ComfyUI Basics - Live Stream - Join me & Have Fun
HTML-код
- Опубликовано: 7 окт 2023
- Donate to my Live Stream: www.buymeacoffee.com/oliviotu...
#### Join and Support me ####
Buy me a Coffee: www.buymeacoffee.com/oliviotu...
Join my Facebook Group: / theairevolution
Joint my Discord Group: / discord 
Хобби
Everything I hear about the philosophy of comfyUI - just loading the stuff that's acutally needed, being able to quickly temporarily disable stuff, being able to load more than one model at the same time!!!!!!!!!!, the complete node graph being saved in the image, the ability to create something like an instant style lora by simply choosing a mix of images - makes so much sense. I am so happy you made this video, I will definitely try this out.
I just love that you are doing a livestream on ComfyUI, being able to chill and watch you build up elemetary configurations is my preferred form of learning, and it makes for the perfect thing to watch while taking a meal, "effortless learning" ♥♥♥
It seems that using the KSampler going into another KSample Latent_Image section with fixed seed, it seems to add just a bit more detail to the newer image.
BasingSoaked! That's a more appropriate pronunciation than the real one. 🤣
(FYI Olivio it is more like Bay zing stoak)
To copy a node quickly, you can hold down the left Alt key and drag the node.
Awesome tip, tks!
so I just...
...downloaded and installed ComfyUI (easiest install imaginable, just run the run file either for cpu or gpu),
...copied one automatic1111 checkpoint into the corresponding ComfyUI models folder (after being to weary of messing something up by changing the config file to link to my automatic1111 folders),
...build up the basic node setup for image generation by following a video by Scott Detweiler called "ComfyUI - Getting Started",
....entered my prompt and
....got my first image after what - maybe 15 minutes? (not counting the download time)
I am sure there will be cursing down the road, but so far, the experience was extremely smooth and chill.
The only two problems I ran into were
1. I can't zoom in and out with the scroll wheel since I'm using a stylus instead of a mouse, so I have to use keyboard (ctrl +, ctrl -) instead.
2. I did change the name of the config file into extra_model_paths.yaml, but then I wasn't sure what exact text I need to paste in where into the file, and if I have to copy the paths manually, since it obviously depends on where on my computer I installed Automatic1111 / Stable Diffusion.
This is fun!
Says me now, until the first time I'll be sweating about something that isn't working :) Although this happened with automatic1111, too, like when I'd get some "mysterious pair of numbers doesn't match other mysterious pair of numbers" message).
---> thanks for this livestream again, Olivio, can now follow along while I'll be watching the rest of it from coffee break to coffee break
solved the zooming issue
(if you have a wacom tablet, open the Wacom control panel, add your browser to the apps you make custom assignments for, and then choose Navigation / zoom for your upper switch click)
sheduler
I am very impressed that you took the time to do a comfy live stream to show how to use comfy.
You are fighting an uphill battle unfortunately.
You got 25 minutes into the stream and I was getting overwhelmed with everything on the screen.
What I would love to see is a video or a livestream about negative prompts.
I have an absolute uncontrollable hate for negative prompts that are a mile long.
Are they really needed or is a lot of it just fluff?
I am convinced that nearly every model has no idea exactly what "Masterpiece" means (I tested adding masterpiece to a prompt and the size of the file went down consistently). Same thing with "Bad quality" why would the AI create a bad quality image on purpose? does it even know what "bad quality" means?
I get wonderful images with very minimal negative prompts. Usually there are body deformity in the image because of the settings being used. ie. aspect ratios, too many weighted aspects, conflicting postive prompts etc. Deformity isn't something that the model does on purpose.
The Stable Diffusion guys say that (especially with SDXL) that negative prompts add (or fix) very little. Definitely try without them. For example though, 100% necessary when using something like StarryAI. Probably not when using SDXL.
ComfyUI is not ideal for people of normal intelligence
yeap just small pps big brains
i tried Fooocus and its pretty good for quick and dirty AI generation. but once you're used to comfyUI it really is so much better. definitely not easier, but better.
🤣🤣
Why?
Because the node graph quickly becomes so intimidatingly complex? I can definitely see this as a downside. In Blender, I tend to find them much easier to grasp which part does what when I build them myself, rather than being confronted with the whole complex end result at the end.
I prayed to the ai God's but they hate me.. 😂 got the error: insightface-0.7.3_cp310-win_amd64.whl is not a supported wheel on this platform.