SDXL - Noise Injection: Can It Boost Image Quality?
Вставка
- Опубліковано 5 жов 2023
- In this video, we're diving into the noise injection method showcased by A Latent Place. We've also played around with the parameters, adding our little twist to it.
Feel free to download the workflow here: comfy.icu/c/u8zF8g and see how this technique can spice up your images.
For once youtube decided to recommend me some great hidden gems of AI channels, you're one of them. Thanks for the videos!
Subscribed, keep doing the hardwork! ♥
Nice video, thanks.
There is Ksampler (inspire), and it provides additional noise with variation strength. Good enough for variations
Isnt that the plasma one ? Ive tried a few and also some additional noise injection nodes all pretty useful depending on what you’re looking for.
You got the idea. :) Nice tutorial. I like the idea of combining more noises in a chain, didn't tried that yet. Looks like you can maybe get a bit of control on small and bigger details by doing this. And thanks for mentioning me. :)
You're very welcome! A shoutout was the least I could do , right? Btw, I'm currently experimenting with the following node : github.com/WASasquatch/PowerNoiseSuite , I'm pretty sure you'll find this also interesting.
@@tech_bytebrain Ah, didn't knew that one yet. Looks interesting. :O
@@tech_bytebrain Ohhh boy, can't wait for your tutorial on this ! :P
What is the rationale in enabling "add_noise" in KSampler and manually making additional noise "manually" with this (A Latent Place) intermediate step? As far as I understand this: you add just another layer of noise between the KSamplers while the KSampler does the same each time (or even each step?). Or am I wrong? Or is this noise from the image not random at all but reflects the growing image?
I've been banging my head on the wall trying to find an explanation for this, as disabling this option always breaks the results
Do you think that blurrying the background can help add more details to the foreground that is composed of a person ?
It can. Sometimes. Ive done some tests on doing some basic image filters and altering background with noise and contrast and blur/sharpen.
It absolutely has some interesting effects. Not always needed and increases render time. But yeah. It does.
The one thing I will say - if you inpaint adding noise has the effect of turbo charging your render.
As an example.
Take a black box. Just black image. Do a mask. And try to render a diamond in that mask. You’ll find you have to go up pretty high (about 0.7) until one starts to show up.
If you do the same thing with a noisy image. (Black with a bunch of grey noise) …. It’ll render around 0.5
Hi thanks for the video. I'm just trying to learn what i can about these nodes and how they fit together. Honestly i dunno wtf i'm doing with a 1000 parameters at every turn, these jerks are almost functioning like gatekeepers
You're welcome. Its always very useful to read the github documentation of the nodes to find out more.
Bleeding edge technologies be like that... but it's not gatekeeping, it's just being too technical and hardcore about it...
Eventually, we're gonna witness people like ByteBrain that can speak both human and technical at the same time. 🤣❤
@@6Saturn9 I know man.. I used to be a dev type like these fucking nerds who had to spend and inordinate amount of time documenting what they *might* have been thinking .. now i'm on a new hobby in retirement loving the bleeding edge but ffs
@meathunderify The pleasure of riding the bleeding edge, is paid by the pain of being your own tutor.
@@blacksage81 nicely put. definitely an exercise in humility
What is language you used in this video??
Where do you get image to noise?
you mean like encoding your image? vae-encode? or more something like the SamplerCustom node?
I mean "Image to noise node" @@tech_bytebrain
@@damrongtonklang9368 you can use the Unsampler or SamplerCustom node.
"This node does the reverse of a sampler. It calculates the noise that would generate the image given the model and the prompt. "
How make it for SDXL?
It is for sdxl
I liked this draw of network , pls need this app or language name
comfyUI