This is a merger node for Z-Image Turbo.
Unlike ComfyUI's standard layer merger, the numerical value indicates the strength of Model B (same as A1111 and SuperMerger).
While it doesn't consume much VRAM (around 10GB), 64GB of main memory may be recommended.
Usage:
Put the file in the custom node folder and restart. The node can be found in the menu: Add Node -> Model Utils -> Z-Image.
Noise Refiner Preprocesses the initial noise state to refine it in a more "pictorial" direction.
Context Refiner Preprocesses text and conditions to convert them into a "context that is easy to use for drawing."
Layer As with the Single block in Flux.1, this is where you actually draw the details.
resident_B Places Model B in VRAM (for those with ample VRAM).
Note: GGUF is not supported.
Description
FAQ
Comments (2)
what so how many of loras u can add to merge them?
is the affect of merging lot of loras not gonna vanish them? also is it gonna be working when the base and edit model will be released?
This tool can be used with LoRA, but is generally designed to handle ckpt models.
There is no limit to the amount of LoRA or ckpt that can be handled.
The merged model output can be saved using the ModelSave node and used as a standalone merged model. No matter how many times you merge, the effect will not disappear. Generally, however, changes made later seem to have a stronger effect.
Details of unreleased models such as the Base model are unknown, but they may not be usable as is. Generally, when a new model is released, you will need to adapt to the structure of that model.