Mastering ControlNet Scribble in Stable Diffusion: Transform Your Sketches into Stunning Images.

Подписаться 14 тыс.

Просмотров 34 тыс.

50% 1

In this video, we'll cover everything you need to know to master the ControlNet Scribble extension in Stable Diffusion. From understanding the XYZ plot and rescale mode to getting the most out of the preprocessor and CFG scale, we'll walk you through all the key features and functions of ControlNet Scribble. Plus, we'll share tips for upscale image techniques.
After this video, you can transform your sketches into stunning images.
📣📣📣I have just opened a Discord page to discuss SD and AI Art - common issues and news - join using the link: / discord
🤙🏻 Follow me on Medium to get my Newsletter:
- Get UNLIMITED access to all articles: / membership
- Laura: / lauracarnevali
- Intelligent Art: / intelligent
📌 Links:
ControlNet Paper: arxiv.org/abs/...
AI Image Upscaler: www.imgupscale...
Unsplash Picture: unsplash.com/p... (Photo by Chris Curry)
📰 Article:
/ mastering-controlnet-s...
00:10 What is ControlNet?
01:54 What is ControlNet Scribble?
02:57 Upload/Draw Your Image
04:16 What is the Preprocessor?
05:14 What is the Resize mode?
06:50 How to use the x/y/z plot
08:54 Preliminary results (no prompt)
09:11 Just Resive vs. Inner Fit
10:47 Inner Fit vs. Outer Fit
12:23 Sampling steps
13:05 CFG scale
15:16 Preliminary results (prompt)
17:52 Final result
07:03 Final thoughts
18:05 Upscale the image
18:56 See results in Google Drive
19:35 Final thoughts
#aiart #stablediffusion #generativeart #controlnet #scribbles

Опубликовано:

30 сен 2024

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист

Посмотреть позже

Комментарии : 51

@CoconutPete 7 месяцев назад

for some reason my canvas isn't showing a color... when I draw... hmm

@marcelocosta2612 Год назад

Wow, I've been looking for a tutorial on how to enhance sketches with Artificial Intelligence for two months now. The search ended with your video. Thank you very much!

@LaCarnevali Год назад

Nice! ☺️

@CoconutPete 7 месяцев назад

I'm having difficulty converting a color painting into something photo realistic.. almost seems like I have better luck with black and white line drawings vs color painting.. any suggestions are appreciated 🥴

@LaCarnevali 6 месяцев назад

try to use a different controlnet model, like softedge

@CoconutPete 6 месяцев назад

@@LaCarnevali will try it out... haven't used softedge yet. getting a little better but have still not had a breakthrough with sharp images that maintain the same structure. But, I have noticed adjusting prompts can help. I'll report back if I have any luck 😀

@LaCarnevali 6 месяцев назад

@@CoconutPete should also use a combination of controlnet. yeah let me know how it goes - not easy sometimes to get good results. Make sure you are using a good checkpoint (also consider that SD XL controlnet are not as good as SD 1.5)

@fullflowstudios Год назад

Hi Laura, I don't seem to have any of the ControlNet enabled or done something something wrong somewhere since I don't have the ControlNet selection pane in my options window. Also I don't seem to have StableDiffusion checkpoint v1.5 model.ckpt rather than the model.ckpt I see in your screen ... Care to help out here very please?

@LaCarnevali Год назад

Hi Hens, thank you for the comment - next time, I ll add this step :) Are you running it locally? When you run it locally, you need to download the ControlNet extensions. You need to "git clone" the ControlNet repo into the "stable-diffusion-webui > extensions" folder. After that, you need to download the safetensors from Hugging Face and move them to the "stable-diffusion-webui > extensions > sd-webui-controlnet > models" folder. ControlNet repo: github.com/Mikubill/sd-webui-controlnet Hugging Face safetensors (under the file and versions tab): huggingface.co/webui/ControlNet-modules-safetensors/tree/main If the above is not clear, I m explaining how to do that in this video at min 9.52: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-SktvO_OtnOQ.html The model.ckpt you see in the video is the v1-5 model but renamed

@smithyjake Год назад

to me, the final image looks worst than the original sketch. That sketch has artistic flare.

@AscendantStoic Год назад

You aren't really supposed to use a high quality detailed sketch for ControlNet, it can understand it's a dog from a much more simple drawing, you just need a simple sketch that captures that basic elements of a dog and minimum amount of details necessary, the examples on the ControlNet Huggingface page show that clearly.

@Ramirezz11 Год назад

@@AscendantStoicagree i think her sketch better with depth model.

@John_Krone Год назад

This is a great tutorial. very thorough

@kenyenmusic7548 6 месяцев назад

At 2:20 you said you’re using something to run it. What did you say? I can’t understand. Thanks in advance 👍

@LaCarnevali 6 месяцев назад

I say I am using Google Colab to run Stable Diffusion. Note: when I made the video, Google Colab was free. If you have a powerful computer you can run it locally.

@YOVOZOL 9 месяцев назад

Very helpful tutorial and this showed me the usefulness of the X/Y/Z plot! Also, perhaps you already know this now, but I noticed the double dog snout near the end and I read somewhere that this could be caused by your final canvas not being a true 512x512 image or multiple of 64, since I think stable diffusion is trained completely on 512x512 images.

@jamesbriggs Год назад

Super helpful thanks!

@fredmcveigh9877 Год назад

The canvas width and height in the controlnet section are only there for when you create a blank canvas to sketch .They make no difference to the controlnet settings .

@fred4838 10 месяцев назад

Thank you Luara just installed. How can I link my output to google drive, or have my output folder to export to my external dive?

@Jeanjean-nq8lk Год назад

Very helpful, thank for all your tutorial.

@NUOTHAENG 11 месяцев назад

i cant change the colour to draw with in the canvas? just white not black

@LaCarnevali 11 месяцев назад

canvas for controlnet is only white and black as you are building a mask

@SyncWaveAnimation 8 месяцев назад

x1.5

@MrSongib Год назад

Nice

@aleksandrekordzakhia4406 Год назад

Hi, I Can't find ControlNet button

@LaCarnevali Год назад

Hi Aleksandre, Are you running it locally? When you run it locally, you need to download the ControlNet extensions. You need to "git clone" the ControlNet repo into the "stable-diffusion-webui > extensions" folder. After that, you need to download the safetensors from Hugging Face and move them to the "stable-diffusion-webui > extensions > sd-webui-controlnet > models" folder. ControlNet repo: github.com/Mikubill/sd-webui-controlnet Hugging Face safetensors (under the file and versions tab): huggingface.co/webui/ControlNet-modules-safetensors/tree/main If the above is not clear, I m explaining how to do that in this video at min 9.52: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-SktvO_OtnOQ.html

@aleksandrekordzakhia4406 Год назад

@@LaCarnevali Thank you

@multimagical3137 Год назад

Brava, bella spiegazione. Questo è lo strumento che più di altri in controlnet fa arrabbiare i fotografi vecchio stampo.

@D.Art3Ambience 9 месяцев назад

Hello!! Thank you very much!! Where I can find the program?

@LaCarnevali 8 месяцев назад

What program? Scribble? huggingface.co/lllyasviel/sd-controlnet-scribble

@epelfeld Год назад

Very detailed and I liked the theoretical part. I understood now why I need to use x/y plot. If you could do also a video about creating ads in SD it will be great to look. I make cups from gypsum but not really good in product photography. I want to have a good looking Instagram with my goods. Found few videos about this topic but they are not working for me. So if you are searching for a new topic - maybe this is it.

@BasicShowTime Год назад

My scribble is not showing at the list, even stalling the lastest version, but the Model Scribble shows, does anyone can help me?

@LaCarnevali Год назад

Hi Lucas, you need to download scribble from HuggingFace using this link: huggingface.co/lllyasviel/ControlNet-v1-1/resolve/main/control_v11p_sd15_scribble.pth Then, you need to move the file in the models folder (see path below): stable-diffusion-webui/extensions/sd-webui-controlnet-main/models If you don't have the sd-webui-controlnet-main extension, you can git clone this repository: github.com/Mikubill/sd-webui-controlnet (you can watch this video at minute 9.52): ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-SktvO_OtnOQ.html

@chan-xs6lu Год назад

i don't know how much i thank you, thanks a lot, i've been looking for this, easy to understand! thank you so much again

@daizy5150 Год назад

just found your channel. great tips and advice. subbed! thank you for the info

@s11-informationatyourservi44 Год назад

thank you for this tutorial. i love your channel! subbed.

@mr.entezaee Год назад

O beautiful and kind lady. Help me how to improve an old photo in this artificial intelligence so that the basic characteristics of the photo do not change and at the same time the details of the photo look extremely natural. If there is a tutorial, I am very eager to make it because I have been looking for such a feature for years. But whatever I do, the characteristics of the photo change, and if I reduce the noise, the details of the photo are reduced. Sorry, I wrote this text with Google because I don't know English.

@LaCarnevali Год назад

Hi, no worries. Img2img is the way to go. The prompt description is important. You can then use Codeformer to improve the quality. For face expression, you can use ControlNet mediapipe. You might want to watch this video: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-DCfhLtv2IRk.html

@mr.entezaee Год назад

@@LaCarnevali 🖤🖤☺