ComfyUI Fundamentals - Mastering Noise

Рет қаралды 6,901

6 ай бұрын

In this video I cover my own observations while working with noise, how it affects the sampling process, looking at how different levels of denoising are affected by different kinds of images and how to build a module which can create useful images for use in the sampling process or as a post process effect.
This is a long one, sorry lol.
Discord: Join the community, friendly people, advice and even 1 on 1 tutoring is available.
/ discord
Workflow-
drive.google.com/file/d/19rjW...

Пікірлер: 41

@lex_darlog_fun 5 ай бұрын

In ComfyUI tutorials, there are mostly two extremes: "today we'll learn how to launch cmd to install Comfy, kids" and "here's a mind-blowing face-melting workflow I use" - and a dead-empty gap in between. Thank you VERY MUCH for filling this gap, your tutorials are a real treasure. It's SOOO HARD to find *ANY* info on "advanced fundamentals": i.e., basic stuff, but slightly above the very first few steps a novice would do. As an example, for MONTHS I'm struggling to understand the difference between various methods that conditioning (encoded prompts) can be combined with, and the lack of docs don't help at all. So @Ferniclestix, if you're unsure of what to tell next, a tutorial simply describing node after node, how to work with *ALL* the conditioning-related nodes - would be greatly appreciated. And, I believe, not only by me. This is one of the most confusing parts of Comfy, which (for some incomprehensible reason) is also one of the least documented ones.

@ferniclestix 5 ай бұрын

You are not wrong, in fact, I think most people don't really understand conditioning, the reasons for this being, its like a whole bunch of stuff from a text prompt to datapoints about an image. I struggle with it a fair bit, one thing I do know is that if you break something you end up learning more about it. so one major thing I've found is you can overload a conditioning line with too much information and it will throw errors lol. which basically means it works with floating point numbers and does math operations to them which can go out of bounds.

@stephantual 4 ай бұрын

this exactly. Ferniclestix and Matteo (Latentvision) are the best Comfytubers (is that a word) for that very reason. They're putting so much work in!

@aaagaming2023 5 ай бұрын

This is what I was looking for. In depth, coherent, relevant, and thorough. Thank you so much!

@dulow 5 ай бұрын

The GOAT. Love everything you put out. Thank you for doing these

@Ra-cd9op 5 ай бұрын

You deserve much more people find your videos. Super helpful to understand the tools correctly! thanks for sharing

@anthony_leckie 6 ай бұрын

Your tutorials are really helpful. Thank you. I got the impact nodes thanks to you and they've made a huge difference when I'm inpainting.

@ferniclestix 6 ай бұрын

great :D impacts great, wish it didn't want to use pipe for everything only downside lol.

@kobe5113 Ай бұрын

really awesome, thank you very much! much appreciated

@jalmarihann 2 ай бұрын

so much great stuff packed into one video, cheers!

@dougmaisner 6 ай бұрын

super helpful!

@visualfinesse7445 6 ай бұрын

Love these videos! Great job.

@WalidDingsdale 6 ай бұрын

your tutorials are wonderful, I learnt alot from them, your efforts are value to the community.

@ferniclestix 6 ай бұрын

:D try my best hopefully it helps people make cooler images :)

@akkitty22 6 ай бұрын

Insta subscribed! Great work, Sir.

@lex_darlog_fun 5 ай бұрын

21:35 What persistance in Perlin Noise means: The noise itself is made from the very simple pattern (set octaves to 0 or 1 - to see the "raw" pattern, it's basically a bunch of blurred blobs). But this pattern is applied multiple times, iteratively - each time, with smaller and smaller scale. (yep, perlin noise is a fractal) *Persistance* is basically a blend factor ("opacity" in Photoshop terms) used when the next octave is applied: it defines how much of the accumulated result is kept at each step. The higher it is, the more pronounced big clusters are (because higher levels contribute more). The lower it is, the more uniform noise you get (because lower octaves are more important than higher ones). The default 0.5 gives a "normalized" result: a middle ground between big "clouds" and small details on them.

@ferniclestix 5 ай бұрын

Thank you very much for the explanation :D

@miro2832 4 ай бұрын

Thank you very much for your videos!

@blacksage81 6 ай бұрын

Thank you, this video was super informative, and while I haven't ran it yet, the module looks clean and the outputs look interesting. Since I started organizing my workflows with modules, I've found that having a graph dedicated to building and storing modules for later use has come in clutch. I reverse engineered one of your workflows and pulled out two modules, the load batch image, its control, and the Blip captioner, and attached an Image Scale to side node, then I used the multiprompt area as a reference, ran the caption through a text concat, to duct tape it to the end of a prompt. I was surprised it worked, I just wish Blip captions were a little higher quality.

@ferniclestix 6 ай бұрын

hah NICE! yeah, blip was a bit of an experiment, unfortunatley it doesnt really provide a whole bunch of words for prompting. I think it was... FABRIC ? you might want to check that out, from memory it lets you mix one image into another in cool ways so it basically turns an image into conditioning. eh, been a while since I looked at it but you might find it interesting.

@FreeEasyAI 5 ай бұрын

Great job, thanks!

@unforkableonion5081 6 ай бұрын

hey there, binge watched all your content which i find very informative! i believe you would absolutely enjoy the rgthree nodes, in particular the context one. it has some cool auto wiring tricks, also it has reroutes that can do what you expect ;)

@ferniclestix 6 ай бұрын

yah :D the only issue I have with rgeethree is how slow it is to actually use the nodes, needs its own UI menu tbh. I don't generally use them much because i like to use the base setup to help new users learn with something familiar if possible and they won't have all the custom nodes so i leave that up to you guys to decide on mostly :D

@blisterfingers8169 6 ай бұрын

Some fun techniques here, thanks. How do you get those previews in the kSamplers?

@ferniclestix 6 ай бұрын

there is an option in the comfyUI manager to turn that on, alternatively you can add it in the comfyUI batch file.

@blisterfingers8169 6 ай бұрын

@@ferniclestix Ooooh, found it! Thank you. Having some fun with all sorts of weird inputs. Halftone's a good one if you like fences. ♥

@davidb8057 6 ай бұрын

Thanks a lot for these tutorials, they are helping me a lot. It's interesting to observe that many nodes that are coming out for Comfy are very close to basic nodes used in "classic" VFX compositing softwares (like Nuke, Fusion or Shake...), nodes about mixing pixel values from two images through different math ops, masks nodes etc. By the way, do you happen to now, in Comfy, at which bit depth is it operating? At 8 bits, every RGB pixel have a range of 256 values (from 0 to 255), at 10 bits it would be 1024, and in the compositing world, we use to work at 32 bits (values are almost infinite and can even be negative). This concept could become relevant when you start to add/subtract/multiply pixel values...

@ferniclestix 6 ай бұрын

depends on the save out node you use, you can output tiff files and someone was messing with it at 32bits a while back, but like render times were massive so there was limited value up that high.

@jmoneydroid 6 ай бұрын

does the "add noise" enabled on the ksampler cause the unwanted effects around 16 minute mark?

@ferniclestix 6 ай бұрын

no, its on because thats how the normal samplers actually behave. it looks terrible because anything between steps on samplers generally looks really bad/noisy.

@user-yn1ir6hf4d 6 ай бұрын

Hey nice work! Just want to ask that are you interested in our new ai generation product? Maybe I can send you beta version to you~

@LordOfThunderUK 6 ай бұрын

Hi, I am trying to make a satiric graphic about the bible character Jonah, I would like to depict him inside the whales stomach, whilst being inside a jacuzzi and eating a prawn cocktail.

@ferniclestix 6 ай бұрын

hah!, ive had real difficulty getting 'inside a stomach' images, you could probably do it using a radial gradient with red and pink tho if you wanted to give it a shot :P

@LordOfThunderUK 6 ай бұрын

@@ferniclestix First, thank you for your response, I am not that savy with ComfyUI but I will view your videos so I can get acquainted with the program.