How to Install ComfyUI in 2023 - Ideal for SDXL!

In this ComfyUI tutorial I show how to install ComfyUI and use it to generate amazing AI generated images with SDXL! ComfyUI is especially useful for SDXL as poor old Automatic1111 can have a hard time using it - especially if you try to use the refiner! Also works great for Stable Diffusion 1.5!

It's really easy to install - especially with this video as a guide. It gives a level a freedom you've never had before, so why not give it a try?

Works on best on Linux but also works on MS Windows and even Mac (apparently)!

== Links! ==

== Increase your knowledge with Stable Diffusion Playlists! ==

#Sdxl #nocode #comfyui

[pedrogorilla]

It’s getting a lot of attention now after SDXL, hopefully there will be many improvements to the user experience in the coming months. It’s truly a powerful tool. Although I love A1111, it has grown into a software mess and feels like a lot of things are put together with duct tape. I think they’ll have to redo the entire backend from scratch at some point. It keeps getting harder to maintain and every new update breaks a lot of things.

[daniwanicki]

I enjoy the flexibility with ComfyUI. I can press one button and it will generate, refine, auto fix face and hands, remove background, then upscale all in one go. I can also queue hundreds of images in the background and not have to worry about my computer crashing while browsing the web. I'm impressed!

[omarbromar]

Had to re-install comfy, video was a pretty good refresher! thanks!

[Puckerization]

Learning ComfyUI just takes a little more perseverance than your normal attention deficit threshold. Once you pass that threshold, the basics all clicks into place. I like the fact that the images created are a record of the workflow and settings. Drag an image into the UI and it reproduces the node workflow with all the parameters.

[remco]

It's "modular" and that looks very promising to me, I love that the workflows are customizable

[Luxcium]

This is what I hope to have more of, I understand that you do have an awesome voice and charismatic accent, but I am also very curious about all there is to know with ComfyUI and all if you are willing to do more videos that explore this amazing UI

[ScottLahteine]

Among the many SDXL options available for Windows / macOS / Linux, ComfyUI is one the most optimal. At the moment (on macOS M1 Ultra) I’ve got ComfyUI, InvokeAI 2.3.5 and 3.0.1, Automatic1111 v1.5.1, Apple CoreML Diffusion, DiffusionBee, and MochiDiffusion. Automatic1111 is the fastest of the bunch with 3.4it/s for a 512x512 image, but ComfyUI is a close second with 3.0it/s. Both of them are pretty efficient with memory usage so they should work fine on more modest systems. We’re lucky to have so many great Stable Diffusion options now. What a time to be alive!

[agaviani]

we love you. Thank you for teaching sharing! Consistent style and consistent character in many different poses is the most interesting topic.

[kalisticmodiani]

Somebody should do the spaghetti and hide it in a nice interface. We'd call it Comfy1111

[Mimeniia]

Wow that Sytan workflow is insane. Thanks for vid, Nerdy.

[JanBadertscher]

You helped me to learn to use conda environments properly thanks to your first SD videos. Now I'd like to help you out too and recommend mamba to you! It's a one to one drop in replacement of conda, so you just subsitute all "conda" with "mamba" and have all the benefits: faster native C++ implementation of conda with parallel downloading, multi threading and faster dependency solving.

Honestly, just start using it and never look back at conda :)

[Kusari]

Much appreciated! I'm a complete newb and was able to follow this tutorial!

[notDreadful]

It does work indeed better, the speeds in ComfyUI are insane.
But that UI is too much for me and for the average SD enjoyer.
I'll probably just stick with 1.5 until Automatic1111 catches up.
Unfortunatly, on the other hand, I've personally seen the power of SDXL and I'm eager for things to move along quicker ahah

Amazing video Nerdy Rodent!

[xmattar]

Fun fact, I run SD 1.5 SD on 512 mb of vram

[Maria_Nette]

I'm quite comfy with the node base ui! Gotten used to it from doing a lot of work from Autodesk 3ds Max lol.

[brgtubedev]

Brilliant introduction to image gen in ComfyUI! Honestly, I've looked at all the other comparable videos for a starting point, and this is the most comprehensive for the installation and basic functions phase of learning. I've got the ComfyUI ControlNet tutorial open in my next tab.

On that note, can I offer some feedback regarding the in video link you placed at the end of the video that leads to the AudioGen AI video? My understanding is that YouTube allows linking to the next video, to encourage "bingeability", so users click through and keep watching once a video ends. Would it make more sense to link to another ImageGen tutorial after this one? It was sort of disjointed for me seeing the AudioGen link, because I wouldn't mind watching another tutorial on ComfyUI ImageGen, but I don't do anything with audio, and I found this video by searching for ImageGen tutorials, so why would I continue on to an AudioGen tutorial... My apologies if this comes off as being too blunt. I really like your channel and have been following for about half a year now, so I just thought I should share my thought process and search intent, since you are using the YouTube in-video link feature.

[wholeness]

What about control net and all the plugins automatic has ?

[AlexanderBukh]

figured how to use it a couple days ago but still, a very good video, thanks for making it.

one thing i wanted to do but saw it here first - present and save two images (or more) at once, say with or without the

[Puckerization]

ROOP works better in ComfyUI (make sure to separately install the restore faces nodes) you can add many ROOP nodes in a daisy chain and assign them to different figures...OR use multiple ROOPs of the same subject images (left, right older, younger) and assign them to the same figure and you'll get a mix with better details.

[bigbo]

I started by learning on ComfyUI, it was annoying at first, but the customization and resources are unmatched. The barrier to entry is also very low, although it is pretty easy to get carried away with upscaling and refining and make a workflow that takes forever to run.