↓cattowerV15vpred_cyber4itercomppvc.safetensors
↓cattowerV16vpred_cyber4itercomppvc.safetensors
↓cattowerV16vpred_cyber4itercomppvcrev2.safetensors
↓cattowerV16vpred_cyber5itercomppvc.safetensors
↓cattowerV17vpred_cyberill3itercomppvc.safetensors
Description:
A merge model of those models; cyber realistic xl v4/5, itercomp v1, movable figure model illustrious xl v1 and catTower v1.5(vpred)/v1.6(vpred).
This model require V-pred supported client; like as a1111dev, forge, reforge, comfyui.
The recipe is included in sample images.
Entry settings:
Euler, 24 steps, CFG 3-5
PP: masterpiece, newest,
NP: worst quality, comic, multiple views, bad quality, low quality, lowres, displeasing, very displeasing, bad anatomy, bad hands, scan artifacts, monochrome, greyscale, twitter username, jpeg artifacts, 2koma, 4koma, guro, extra digits, fewer digits, jaggy lines, unclear, oldest
This model has wide range CFG tolerance. If you want more NP effect or off-white and vividness, increase Steps and CFG. up to around 7-9, there are few side effects. v1.6_rev2 is reduced version grayish white background issue in low CFG.
Note:
To set the model type as V-pred automatically, After model export, the keys "v_pred" and "ztsnr" were written into the state_dict of the model using this python function.
import torch
from safetensors import safe_open
from safetensors.torch import save_file
def append_vpred_keys(input_filename, output_filename):
with safe_open(input_filename, framework="pt", device="cpu") as f:
state_dict = {key: f.get_tensor(key) for key in f.keys()}
state_dict["v_pred"] = torch.tensor([0])
state_dict["ztsnr"] = torch.tensor([0])
save_file(state_dict, output_filename)