cf5682152cf6e34a30afe2646672494ec1010d5b
Diffusion-Model-Detailer
ComfyUI nodes that allow the user to control the generation of diffusion models to increase and decrease level of detail. The model patch has no trainable parameters and can be applied to theoretically any diffusion model in existence. whether or not the implementation currently works for every diffusion model in existence is unknown.
Installation
You can now install it with the ComfyUI manager. if you experience any issues please let me know.
Usage
Detail model wrapper node
- scaling factor: determines the strength and "direction" of the modification. positive values decrease the denoising model's agressiveness during affected steps and negative values increase it's aggressiveness at affected steps. this in turn means that positive values increase "detail" and negative values decrease "detail"
- scaling narrowness: this determines how "narrow" (or wide) the affected steps are. a value of 0 affects every single step at the strength of the scaling factor. higher values decrease the width (or increases the "narrowness") of the affected step area according to an exponentiated sine wave (i.e. the schedule is a sine wave raised to the power of "scaling narrowness")
Planned Features
-
Add a "depth" parameter to alter when the scheduled noise injection/removal peaks.
-
Migrate to using the proper model patcher comfy commands.
-
Make the nodes able to run inside a folder so comfy manager will be able to install the nodes automatically.
-
Add some example image grids
-
Add some example workflows
Languages
Python
100%