laksjdjf

'''
It supports only SD-v2 models.

usage:
python simo2kohya.py --unet <simo's unet weight path> --text <simo's text_encoder weight path> --save_to <save path>
(--text is optional)

This code may no longer be available due to updates from both @kohya-ss and @cloneofsimo.
'''

laksjdjf

#Put it on stable-diffusion-webui/scripts
#Idea:https://twitter.com/Birchlabs/status/1640033271512702977
#Related to https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/9129
import torch

import modules.scripts as scripts
import gradio as gr

from modules.script_callbacks import CFGDenoiserParams, on_cfg_denoiser,  CFGDenoisedParams, on_cfg_denoised
from modules.processing import StableDiffusionProcessing, process_images

laksjdjf

==============================================================================================================
Layer (type (var_name))                                      Input Shape               Output Shape
==============================================================================================================
UNetModel (UNetModel)                                        [1, 4, 64, 64]            [1, 4, 64, 64]
├─Sequential (time_embed)                                    [1, 320]                  [1, 1280]
│    └─Linear (0)                                            [1, 320]                  [1, 1280]
│    └─SiLU (1)                                              [1, 1280]                 [1, 1280]
│    └─Linear (2)                                            [1, 1280]                 [1, 1280]
├─ModuleList (input_blocks)                                  --                        --
│    └─TimestepEmbedSequential (0)                           [1, 4, 64, 64]            [1, 320, 64, 64]

laksjdjf

import torch

import modules.scripts as scripts
import gradio as gr
from modules.processing import StableDiffusionProcessing, process_images

class Script(scripts.Script):

    def __init__(self):
        pass

laksjdjf

import torch

import modules.scripts as scripts
import gradio as gr
from modules.processing import StableDiffusionProcessing, process_images

class Script(scripts.Script):

    def __init__(self):
        pass

laksjdjf

================================================================================================================================================================
Layer (type (var_name))                                      Input Shape               Output Shape              Param #                   Kernel Shape
================================================================================================================================================================
SdxlUNet2DConditionModel (SdxlUNet2DConditionModel)          [1, 4, 128, 128]          [1, 4, 128, 128]          --                        --
├─Sequential (time_embed)                                    [1, 320]                  [1, 1280]                 --                        --
│    └─Linear (0)                                            [1, 320]                  [1, 1280]                 410,880                   --
│    └─SiLU (1)                                              [1, 1280]                 [1, 1280]                 --         

laksjdjf

# python convert_lora_sdxl.py <input_file_name> <output_file_name> <sd2diff or diff2sd>

import torch
from safetensors.torch import load_file
from safetensors.torch import save_file
import os
import argparse

def load(file):
    if os.path.splitext(file)[1] == ".safetensors":

laksjdjf

import torch
import comfy
import copy

def chunk_or_none(x, chunk_size, index):
    if x is None:
        return None
    return x.chunk(chunk_size, dim=0)[index]

def chunk_or_none_for_control(x, chunk_size, index):

laksjdjf

'''
1. put this file in ComfyUI/custom_nodes
2. load node from <loader>
'''

import torch
from comfy.ldm.modules.diffusionmodules.openaimodel import forward_timestep_embed, timestep_embedding, th

# https://github.com/ChenyangSi/FreeU
def Fourier_filter(x, threshold, scale):

laksjdjf

import torch
import numpy as np 
from rembg import remove

class RembgMask:
    @classmethod
    def INPUT_TYPES(s):
        return {
            "required": {
                "image": ("IMAGE", ),