Detail Daemon for ComfyUI with 8+GB VRAM

[@muerrilla]

yay! I'm somewhat featured on the Nerdy Rodent Show!🍾

[@NerdyRodent]

😃

[@kunalpuri9492]

Thanks for the good-humoured updates!

[@NerdyRodent]

And thank you for watching 😃

[@MikeHowles]

This is great. There goes my evening.

[@christofferbersau6929]

looks great, I wonder how it would work with tiled upscaling

[@stevenls9781]

Weird, I tried to use the Multiply Sigma with a Face generation and all I got was weird little blue and red objects in the background and not the skin details coming back in. Suffering through a very plastic looking output at the moment

[@TheSORCERER-p9l]

which one did you use for the woman with the red hair, that looks like the answer for Pulid plastic skin, just not sure what you did the achieve that.

[@PixelsVerwisselaar]

😱😱👌 nice. Thank you sir

[@dadadies]

How is it different from img2img? I see it is much better it seems but still if you know.

[@AorSoda]

Is there any way to make it compatible with the sd 3.5 model?

[@Jeremy-sk9qh]

Can you use the efficient nodes with this some how?

[@seyedamirhosseinKarrabi]

Hi! Thank you so much for the amazing content; I recently subscribed to your channel, and I’m loving everything you share! I have a question about one of your videos, specifically this one, where you teach how to repose any character in SD models. Is there any guide available for Flux? I’m curious if it's possible to do the same with Flux models. Thanks again!

[@skyworks6009]

Go Nerdy 👍

[@NerdyRodent]

😉

[@MissingModd]

My Nerdy Rodent is famous!

[@NerdyRodent]

😉

[@banzai316]

He’s our Neo!

[@kariannecrysler640]

🎃🎃🎃🎃🎃🎃🎃🎃🎃 My nerdy friend 🤘😉💗 Happy Halloween 🎃

[@NerdyRodent]

Happy Halloween!

[@murphylanga]

Hello, thank you very much for this information, something new that I didn't know before. I really enjoy watching your videos. Is it okay for you if I link one or two videos on my channel? Best regards, Murphy

[@NerdyRodent]

Hi, sure! Links are fine 😉

[@MilesBellas]

Batch processing option ?

[@NerdyRodent]

Latent batch or queue are two options

[@transfear]

Does it work with split sigmas & noise injection?

[@NerdyRodent]

Yup, though split sigmas changes where the middle is…

[@60FpsGoodness]

On the detail demon, you set it all the way up, for almost all the steps, thats the wrong way to go about it. You need to smooth the curve, and make sure you lower it way before reaching the last steps. Also 0.5 may be a bit much as well. At least for sdxl I find that 0.15 is already too much in most cases, depending how you set the bias and starting and end points, ofc. You could also play with the bias and starting point, for very different and amazing detailed results.

[@NerdyRodent]

I used the defaults to match the experience that everyone will have

[@ahsookee]

@@NerdyRodent i had different defaults

[@rishabhp1762]

Hey i cannot download the lying sigma its shows error

[@NerdyRodent]

I’m using Chrome

[@ai-centerd]

yeah

[@DanielPartzsch]

Thanks. So basically lying sigmas it is then if you don't want to be unpredictable things to happen...

[@LouisGedo]

👋 hi

[@NerdyRodent]

👋

[@joeyhandles]

Sigmas (stateless)

[@aestheticPhantasm]

"Day"mon

[@KINGLIFERISM]

half an hour. not worth it.