33 lines
1.4 KiB
Markdown
33 lines
1.4 KiB
Markdown
# SEGA for Anima
|
|
|
|
SEGA anima is implementation of ["SEGA: Spectral-Energy Guided Attention for Resolution Extrapolation in Diffusion Transformers" (arXiv:2605.22668)](https://arxiv.org/abs/2605.22668) for Anima model in ComfyUI.
|
|
|
|
## Installation
|
|
|
|
### Using git
|
|
|
|
From root folder of ComfyUI, goto `./ComfyUI/custom_nodes` and open terminal or command prompt and type following:
|
|
`git clone https://github.com/Filexor/SEGA_Anima.git`
|
|
|
|
### Via downloading zip
|
|
|
|
Click green "\<\> Code ▾" button and then click "Download ZIP" and extract folder to `./ComfyUI/custom_nodes` .
|
|
|
|
### Via ComfyUI Manager
|
|
|
|
1. Open ComfyUI Manager and click "Custom Nodes Manager".
|
|
2. Find "SEGA for Anima" and install it.
|
|
|
|
|
|
## Usage
|
|
|
|
You can find "SEGA for Anima" node in "Add node" -\> "model" -\> "patch" -\> "anima"
|
|
Insert "SEGA for Anima" node to between "Load Diffusion Model" node and "KSampler" node.
|
|
|
|
### What should I do when output image is bad
|
|
|
|
1. If image does not appear, first thing you have to do is decreasing "training_resolution". Too low "training_resolution" results structual inaccuracy.
|
|
2. Adjust CFG. Normally, higher CFG is needed for high resolution output.
|
|
3. You may adjust "theta", "base_mscale_coefficient", "mscale_alpha" or "d_mul".
|
|
4. Creator of the this node does not know what rest of parameters do. You may do experience with those.
|
|
5. If changing none of above work, Increasing steps may work. |