your workflow
hahaha no way someone recognized it xD
Oh wow that's actually pretty useful, I would love to try it, tyty!
@@Not4Talent_AI Since 2004 I spend like 25000 hours on Dofus (no I'm not an addict).
Normally, if you have a paid Photoshop subscription, you should have access to the beta. I haven't had any particular problems using it.
@@losing_interest_in_everything I used to play that so much, come back to it every once in a wile. Even met my gf on dofus touch 😂😂
I should try it, ngl. Im just that type of person that wont update a program they are used to using just because there is a minor change in something. (Would probably get used to it in 3 days). 😂
I use AI to helping with the planning stage of the art, then I use that as inspiration to draw from, seems like everyone wants to 100% use AI start to finish which I think isnt the point of AI art.
nice!!! I dont think using AI for a 100% of the process is the best way either, but it is a nice way to create and compensate for your shortcomings as an artist, while you learn how to overcome them
btw you can open cmd by typing "cmd" in the path in file explorer. Just make sure you're in the right folder
Can you do that for administrator too? That would be really nice ngl 😂
Thank you!
@@Not4Talent_AIYou can click the “file” button in upper left corner of file explorer and get to an admin prompt.
@@jdietzVispop cool, ty!! Ill keep it in mind
this is next level!! newb here but would you consider making a tutorial about making a scene? one place. multiple. same clothes.color. with different angles. consistency is what i struggle with. faces with loras is not bad but a scene. 3d background (sketchfab) for the photo angles and control net then impaint/seg like you went over. would love an expert to walkthrough that specifically.
thanks!!!
Now I'm doing an expressions video that will come out as soon as I'm feeling well and can record voice over.
Next I'll try to make a tutorial on lora sliders, but it shouldnt take too long. And after that I had planned a video on how to get spacial consistency and all that stuff. Tbh not fully sure how to do it for now. But I'll be experimenting a lot to bring ya'll the best methods I can
you should mention how the seed works in Inpaint too, but whatever. xd
Tbh, I dont know much about that, never use it. Could you explain please?
Ty!
@@Not4Talent_AI it can overbake the image afaik if you lock the seed and all that, you can try it out just in paint one area than use that as base image then repeat.
@@MrSongib I will, thank you!!
ill wait for the commercial version
try adobe Firefly then!
The problem for me is that, composition stuff and can be annoying af, since it's random. ill try some of your method sometimes. i did try to get my sketch working long time ago and I end up giving up at the time I just messing around with inpaint and SD base model so kinda hard to control stuff. xd
It is yeah, thats why I create a bunch of different controlnet models, to see if I can controll that better. Another option is photobashing and latent coulple with contolnet.
At that point you are using AI to mix stuff for you and that is it, but it may work😂
@@Not4Talent_AI latent couple like gambling aswell but can be good, last time I tried it's not working anymore on my 5700xt. maybe I should try your command at some point, maybe it can fix my problem.
@@MrSongib it might, if not, try reinstalling fully, maybe something got lost at some point
mazing一个
@Not4Talent hmm wouldn't there be a way to utilize unreal engine to create the scene? you could use cheap assets or free assets to create the beach and use the basic character models for the positioning of the characters?
Yeah, ofc. Or just painting directly xD But that wasnt really the point. To be honest creating that scene in unreal would be very hard due to the fact that you need rigged 3D models for all characters, then the time of modeling and stuff for the beach. The result would be 1000 times better and in video format, but the time invested in the models would be veeeeeery high ino
@@Not4Talent_AI hm probably so. i was thinking of just basic unreal models that come with the software for proper positioning of the characters so you could frame a scene. i think in the future that might be the more efficient way to do it but we aren't there yet. fully customizable themes with trained models i think would be the most top notch. will be very interesting to see how this technology improves in the near future
Ohhhh I see, I think I misunderstood the initial comment. But yeah, I hope to see this tech evolve to make stuff like that easier and faster @@Atticus_Moore
I wonder if Blender will ever support exporting rig data for OpenPose?
I think it does, just export an albedo map I guess@@APaleDot
This really should of been 3 videos one for each image... too much to process and follow along I just gave up after my last installation error.
Yeah, I didn't mean it to follow along but it is a common complaint I had. Next one I'll try to keep it one at a time, or at least easier to make ones. Sorry
Dude, your tutorials are good, but from a certain point you really go off on a tangent. From the middle of the video things become confusing and difficult to follow.
Do you feel this happens in every tut. Or just this one?
I can see how in this one I go off. But not sure about the others. I tend to go on tangents irl too, so ut wouldnt surprise me if people feel like thats the case. I need to understand it to fix it tho.
Thanks for the feedback!!
@@Not4Talent_AI Don't get me in the wrong way: your tutorials are gold, but from the middle of the video you give too much things for granted as if this was an advanced guide for expert people. Maybe, this is stuff too advanced for me. I will check for tutorials about Controlnet before rewatching this.
Oh, that could be. I tend to make tutorials for people that have some experience with the program already. As explaining everything everytime would be a pain.
Maybe I could create a series of shorts or more noob friendly videos so I can refere people to those instead of explaining it in the video?
Ill think about it.
Thank you so much!
@@Not4Talent_AI Ok, man. I will watch all of your tutorials from the first one, then I'll come back to see if I can handle this. Thank you.
Thanks! And srry for the extra effort. Still need to learn a bunch bout how to make proper tutorials 😂
That's a lot of work, I think you could get the same result using photobashing and actually painting it lol
Yes! You absolutely could. I talk about that in another video a little more slightly. I'll make a new video with more workflows eventually.
Though it is true that for some you could just photobash, for the volleyball image, for example, it is easier to make the controlent references and then phot bash from there. In my opinion ofc. There is soooo much testing to be done still.
I think I need to optimize more what can be done with controlnet, what with photobashing, what with inpainting... etc
FYI - You didn't link the video here you pointed to as a reference to watch @ 5:27
ty! its on the the suggested content. (could put the actual video on cuz I hadnt recorded it at the time xD mb) th-cam.com/video/SvGIkWFh-eo/w-d-xo.html
Wouldn't it be easier to learn how to draw?
Not really. you would get better results for sure. But learning how to draw takes way more time.
I'd like for this to act as a gate way for people to learn how to draw and improve upon AI's creations
Different people have different interests and skill sets.
Geez....at this point you can just paint it. It's faster lol. Great video though!
hahahhahahaha yeah, if you already know how to paint you'd get better results and faster just drawing xD
Seeing how someone like samdoesarts can paint masterpieces in 1hour, this is pretty lame
I'm still not anywhere near that drawing level tho xD
well explained and very entertaining video, i was just wondring if you can share with me the resault images you got in higher quality? the instagram ones are low quality (512x282), In any case, thanks for your help ^^
sure, add me here on discord and I'll send them! Not4Talent#9861
@@Not4Talent_AI ok, thank you ^^
Really helpful and informative videos, thanks for making them. keep up the good work pls :D Maybe you can make a video explaining what AfterDetailer is? thanks again.
I'll look into it! Thank you very much!
How much vram are you utilizing?
1080 Ti, so 11gb vram!
need more detailed describing of the process, some steps are skin so it hard to track the process
Agree! I have a bit of a new process, but I'll be posting 2 vids soon enough about it.
1 will be on this style, faster in nature. But then I'll also post a, howeverlong video of an image start to finish, while explaining a bit of my though process while making it. It might have some speed-ups on stuff like generating images, or iterating. But the rest will be slow and chill