When you select "only masked" and then set the resolution lower it's only using the lower resolution for the inpainted area. The reason the image overall still looks good is because it's the same resolution. Using a higher or lower resolution while inpainting a mask doesn't have any impact on anything other than the inpainted area. Using this you can actually get more detail in your inpainted area by maintaining the high resolution (either with 1x if using rescale-by or by manually typing a higher resolution) it will generate for example the face at the selected full-size resolution and then shrink the inpaint down to fit inside the overall resolution of the image.
Thank you for your insightful comment! You're absolutely right about how the 'only masked' function and resolution settings work in Automatic 1111. I must admit, there was a misunderstanding on my part regarding this. Your explanation is spot on and it's very helpful to me and, I'm sure, to other viewers as well. I will make sure to address this in a future video to correct this misunderstanding and to further enhance the learning experience for everyone. I truly appreciate your input and contribution to our community.
How do you make sure that you don't go too high in resolution and clash with the larger picture in amount of details? Is there a way to calculate the actual size of the selection? I'd be curious to use x/y/z plots for the whole process to save a lot of time.
But wouldn't the resolution make a difference to what result you get? Generating a 1024x1024 image gives you a different image compared to 512x512. So if he wanted to get something close to original as far as general image properties, it would make sense to do the inpainting at that original resolution. Unless it really doesn't matter in this particular case...
I got into stable diffusion last year and i had a lot of fun figuring out how everything works by myself. I'm pretty much self taught so i knew there would be some features i didn't know about so i thought i would finally look up how other people do their generations. Boy did i need to hear about the inpainting function. I knew what it was supposed to do but i could never figure out the specifics of how to get it to work. This is a major game changer for me! Thank you so much! I love the german accent by the way. You sound like a very friendly person.
I'm so glad to hear that the inpainting function has made such a difference for you! It's always rewarding to discover new features that enhance our creative process. Keep experimenting! Thanks for the remarks on my accent. 😊
I'm really glad to hear that you found the tips helpful! I strive to make my content as clear and informative as possible, so it's fantastic to know it's hitting the mark. Thanks for the kind words!
That's fantastic to hear! The AI-powered img2img upscale truly is a game-changer for revitalizing images. I'm thrilled it helped you save an image. Thanks for watching and sharing your success!
Thank you so much for your kind words! I'm really glad to hear that you found the tutorial helpful and of high quality. It's comments like yours that motivate me to keep creating and sharing more content. Stay tuned for more tutorials!
This video has some serious early 2010 movie maker vibes - you just know it's gonna be good and helpful!. And bam! I wasn't dissapointed - super precise & to the point. A complete workflow highlighting each step, thanks so much 👍 The german accent is just the cherry on top 🍒
Thank you for the fantastic feedback! Thrilled to hear the video hit the mark with my content. Your appreciation, especially for the German accent, brings a huge smile to my face! 🍒 Exciting news: the next video will also focus on a workflow, this time harnessing the power of ControlNet. Your support inspires me to keep creating. Don't miss out-thanks for being an awesome part of our community!
@@AIKnowledge2Go good to hear that I could put a smile on your face 😋 I’m totally excited for the next video as well! ControlNet is so far the one thing I haven't touched, but would love to know more about and slowly master.
Wow, thank you so much for the high praise and for subscribing! I'm thrilled to hear that you found the tutorial to be so valuable. Your support means a lot, and it motivates me to keep creating high-quality content. Stay tuned for more AI insights and tutorials!
I know this might get lost here, but, thank you, you really made my day, been a bit sad lately with life and everything and I just wanted to learn something new to keep myself occupied. With something. Just to not think about it all and produce art. You made that all possible. So thank you and I hope that you know that you made someone's life better.
I'm deeply touched by your words, and I'm so grateful to know that my videos have been able to provide you with some comfort and a positive distraction during this time. Life can be challenging, but remember that it's okay to take time for yourself and do something that brings you joy. The beauty of creating art is that it allows us to express ourselves, to lose ourselves in the process, and to make sense of our experiences. Never hesitate to reach out if you have any questions or just want to share your creations. Thank you for being a part of this community, and please take care of yourself.
Thank you for your suggestion! The combination of controlnet tile and ultimate SD upscale is indeed a powerful technique for getting more detail. I have plans to cover this and other advanced techniques in an upcoming part II tutorial. Stay tuned for that, and I appreciate your input!
I'm sorry to hear that you're feeling overwhelmed. I'd suggest starting with my basic tutorial to build a foundation, then gradually progress from there. Here's the link to help you get started: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-SHikMK39Q30.html.
Thank you for your kind words and subscribing. There's actually an improved version of this workflow you can find in this video ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-wyDRHRuHbAU.html it uses control net so it's a little more advanced. Happy creating
Thank you for your feedback! I'm glad to hear that you found the video informative and valuable. The next Video releases in a few minutes actually :) Stay tuned!
I'm glad my antics gave you a chuckle! I admit, sometimes my attention wand... - "Oh, look a bird!"😂 Jokes aside, there was some re-recording involved in creating the video. When it came to editing, I figured it would be easier to add some explanatory text rather than reshoot everything again. Thanks for noticing and watching!
Hi und danke für dein Feedback! Den Akzent bekomme ich einfach nicht raus 😂. Zum Glück scheinen ihn die meisten englischsprachigen Zuschauer eher amüsant als nervig zu finden.
Thanks so much for your excitement about Part 2! 🌟 Your wait is over: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-mrWmEWEZwDw.html. Just a heads-up: since Parts I and II have been around for a bit, some of the settings might have changed. I'm planning to update them as soon as I can find a slot in my busy schedule. Stay tuned and happy creating! Your support means a lot! 🚀
@@Asterex versuchs mit negative embeddings wie z.b.civitai.com/models/116230/bad-hands-5 Ansonsten hat After Detailer auch die möglichkeit Hände zu "inpainten" zu after Deatiler hab ich auch ein Video. Allerdings ist da der Fokus auf Gesichtern.
with the same settings and just some changes to the negative prompt I was able to get much better faces, limbs, and less disfigured limbs with initial generation making the later inpainting much quicker.
You've made a great observation! This video primarily showcases this particular workflow. However, it's worth noting that when your prompts become more complex, or you're using multiple 'loras' as is often the case for creating stunning art in SD 1.5, the quality of faces tends to diminish. That’s exactly where this workflow proves to be incredibly useful.
Absolutely! Adetailer has been a game-changer. If you haven't seen it yet, check out my video on it; I delve into its benefits and how it can save you a lot of time. Cheers!
My 1080 can't upscale so much. Welp. But nice demonstration. That model is right up my alley. I hope the dev will resume maintaining it after his hiatus.
I understand your concerns with the 1080. To address the VRAM issue, you might want to try adding the `--lowVram` or `--medVram` parameters when starting with webui-user.batch. Another approach for upscaling with limited VRAM is to utilize Controlnet Tile. It essentially breaks your image into smaller tiles and scales each one separately. This can be especially helpful for hardware with memory limitations. Hope that helps
Where do I find the Controlnet Tile? I'm not ready to follow your tutorial yet cos my internet is being throttled and that file for the revAnimated model you suggested is taking hours to download :/@@AIKnowledge2Go
I'm glad to hear that tip was helpful for you! I dont know why this is not fixed already. Maybe one day i'll write a bug ticket in Automatic 1111 github on the other hand, this move has become second nature to me. So even if this bug is fixed one day i will clear the mask anyway. :) Thanks for watching and happy creating!.
A good beginner/intermediate tutorial, I have been using a very similar workflow for a while, except I gave up using extras to upscale (because it never looked as good as I wanted) and now use SD Ultimate upscale Script in conjunction with Control Net Tile (CN tile not 100% necessary but can improve the image) as a final step, then maybe some touchups in GIMP if needed.
That sounds like a solid workflow you've developed there! Thank you for sharing it with us. It's always interesting to hear about different methods people are using. Your suggestion about using Control Net Tile has come up a few times and I can see the value it brings. In light of these discussions, I'm planning to create a Workflow Part II tutorial where I'll cover this more advanced technique. Stay tuned for that!
I'm thrilled to hear that this has reignited your interest in building virtual worlds! It's such a fascinating field with so much potential. I'm glad my content could be part of your renewed journey. Thanks for sharing your experience and happy world-building!
I've been tinkering with BlenderAI, Stable-Diffusion that works inside of the Blender modelling system. Blender renders a source image which is passed into Diffusion (along with animated prompts!) The Blender scenes (3d animations) are a great starting point, but I can see that using this work-flow is going to greatly improve my results. My current project is to 'convert' a short video clip into a nightmarish vision using BlenderAI to re-work frames of video. Being able to play with 'sliders' on a frame-by-frame basis is pretty wild!
I use Draw Things on my iPad and iPhone but even so this video has been very helpful! Your presentation is clear and simple, with no filler. Fantastic!
Draw Things is indeed a great piece of software. I'm thrilled that you found the video helpful and appreciate your kind words about the presentation. Thanks for taking the time to comment!
Do you struggle with prompting? 🌟 Download a sneak peak of my prompt guide 🌟 No Membership needed: ⬇ Head over to my Patreon to grab your free copy now! ⬇ www.patreon.com/posts/sneak-peek-alert-90799508?Link&
You're getting a hand and a robot thing in your leg in painting because it's still part of the original prompt. It's important to trim out portions that aren't applicable to what you're inpainting.
You're absolutely right. My skills in inpainting have improved since then. I'm considering re-recording this video in the future. Thanks for pointing that out, happy creating.
Vielen Dank für dein positives Feedback! Es freut mich zu hören, dass dir das Video gefallen hat. Ich hoffe, dass die Tipps und Tricks dir beim Ausprobieren helfen werden. Viel Spaß dabei!
adding keywords for face tend to decrease chances of getting mangled faces you will have to fix later. Try (highly detailed face, textured skin, detailed eyes)
That's a great tip, thanks for sharing! However, it's also worth mentioning that there will come a time when you'll want to upscale your image to enhance the overall quality. In that case, you might still need to deal with some mangled features, but hopefully fewer with your suggested keywords.
Good video and explanation! I'm just a bit confused about one thing, during inpainting is it really necessary to have an accurate prompt, for what you want specifically in that area? And what happens when you leave it blank, does it just try to "autofill"?
Hey there! Your prompt can guide the AI for better inpainting, but feel free to experiment and see what surprises the blank canvas brings! What i didn't do in my Video is to change the prompt. I actually should have changed it. Leaving it blank can work, but i suggest you use ControlNet Inpaint.
Hello, could you please tell me how you enabled the bars on the side of your image on 4:16? The ones that let you pull the image up and down and to the sides. I can't find anywhere on how to enable them, for me images just get squeezed smaller and for really wide images that makes it so hard to work with.
Oh, those scroll bars were actually a result of using ctrl + mousewheel to zoom in on my browser. It appears that feature might have been changed or removed in newer versions. 😞 However, I'd recommend the canvas zoom extension for a1111. You can find it under extensions -> available -> load from, or check these direct links. haven't tried it myself but it seems to do what you are looking for: github.com/richrobber2/canvas-zoom Hope this helps!
Danke dir für das hilfreiche Video, Chris! Gibt mir eine super Hilfe zum Start! Funktioniert der Workflow so grundsätzlich auch für fotorealistische Bilder (oder auch für anderes)? Natürlich dann mit anderem Modell und Parametern denke ich...
Danke für das Feedback! Es freut mich sehr, dass dir das Video geholfen hat. Ja, ich verwende diesen Workflow für alle Arten von Bildern, sei es Anime oder fotorealistisch. Gelegentlich nutze ich den "after! detailer" speziell für Gesichter. Falls du damit noch nicht vertraut bist, habe ich ein Video dazu auf meinem Kanal.
Thank you for your enthusiasm! It's great to know you found the video helpful. I'm excited to continue creating more content for you and the community. Stay tuned for more!
"Before watching your video clip, I kept trying to use 'Hires fix' foolishly, and the result is that the pictures look very, very bad. Thank you very much." and I'm a newbie
I totally get that! The 'Hires fix' can be quite tempting for many, especially when you're just starting out. I've been there. Glad my video could steer you in a different direction. Happy creating.
@@AIKnowledge2Go "My English is quite poor, I can only follow what I see. I have to watch your video clip a few times before I can do it, well, a bit slow but anyway I feel lucky that you shared your experience. Once again, thank you very much." Chat GPT Translate :D
Hey, wirklich ein sehr gutes Tutorial, danke dafür. Ich habe noch ein Problem, vielleicht kannst du helfen: Wenn ich in InPaint einen bestimmten Bereich maskiert habe, sagen wir zum Beispiel die Beine, und dann auf generate drücke, macht er mir keine neuen Beine, sondern es sieht so aus, als würde er das komplette Bild noch mal in dem maskierten Bereich erstellen. Hab die Einstellungen aus dem Video übernommen und auch mit verschiedenen Noise Strength experimentiert.
Danke für das Feedback. Wenn er das oder ein ähnliches Bild noch einmal erzeugt in dem maskierten Bereich, hast die denoising strengh zu hoch. Was ich in dem Video nicht zeige (weil ich es damals nicht besser wusste) ist, das du dein Prompt anpassen sollst/kannst/musst. Wenn du z.b. ein Gesicht inpainten möchtest, dann schreib sowas wie "image of a face of..." was du im Prompt drin lassen sollest ist alles was das rendering betrifft (HDR, 4K, cinematic shot). Inpainten ist immer auch ein bisschen Trial and error. Hope that helps.
I'm really glad to hear that you found the tips and workflow helpful! Sharing this knowledge and helping others is exactly why I create these videos. Your support and feedback are much appreciated. Thank you!
I'm so glad to hear that the video was helpful for you! It's wonderful that you're now able to use the tool more effectively. Thanks for watching and happy creating!
To see the 'resize by' option, you may need to update your version of Automatic 1111. The latest version as of today is 1.3.2. I've confirmed that this feature isn't related to any extensions, as I still had access to it even after disabling all of my extensions. Please try updating your version.
If you're interested in learning how to save time inpainting body parts using a handy tool in Automatic 1111, I recommend checking out this video next: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-y3DxX9s0NhQ.html
Really good video :) btw you think the "vae-ft-mse" is worth having? since I got it but can't see it anywhere in my stable diffusion so if its worth trying to get it to work?
@@Marcus-si7su Thanks for your feedback. You can set up vae-ft-mse in A1111 under Settings->Quicksettings list-> there you add Sd vae and restart A1111 then you get the dropdown. But you will have to download the VAE.
To the entire ai art community: You lack of creativity is astounding, here you have this amazing new art tool that allows a new multiverse of possibilities, and all you guys come up with is essentially the same few pinup girls. It's quite cringe guys.
This is more or less my workflow, only that I prefer DDIM over Euler a, and DPM++ 2S a Karras over DPM++ 2M Karras. I didn't know when I was first experimenting, but the two that I picked add noise between steps, which further randomizes and creates variation. Most other samplers(ones that aren't DDIM or have 'a' in the name), end up converging at about 150 steps to the same result because they don't add noise between steps.
Interesting! I've noticed similar patterns with different samplers. The noise addition in DDIM and those with 'a' really makes a difference in creating unique variations. I'll definitely consider experimenting more with DPM++ 2Sa for refinement. Thanks for sharing your insights!
I would use it while inpainting. While this image has only one person, sometimes you want to create setups with multiple people. And masking out a face and writing the prompt for the face in After Detailer, or using a Lora/TI/Lycoris, can be helpful. I hope that helps you on your creative journey.
Hello @AIknowledge2Go, this is a really helpful video. Thank you for coming up wirh such a great content. I have a quick question here and would appreciate if you could provide some valuable suggestion. I have always been having challenge in worling with image2image generation, seems like the final result is no where close to my input image. For example, if I want to make minor edit to my normal human images like change clothes color, changing hairstyle, changing hairlength etc by keeping rest of the entire details intact should i go for image2image option in SD or should i use some other method for this? I am using absolutereality Checkpoint for this to ensure the pics are realistic. Any advise / suggestion would be greatly appreciated
Thank you for your kind words! If you want to make minor edits to your normal human images while keeping the rest of the details intact, using the inpaint option of Image2Image in Automatic 1111 is a good choice. However, for specific modifications like changing clothes color, hairstyle, or hair length, you may need to experiment with different prompts and parameters to achieve the desired results. In addition to Image2Image, you can also explore the ControlNet model for inpainting, as it can be effective in preserving the overall details while making specific modifications. Remember to adjust your prompt accordingly to focus on the areas you want to edit. It's important to experiment and iterate with different prompts, models, and parameters to achieve the desired outcome.
@@AIKnowledge2Go Thank you so much for your response . I totally agree , detailed prompt + config details are the key here, which I haven't mastered yet, still learning 🙂. Btw, if you don't mind me asking I would like to know if there is a way I can have a model trained to simply take input and change hairstyle on the same pic as output. Considering I have to work on multiple images, it would be difficult to keep writing prompt on each image to get the desired result. Thank you so much for looking into this 🙂
mehr als eindeutig du bist deutsch ^^ allerdings muss ich sagen es war schwer zu verstehen, was was wirklich macht. ich nutze easy Diffusion als web ui auf meinem poweredge r630. es war viel information gepackt in einem video allerdings hätte ich gern ein paar Beispiele gesehen. würde mich gern über eine video serie freuen die frei nach dem motte. lasst uns heute mal ein bild erstellen und dann vielleicht irgendwelche themen nehmen oder random ein paar worte zusammen werfen oder so. also quasi was man so als anfä´nger auch tun würde. wäre sicher interressant dich einfach mal bei dem prozess zu begleiten und auch deine hinter gründe zu hören wie und warum du das so jetzt machts. aber sehr informativ und hat mir schon ein ganzes stück weiter geholfen
Ha da hast du mich erwischt 😂 Deutscher durch und durch😂 Du hu hast absolut recht, das Video richtet sich auch eher an Leute die bereits Berührungspunkte mit Automatic 1111 hatten. Ich kann dir dieses Video von mir empfehlen da erkläre ich die einzelnen Settings: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-SHikMK39Q30.html Wenn du der Liste folgst da hab ich einzelne Themenbereiche in Reihenfolge abgedeckt. Ich hoffe das hilft dir weiter.
Do you have some source to adjust our research, a lot of parameters I've seen in this video, weren't showed anywhere, I assume part of it is emperical knowledge, still, if you can link us to great indept tutorial explan,ation it would save me a lot of hardwork ! Still I learn a ton from that video, and helped me adjust some issues i've got manytimes !
Your response is a good start, as it acknowledges the viewer's appreciation and seeks to clarify their specific needs. You could enhance it by offering direct links to your relevant videos and expressing a willingness to assist further. Here’s a revised version of your response: Hi there! I'm thrilled to hear that my video was informative and helpful for you. Regarding the settings you're curious about, could you specify which parameters or aspects you're looking into? This will help me guide you better. Also, I have two basic videos on prompting and basic settings on my channel which might be just what you're looking for. I'll drop the links here for easy access: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-MftRapF4AaU.html ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-SHikMK39Q30.html Feel free to check these out. Also i have whole tutorial Series on stable diffusion.
hi there! thank you for the guide and tutorial. Very useful information especially to users like me that only know how to put in prompts and click generate. Thank you again!
You're very welcome! I'm glad to hear that you found the guide and tutorial helpful. It's always my aim to make the process of AI art generation more accessible and easy to understand. Don't hesitate to explore my other videos for more tips and tricks. Thank you for your kind words and support. Stay tuned for future videos!
You're very confident, first video I watched of you. How long did it take you to self learn this? There's very little I completely understand from tweaking arbitrary values and knowing how that changes the image. I just read about how to get the software running on google colab which I never knew existed or how anyone would use it until now.
I appreciate your observation! It took me several evenings of dedicated learning and experimentation to grasp everything fully and understand the intricacies of the values. If Google collab works for you great. I read that A1111 would not run anymore on collab, but that info is 2 Month old at least. If you find yourself challenged by the basics, I recommend starting with my foundational tutorials. Here's a playlist that can guide you: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-SHikMK39Q30.html.
Do you do tutorials for us cheapskates who can't afford a powerful enough computer? I'm on a $1000 dollar Windows Surface 4, using GPU Radeon Ryzen (?) and it's taken me almost 24 hours to get it up and running to an acceptable level (lots of code changing and deleting venv files, and setting (having lol) the arguments to make it work on this computer. It's been an absolute nightmare for me, but a small render with a handful of prompts is now coming in at under 2 minutes. Another problem I;m having is running out of GPU memory when I'm asking it to do when I use too many prompts or sampling steps etc. It's been a real head scratcher! @@AIKnowledge2Go
When you refering to refiner you mean SD XL Models right? Actually i havent done upscaling with it. My system crashes when i go higher than 1024 x 1024 with sdxl models. I still use a lot SD 1.5 because as of now in my opinion you can get better results with 1.5 if you know what you are doing.
Omg you`re a life saver!!! I have one issue tho and maybe more people would relate: Inpainting does not work for me, i tried giving it prompts, paining one small area, multiple, restarted the whole webui, just nothing seems to work. I can see the image being rendered nicely and it looks good but i get the same result in my folders. Does anyone know any fix?
Thank you for the kind words! Regarding your issue, I've experienced similar problems when using ControlNet for inpainting in the current version. If you're not using ControlNet, perhaps a fresh installation might help resolve the problem?
@AIKnowledge2Go Hello! I've solved my issue, I don't have them right now but I've had to put some commands in the webui batch file and it works fine. I don't know why, maybe because I have an AMD computer
Thank you so much for your kind words and support! I'm thrilled to hear you found the video helpful. Cheers to more learning and exploration together! 🍻
Hi yes it is. Maybe you want to use controlNet for this. Here is a newer version of this video. ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-wyDRHRuHbAU.html To remove objects just write what you want to have instead on the background for prompt when inpainting. You need to experiment with the denoising strength.
Nice video.. 1 question, when I try pinpointing do you need to change the prompt at all? I keep getting strange results, for example, I inpatient the face, but then I get a whole corrupt face, sometimes with bodies and all sorts of things in it. Any tips?
Thank you for your feedback. Regarding pinpointing, adjusting the prompt can often help in achieving the desired results. It sounds like the issue might be related to the denoising strength; consider reducing it. If you're having specific problems with faces, I recommend checking out my video on 'After Detailer'. This tool can realy make inpainting face effortless
Hi, Can you explain this please. why do you need to generate first set of images with Euler A? when you can use the DPM++ 2M KARAS in the first place. first set of images generated deformed faces. I do not understand what are the benefits of doing the additional steps? thanks
Hello! I appreciate your curiosity. The choice to start with Euler A isn't mandatory, it's more of a strategic move. Recently, I've been leaning more towards using DPM++ 2M Karras from the get-go. As for the additional steps, it comes down to the limitations and preferences in image resolutions. Stable Diffusion 1.5 typically restricts us to generating images at 768 x 512 or 512 x 768, and occasionally 768 x 768. I like to stretch those boundaries to achieve higher resolutions, like 2K or even 4K. By using the Image2Image upscale method, I feel I have a tighter grip on the upscaling process, compared to the highres fix workflow. Think of it like having a finer zoom lens for the details when you're aiming for that perfect shot in photography. It's about crafting the image with as much precision as possible. I hope this gives you a clearer picture of why those extra steps might be worth the effort!
First of all, a greeting and thanks for the video, a beautiful image. I wanted to ask you why you didn't use the High res fix, from what I understand what it does is the same as you did but saving you a step, that is, starting an imgtoimg process. What is the reason that you advise against its use?
Thank you for your comment and your kind words about the video. @hoasiai is spot on. The High Res Fix does indeed have the potential to significantly alter the image, as does changing the sampler. My preference is for a straightforward workflow. I start with prompt engineering, and once I'm satisfied with the composition, I move to Image2Image to boost the quality when changing the sampler, without affecting the composition. I hope this clarifies my approach, and thanks again for your question!
Klasse Tutorials! Durch deine Videos sind meine Ergebnisse um ein vielfaches besser geworden! Ich habe nur ein Problem mit den "Inpaintings". Ich habe es Schritt für Schritt so gemacht wie du in deinem Video, aber bei mir verändert sich so gut wie nichts. Egal welche Einstellung ich verändere, selbst wenn ich die "Denoising strength" komplett hochstelle oder den "Seed" auf -1, bekomme ich nach dem Generieren 4x nahezu das selbe Ergebnis. Hast du zufällig ne Idee woran das liegen könnte?
One approach is to first generate an image without a person and then use inpainting to add the character. Alternatively, you could use a segmentation model in ControlNet to achieve this. If you're looking to remove the background from an existing image with a character, using an external background removal tool would be the way to go.
To have more control I suggest you use controlNet. I have a newer version of this video right here: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-wyDRHRuHbAU.html
The ideal computer for running Automatic1111 largely depends on your budget. If cost is not a concern, I would suggest going for a system with an Nvidia RTX 4090 graphics card and an Intel I7-9700K processor or faster. However, keep in mind that when it comes to running Automatic1111, the GPU and VRAM are significantly more important than the CPU.
How much was your computer? lol I mean, I'm utterly jealous by how fast yours works, and I don't know where to even start looking for good brands @@AIKnowledge2Go
I'd actually like to know more about the syntax, what to set in brackets, the values and what options there are I see a lot of stuff like (purple hair:1.2) sometimes with or without colon, sometimes with multiple brackets and so on... While I use prompts like that myself and play around with it, i feel like I roughly get how to use it, but i could very well be completely wrong on this as well If you already have a video on this just let me know, otherwise this would make a very good video imo
Absolutely understand your query! Check out this video for a detailed understanding: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-IEYMVIbPbQQ.html However, if that feels too advanced, I recommend starting with this one: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-SHikMK39Q30.html I'm also currently working on a comprehensive guide on prompting. Stay tuned!
Took hours to generate the first set of 8 images... My graphics card is "MSI Gaming GeForce RTX 3070 LHR 8GB GDRR6 256-Bit HDMI/DP Nvlink Torx Fan 4 RGB Ampere Architecture OC Graphics Card (RTX 3070 Gaming Z Trio 8G LHR)" Is this normal/expected? Every other setting I think I got to match yours in 1111. Thanks for the great video!
10 Minutes still is very long with an RTX 3070. I needed about 1,5 Minutes for 8 Images with my old 2080 Super. Do you have Xformers installed? In your stable diffusion web ui folder find the webui-user.bat file. Open it with a texteditor like notepad. Add --xformers if your "set COMMANDLINE_ARGS=" does not have it. so should look like this: set COMMANDLINE_ARGS= --xformers or similar, also between set COMMANDLINE_ARGS= and call webui.bat write in a new line "git pull".This keeps automatic 1111 up to date. For Xformers to install sometimes takes 3 - 4 restarts of the automatic 1111 server.It's Strange. After you saved the file you have to start A1111 via the webui-user.bat instead of the webui.bat file. Hope that helps.
Dman this is very nice workflow , i usually use ad detailer to fix face from the start , and for upscaling i go to img2img and use tiling with ultimate upscaling. I dont know why....thats just the workflow ive had. Ill try this now
It's always interesting to hear about others' workflows! The beauty of these AI tools is that there are so many ways to use them. It sounds like you have a process that works well for you, which is fantastic. My workflow is just one way to approach creating AI art, and I encourage you to try it out and see if it suits your style. Thanks for watching and happy creating!
nice video bro, subscribed. One question, if i just need to get rid of something and i want that part to blend in with the background what should i do? how i copy patterns, for example wood tiles in from parquet?
Thank you so much for your kind words and your subscription. For your question, if you want to remove something and have it blend with the background, you can use the 'Inpainting' option in the Automatic1111 workflow. You will have to change the 'Masked Content' setting to 'Latent Noise' and also modify your prompt accordingly. Please make sure to set the 'Denoising Strength' to 1, otherwise, you might end up with a pixelated result. As for copying patterns, I would recommend using the ControlNet tool. While I don't have a specific tutorial for your exact case, I do have two very comprehensive tutorials on how to use ControlNet on my RU-vid channel. Here is one that might be helpful to you: Basic:ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-dgRtt_yt3yA.html Using two contorlnet models at once: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-RY5XqF38DFg.html
Yes there is. Just change the prompt. In fact that is the reason why is have this little "spider thingy" on her leg, because i did not change the prompt. Happy Creating.
Love this video. I mainly learn through RU-vid Vids, and I've been an avid Stable Diffusion user since the onset. Keep making content like this that explains your actions & choices, and I know you're gonna have tons of subs soon.
Thank you so much for your kind words and support! I'm thrilled to hear that you find the videos helpful. I will certainly continue to create content that explains my process in detail. Your feedback and encouragement mean a lot to me. Stay tuned for more content!
would you believe that as soon as you said "clip skip" I paused the video and started googling how to pin it at the top of the UI instead of watching 0.001 more seconds until you explained it LOL
Great question! It might seem super fast, but that's actually an illusion - I sped up the video for better engagement. The creation you saw was done with my old Nvidia 2080 Super. However, my current setup includes an Nvidia 4080 RTX and an Intel Core™ i9-13900K, which is indeed quite fast. To give you an idea, it can render 8 images at 512 x 768 resolution in about 35 to 40 seconds.