SDXL in ComfyUI!

Рет қаралды 8,833

11 ай бұрын

Finally, Stable Diffusion XL 0.9 (SD XL) is out (research-only)! #stablediffusion #sd
SD XL 0.9 is the new Stability AI model, which outperforms the previous SD 1.5 and SD 2.0. More realistic pictures, although still some limitations...
In this tutorial, we are going to build together the ComfyUI workflow to run the new SD XL Base & Refiner! We are going to download the new model from Hugging Face, ComfyUI for Windows, Mac, and Colab, build the workflows, check for outputs, and save&reopen created workflows!
📣📣📣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
📰Article:
/ stable-diffusion-xl-in...
📌 Links:
SD XL Paper: arxiv.org/abs/2307.01952
SD XL Base Download: huggingface.co/stabilityai/st...
SD XL Refiner Download: huggingface.co/stabilityai/st...
ComfyUI GitHub: github.com/comfyanonymous/Com...
Colab: colab.research.google.com/git...
A1111 Extension (no ControlNet, no LoRA): github.com/lifeisboringsoprog...
Warp: app.warp.dev/referral/2EYMJL
00:14 SD XL model briefly explained
01:37 Limitations of the new model
03:27 ComfyUI
04:14 What to expect from this tutorial
05:18 ComfyUI performance for Windows users
05:18 ComfyUI performance for Mac users
06:29 ComfyUI performance on Colab
07:00 ComfyUI is available on RunPod
07:20 Download ComfyUI on Windows
10:44 Download the SD XL models from Hugging Face
14:03 Load the default ComfUI workflow
14:40 Download ComfyUI on Mac
19:31 Run ComfyUI on Colab
20:03 Different ways of using Colab
20:07 Connect to Google Drive
20:46 Connect to an external computer
21:42 Set Runtime Type to GPU in Colab
22:39 Download any model with Colab
24:07 Download SD XL models with Colab
25:29 Create an access token in Hugging Face
28:03 How to use ComfyUI
29:31 Create nodes and links
30:34 Build the workflow to run SD XL Base
30:41 Personalize nodes
31:38 What is CLIP?
33:10 Add the VAE loader (for external VAE)
34:30 Update prefix when saving images
35:31 Build the workflow to run SD XL Refiner
37:08 Collapse nodes
38:40 Latent image for the refiner model
40:37 Denoise in the Refiner KSampler
41:45 Compare outputs base vs. refiner
42:35 Save the workflow
42:53 Load a saved workflow
43:52 Where to find the output
44:05 Re-initialize ComfyUI with Colab
44:37 Re-initialize ComfyUI with Mac
44:42 Re-initialize ComfyUI with Windows
#aiart #stablediffusion #generativeart #stabilityai #stablediffusiontutorial
#sdxl

Пікірлер: 39

@Superficialgallery 11 ай бұрын

You are getting better and better all the time. Seriously good stuff.

@jejakpotret 11 ай бұрын

This is what most I'm waiting for, thanks a lot, you are the best !

@edsonjr-dev 7 ай бұрын

Thank you for this incredible tutorial, my like and watchtime 100% 💜💜💜

@cunningfilms8728 11 ай бұрын

Amazing detail as always thank you!

@magnoalm1 11 ай бұрын

Thanks Laura, I finally understood how the ComfyUI interface works

@jokosalsa 10 ай бұрын

Laura, great channel. Great explanations. Subscribed

@ronnykhalil 11 ай бұрын

thank uuuu!!!! so helpful

@avinashu118 11 ай бұрын

10:25 u can edit the "extra _model _path.yaml.example" file name to "extra_model_path.yaml" file in `comfy ui\ComfyUI_windows_portable\ComfyUI` and inside the file ,under checkpoints u can copy the path name from the automatic 1111 model folders ,it will automatically load the models files when u run comfy ui ,so there is no need to copy the checkpoint again inside the models folder in comfy ui, it saves u a lots of space great video btw!

@Luxcium 4 ай бұрын

_I wanted to learn how to use ComfyUI with a LoRA_ but *I am happy to be learning how to use ComfyUI with Laura 🎉🎉🎉🎉*

@DanielBlumberg 11 ай бұрын

Really liked this walk through. I was able to finally figure out comfy on windows... Mac did not get it to install. Could you do a review of Warp since you used it instead of terminal?

@macgee 11 ай бұрын

nice walk through thanks laura - reason your first render didn't look good is because this model performs best at 102x1024. i got decent results with 768 as well but the refiner didn't do it's job nicely 1024 is exceptional on a simple prompt this UI is incredible to learn ins and outs SD but also to build you own custom routings many video compositing applications are node based, it's so flexible

@sitgesstudio 5 ай бұрын

a computer geek with a nice tan, thank you grazie mille

@lakislambrianides7619 11 ай бұрын

Great video. I don't see any cancel button for cancelling rendering

@agustinpelliza9501 8 ай бұрын

Hey Laura. The manager doesn´t appear in my interface. I have the "ComfyUI-Manager" folder added to the costom_nodes one. Any fix?

@WildWorldTravelers 8 ай бұрын

when I try to copy the link on SSH appears a message that says: You don't have any public SSH keys in your GitHub account. You can add a new public key, or try cloning this repository via HTTPS. Im a little confuse im new in this, please help!

@LaCarnevali 8 ай бұрын

Might need to follow this: docs.github.com/en/authentication/connecting-to-github-with-ssh

@mojoejojo6675 11 ай бұрын

First.😊 still watching..

@RSV9 11 ай бұрын

Hello, is there a possibility to install and run Stable Diffusion SDXL 1.0 on a mac mini m2 pro with A1111 ? I can't use it on Windows because my graphics card only has 4 GB.

@LaCarnevali 11 ай бұрын

you can, but the performance will be very slow - you better go with Colab/runpod (comfyui colab is free)

@RSV9 11 ай бұрын

@@LaCarnevali It turns out that Stable Diffusion and SDXL can be run with ComfyUI on my Windows laptop with an RTX3050Ti with only 4GB of memory. Due to the low power of the card it is slow but it works although some said that at least 8GB is needed. I'll see how it works with Colab, thank you very much

@carlobrozzo5651 10 ай бұрын

@@LaCarnevali performance is the only issue on Mac (I'm running a M1 Max)? Seems you said you got troubles to install on Mac, do you plan a video for that in the future? Thanx for everything, your videos are so well done!

@orfeaus4240 11 ай бұрын

When I try to run ComfyUI on M2 Mac I get the error AttributeError: module 'torch' has no attribute 'mps'. I have PyTorch installed so I'm confused as to what's triggering this error.

@LaCarnevali 11 ай бұрын

Maybe this post can help? stackoverflow.com/questions/72535034/module-torch-has-no-attribute-has-mps

@WildWorldTravelers 8 ай бұрын

anyone can tell me if I can install this for Mac M1 32gb? and if i can do video2video? thank you!!

@LaCarnevali 8 ай бұрын

Mac will struggle - I would recommend Colab, RunPod, Diffusion Hub, Think Diffusion

@WildWorldTravelers 8 ай бұрын

Laura! Thank you for answering! After I installed local in my mac I finally decide to open in colab, I was able to open it in colab with some errors, should I need to delete all the previous folders that I had in my mac when I was running it local?

@LaCarnevali 8 ай бұрын

If you are not using it locally and you want to save space on your mac, then I would delete the folders :) @@WildWorldTravelers

@lozerkit4949 10 ай бұрын

make more vedio please..interested

@GB-- 11 ай бұрын

Good evening Laura, how are you? Could you please make a new installation video of "stable diffusion sdxl 1.0" on Macbook M1, please? T-T

@LaCarnevali 11 ай бұрын

I don't think you will be able to run it smoothly on M1. Have you tried ComfyUI on Colab (free) or A1111 on Colab/RunPod? It will make your SD experience nicer ;)

@AliKhan-vt3uk 10 ай бұрын

How to get consistent character in all images?

@LaCarnevali 10 ай бұрын

you can use the same seed to generate images, or you can try with adding 'character sheet' in the main prompt

@RyanSmith-vd7ue Ай бұрын

create loras

@manunoyonify 11 ай бұрын

Eh, un altro PS: maybe u can join Rivolt ? (change la première lettre par la quatrième de l'alphabet ou YT ne consent point....)

@wynard 11 ай бұрын

Not sure I like this UI so I'm going to wait. SD XL is a preview for now anyway...

@Luxcium 4 ай бұрын

I got banned from Midjourney (for not agreeing with the decision of the moderation team of interpreting the rules in a way that was not clearly stated in their TOS) I am now unable to use this workflow anymore and having a good computer with a a good but outdated GPU with 12 GB I realize that it would be nice if I could just leverage TPUs and GPUs for free or at a discount per second rate it would be great to increase my productivity and speed 😅😅😅😅