ComfyUI - Node Based Stable Diffusion UI - THIS IS THE FUTURE!!!!!

ComfyUI is the Future of Stable Diffusion. With this Node Based UI you can use AI Image Generation Modular. ComfyUI gives you the full freedom and control to create anything you want. The ComfyUI Nodes support a wide range of AI Techniques like ControlNET, T2I, Lora, Img2Img, Inpainting, Outpainting. ComfyUI also allows you apply different prompt to different parts of your image or render images in multiple passes.

#### Win a RTX 3080Ti ####
Sign up FREE to Nivida GTC for a Chance to Win a RTX 3080Ti:

#### Links from the Video ####

00:00 Intro
01:53 Install ComfyUI
05:26 ComfyUI Examples
12:22 Node Basics
15:20 All Nodes

Support my Channel:

[OlivioSarikas]

#### Win a RTX 3080Ti ####
Sign up FREE to Nivida GTC for a Chance to Win a RTX 3080Ti:
#### Links from the Video ####

[augustday]

I am just infinitely grateful to the Stable Diffusion devs for making the original model and software open source, which has allowed such an incredible community to form around AI art. I hope someday that similar open source projects arise for other important tools such as text transformation and speech synthesis. The future of AI should be open source, I firmly believe that.

[AlistairKarim]

Because of its similarity to other node-based guis, it already looks like some sort of industry standard. Shader editors, blender, ue, nuke. Yes, you probably right, it's a future. At least for professionals in media. Compositing stuff is potentially awesome.

[synthoelectro]

Nodes are a pretty normal thing in the world of 3D rendering, such as Blender, music production, video editors, etc. This is very nice.

[SylvainSangla]

For those running out of space on their HDD or SSD, you can create symbolic links of your models instead of copying them, in order to use them on different UIs.

[jimdelsol]

I really like where this is going. Makes me think a lot about the nodes in blender. I was thinking to myself yesterday that i'd like to be able to have img2img with a "gradiant denoising" over the image, now this makes me think we could have regions with different denoise values painted with a mask node.

[RetzyWilliams]

Wow - I love using unreal engine's node system so I can see some really interesting possibilities with this. Thank you very much for taking us on the tour!

[Mimeniia]

This UI is a massive step up. Super complicated for the layman but immense control. Meta-data in the images...OMG...just drag the image into the UI and boom the whole workflow appears.
Batch size in Automatic1111 is amount of images it will generate all at once. Batch count is the amount of images it will generate one at a time. Seems like they confused it here or maybe it does generate, for example, 3 images at once.

[LeKhang]

Crazy. Absolutely CRAZY. It's incredible to see how much AI Art has evolved just in the last several months. I can't imagine what will it look like 5 years from now.

[JasonVailUXGuy]

Wow! This is fantastic. Coming from the visual effects world, this is absolutely a game changer and could easily setup automations and rich compositions.

[netandif]

Thanks for keeping us up to date with the latest innovations, as always. In this case i would say this is extraordinary news!
The node-based approach is pretty amazing. I know the reason why they call it "ComfyUI" 🥰

[azmodel]

This is CRAZY. I'm looking forward to watch more about this and how it works. Like a whole progress from 0, instead of a walkthrough of an already set one.

[nietzchan]

Looks pretty neat, like a Blender node workflow this is actually pretty intuitive
the best part is you can visualize and store your workflow in one page

for example you could have multiple Ksampler set up linked to same input so you could directly compare multiple output of different settings, or you can use it to compare different vae, checkpoints, negative prompts, etc. While you can already do that with automatic1111, to be able to visually see through multiple process and output in one page would make it easier in fine tuning the outputs

[SakuraSuzie]

You can use mklink to link to your existing models, embeddings, lora and vae

for example:

F:\ComfyUI\models>mklink /D checkpoints

F:\ComfyUI\models>mklink /D embeddings

F:\ComfyUI\models>mklink /D vae

F:\ComfyUI\models>mklink /D lora

You will need to run CMD as administrator.

[sathishkannan]

This would be a great editor when text to video comes in. It will definitely change the movie industry.

[FraYoshi]

This is just Crazy!!
You have definitely blowed my mind today!! 🤯

[USBEN.]

I love procedural workflows. This is very welcome.

[zimnelredoran]

This is big, reaaaally big. I'm eager to get my paws on it and give it a try! Thanks for the video, Olivio, always on the edge :D

[You-evkl]

I am forever grateful to how accessible this all is and how you don't have to be a computer wiz to use it. Great video as always Olivio.

[Sylfa]

Something that people seems to not know, which can be very helpful if you want to use/try multiple SD implementations, is that you don't have to copy the file. It's a little trickier than simply copy-paste, but all you have to do is run the command: mklink /H "New File" "Original File"

This creates a new file which shares the same file data as the original. Essentially this means that they look like normal files, and both say they are 4+ Gb, but they only actually take up that space once.

There's a downside, of course, but it's negligible in this case. If you edit the files content then you change all the copies at once. Also, it won't work across different hard drives. But they are safe to delete, the data remains until the last of them is deleted, they are also safe to rename and move. Copying will just give you a regular file copy though, so you need to run the command each time you want a new one.

You can do the same on Linux, but I don't know the command to use there, so I'll leave that up as an exercise to the reader.