[GH-ISSUE #14] Feature request: ComfyUI nodes

bsakaguchi commented

2026-10-31 21:01:32 +00:00

Owner

Originally created by @Jonseed on GitHub (Oct 10, 2024).
Original GitHub issue: https://github.com/muerrilla/sd-webui-detail-daemon/issues/14

It would be great if this could be used as a node in ComfyUI. I'm not sure if there is anything like it in ComfyUI. There are many nodes that say they can adjust the noise in your workflow, such as ComfyUI_Noise, but I'm not sure they do it like Detail Daemon.

Originally created by @Jonseed on GitHub (Oct 10, 2024). Original GitHub issue: https://github.com/muerrilla/sd-webui-detail-daemon/issues/14 It would be great if this could be used as a node in ComfyUI. I'm not sure if there is anything like it in ComfyUI. There are many nodes that say they can adjust the noise in your workflow, such as [ComfyUI_Noise](https://github.com/BlenderNeko/ComfyUI_Noise), but I'm not sure they do it like Detail Daemon.

bsakaguchi closed this issue

2026-10-31 21:01:32 +00:00

bsakaguchi commented

2026-10-31 21:01:33 +00:00

Author

Owner

@muerrilla commented on GitHub (Oct 10, 2024):

Gotta look for nodes that allow you to set your own sigmas or adjust the existing schedules. I'm pretty sure there are many out there.

@muerrilla commented on GitHub (Oct 10, 2024): Gotta look for nodes that allow you to set your own sigmas or adjust the existing schedules. I'm pretty sure there are many out there.

bsakaguchi commented

2026-10-31 21:01:33 +00:00

Author

Owner

@Jonseed commented on GitHub (Oct 13, 2024):

There is this node, CustomScheduler, that lets you set the sigmas at each step, but how would I use it to achieve what Detail Daemon does? Set each step sigma individually?

@Jonseed commented on GitHub (Oct 13, 2024): There is this node, [CustomScheduler](https://github.com/BlakeOne/ComfyUI-CustomScheduler), that lets you set the sigmas at each step, but how would I use it to achieve what Detail Daemon does? Set each step sigma individually?

bsakaguchi commented

2026-10-31 21:01:33 +00:00

Author

Owner

@Jonseed commented on GitHub (Oct 16, 2024):

I wonder if this CADS function is similar to Detail Daemon (also available as an extension in Auto1111). Or is that something completely different than Detail Daemon? It claims to increase diversity of outputs, not necessarily add detail.

@Jonseed commented on GitHub (Oct 16, 2024): I wonder if this [CADS function](https://github.com/asagi4/ComfyUI-CADS) is similar to Detail Daemon (also available as an [extension in Auto1111](https://github.com/v0xie/sd-webui-cads)). Or is that something completely different than Detail Daemon? It claims to increase diversity of outputs, not necessarily add detail.

bsakaguchi commented

2026-10-31 21:01:33 +00:00

Author

Owner

@blepping commented on GitHub (Oct 24, 2024):

ComfyUI users can give this small sampler wrapper a shot: https://gist.github.com/blepping/50103df387945b3fb2691a46812b4a64

put the file in custom_nodes, it will add a LyingSigmaSampler node. nothing special to use it:

the sigma just gets multiplied by 1 + dishonesty_factor so for example 1 + -0.05 is 0.95 - we reduced the sigma by 5%. you can also chain them if you want to use different factors at different points in sampling (negative values will add detail, positive will generally smooth things out). random example:

it should work with most samplers/models.

@blepping commented on GitHub (Oct 24, 2024): ComfyUI users can give this small sampler wrapper a shot: <https://gist.github.com/blepping/50103df387945b3fb2691a46812b4a64> put the file in `custom_nodes`, it will add a `LyingSigmaSampler` node. nothing special to use it: ![image](https://github.com/user-attachments/assets/a161dc17-4876-4647-bc8c-aab922905267) the sigma just gets multiplied by `1 + dishonesty_factor` so for example `1 + -0.05` is 0.95 - we reduced the sigma by 5%. you can also chain them if you want to use different factors at different points in sampling (negative values will add detail, positive will generally smooth things out). random example: ![image](https://github.com/user-attachments/assets/33f9a33b-43f5-4be6-8f80-c1b07497f3fc) it should work with most samplers/models.

bsakaguchi commented

2026-10-31 21:01:33 +00:00

Author

Owner

@muerrilla commented on GitHub (Oct 24, 2024):

I wonder if this CADS function is similar to Detail Daemon

It is not. As far as I understand CADS is degrading the conditioning tensor by adding noise to it at early sampling steps, which leads to more diverse output over different seeds. DD adjusts the sigmas which affect how much the image latent tensor is denoised at each step, affecting the amount of detail in the image.

@muerrilla commented on GitHub (Oct 24, 2024): > I wonder if this [CADS function](https://github.com/asagi4/ComfyUI-CADS) is similar to Detail Daemon It is not. As far as I understand CADS is degrading the **conditioning tensor** by adding noise to it at early sampling steps, which leads to more diverse output over different seeds. DD adjusts the sigmas which affect how much the **image latent tensor** is denoised at each step, affecting the amount of detail in the image.

bsakaguchi commented

2026-10-31 21:01:33 +00:00

Author

Owner

@muerrilla commented on GitHub (Oct 24, 2024):

ComfyUI users can give this small sampler wrapper a shot: https://gist.github.com/blepping/50103df387945b3fb2691a46812b4a64

This is the way! You can also do what Nerdy is doing here if you'd rather provide the list of sigmas than chain up those LyingSigmaSampler nodes.

P.S: Also what's up with the naming? Those poor manipulated sigmas are not lying, they're just different! 😆

@muerrilla commented on GitHub (Oct 24, 2024): > ComfyUI users can give this small sampler wrapper a shot: https://gist.github.com/blepping/50103df387945b3fb2691a46812b4a64 This is the way! You can also do what [Nerdy is doing here](https://www.youtube.com/watch?v=Sc6HbNjUlgI) if you'd rather provide the list of sigmas than chain up those LyingSigmaSampler nodes. P.S: Also what's up with the naming? Those poor manipulated sigmas are not lying, they're just different! 😆

bsakaguchi commented

2026-10-31 21:01:33 +00:00

Author

Owner

@blepping commented on GitHub (Oct 24, 2024):

You can also do what Nerdy is doing here if you'd rather provide the list of sigmas than chain up those LyingSigmaSampler nodes.

i don't think you can do it by just manipulating sigmas. doing something like multiplying the sigmas by 0.95 or whatever is super easy in ComfyUI but that also means you're reducing how much noise is added initially.

assuming i understood your code correctly, the key point is we add the initial noise and do the sampler steps based on the original sigmas but the sigma we call the model with is adjusted. so if sigma_max is 1.0 for example, we add initial noise at 1.0 strength, the first sampler step is from 1.0 to whatever the next step is but we call the model with sigma 0.95 instead of 1.0.

P.S: Also what's up with the naming? Those poor manipulated sigmas are not lying, they're just different!

i guess i'm the liar, not the sigmas. :) there's definitely some lying going on in my sampler at least!

@blepping commented on GitHub (Oct 24, 2024): > You can also do what [Nerdy is doing here](https://www.youtube.com/watch?v=Sc6HbNjUlgI) if you'd rather provide the list of sigmas than chain up those LyingSigmaSampler nodes. i don't think you can do it by just manipulating sigmas. doing something like multiplying the sigmas by 0.95 or whatever is super easy in ComfyUI but that also means you're reducing how much noise is added initially. assuming i understood your code correctly, the key point is we add the initial noise and do the sampler steps based on the original sigmas _but_ the sigma we call the model with is adjusted. so if `sigma_max` is `1.0` for example, we add initial noise at `1.0` strength, the first sampler step is from `1.0` to whatever the next step is but we call the model with sigma `0.95` instead of `1.0`. > P.S: Also what's up with the naming? Those poor manipulated sigmas are not lying, they're just different! i guess i'm the liar, not the sigmas. :) there's definitely some lying going on in my sampler at least!

bsakaguchi commented

2026-10-31 21:01:33 +00:00

Author

Owner

@muerrilla commented on GitHub (Oct 24, 2024):

i don't think you can do it by just manipulating sigmas. doing something like multiplying the sigmas by 0.95 or whatever is super easy in ComfyUI but that also means you're reducing how much noise is added initially.

Oh I see. That would indeed be different.

@muerrilla commented on GitHub (Oct 24, 2024): > i don't think you can do it by just manipulating sigmas. doing something like multiplying the sigmas by 0.95 or whatever is super easy in ComfyUI but that also means you're reducing how much noise is added initially. Oh I see. That would indeed be different.

bsakaguchi commented

2026-10-31 21:01:33 +00:00

Author

Owner

@Jonseed commented on GitHub (Oct 30, 2024):

With @blepping 's great help, there is now a proper port of Detail Daemon for ComfyUI:
https://github.com/Jonseed/ComfyUI-Detail-Daemon

Thank you @muerrilla for your great original work on this!

@Jonseed commented on GitHub (Oct 30, 2024): With @blepping 's great help, there is now a proper port of Detail Daemon for ComfyUI: https://github.com/Jonseed/ComfyUI-Detail-Daemon Thank you @muerrilla for your great original work on this!

bsakaguchi commented

2026-10-31 21:01:34 +00:00

Author

Owner

@muerrilla commented on GitHub (Oct 30, 2024):

With @blepping 's great help, there is now a proper port of Detail Daemon for ComfyUI: https://github.com/Jonseed/ComfyUI-Detail-Daemon

Thank you @muerrilla for your great original work on this!

Awesome! Glad you found it useful.

@muerrilla commented on GitHub (Oct 30, 2024): > With @blepping 's great help, there is now a proper port of Detail Daemon for ComfyUI: https://github.com/Jonseed/ComfyUI-Detail-Daemon > > Thank you @muerrilla for your great original work on this! Awesome! Glad you found it useful.

Rows
Columns

[GH-ISSUE #14] Feature request: ComfyUI nodes #9