# -*- coding: utf-8 -*-
##--------------------------------------------------
## Stable Diffusion with diffusers(092) Ver 0.00
## img2img GUI interface
## 2025.07.03 Masahiro Izutsu
##--------------------------------------------------
## sd_092.py
## Ver 0.00 2025.07.03 GUI 対応版
## Ver 0.01 2025.07.13 汎用コードを sd_tools に分離
# タイトル
title = 'Stable Diffusion with diffusers(092) Ver 0.01'
import warnings
warnings.simplefilter('ignore')
# インポート＆初期設定
import os
import numpy as np
#import random
import time
import csv
import cv2
import PySimpleGUI as sg
import sd_091 as sd
import my_logging
import my_csv
import my_thumbnail
import my_dialog
import my_imagetool
import sd_tools as sdt
# 定数定義
DEF_THEME = 'BlueMono'
CANVAS_SIZE = 512
KEY_CANCEL = '-Cancel-'
KEY_IMAGE = '-Image-'
KEY_INPUT_IMAGE = '-Input_Image-'
KEY_INPUTPATH = '-Input-'
KEY_OUTPATH = '-Output-'
KEY_MODEL = '-Model-'
KEY_IMAGE_SEL = '-Image_sel-'
KEY_MODELSEL = '-Model_sel-'
KEY_PROMPT_JP = '-PromptJP-'
KEY_PROMPT = '-Prompt-'
KEY_MAXSIZE = '-MaxSize-'
KEY_SEED_INPUT = '-Seed_input-'
KEY_SEED = '-Seed-'
KEY_STEP = '-Step-'
KEY_SCALE = '-Scale-'
KEY_STRENGTH = '-strength-'
KEY_CPU = '-CPU-'
KEY_GPU = '-GPU-'
KEY_DEVICE = '-Device-'
KEY_EXIT = '-Exit-'
KEY_GENERATE = '-Generate-'
KEY_LOOP = '-Loop-'
SKEY_RESULT_PATH = '-result_path-'
SKEY_RESULT_FILE = '-result_file-'
SKEY_MODEL_DIR = '-model_dir-'
SKEY_DEF_OUTPATH = '-default_image-'
KEY_TIME = '-Time-'
DEF_IMAGE = './sd_results/sd_00000_12345678.png'
DEF_INPUT_IMAGE = './images/kaisendon.jpg'
CSV_LOG_FILE = 'result.csv'
# ウィジェットのデータの取得（1回目）
def get_paramlist(window, values, param):
param[KEY_MODEL] = window[KEY_MODEL].DisplayText
param[KEY_MAXSIZE] = int(values[KEY_MAXSIZE])
param[KEY_SEED] = sdt.get_random_seed_value(values[KEY_SEED_INPUT])
param[KEY_STEP] = int(values[KEY_STEP])
param[KEY_SCALE] = float(values[KEY_SCALE])
param[KEY_STRENGTH] = float(values[KEY_STRENGTH])
param[KEY_DEVICE] = 'cpu' if values[KEY_CPU] == True else 'cuda'
param[KEY_PROMPT_JP] = values[KEY_PROMPT_JP]
param[KEY_PROMPT] = sdt.trans_jp2en(param[KEY_PROMPT_JP])
param[KEY_LOOP] = int(values[KEY_LOOP])
if param[KEY_LOOP] < 1:
param[KEY_LOOP] = 1
if param[KEY_LOOP] > 50:
param[KEY_LOOP] = 50
# ウィジェットのデータの取得（2回目以降）
def get_paramlist2(window, values, param):
if int(values[KEY_SEED_INPUT]) > 0:
param[KEY_SEED] = param[KEY_SEED] + 1
else:
param[KEY_SEED] = sdt.get_random_seed_value(values[KEY_SEED_INPUT])
# 結果を CSV ファイルに残す
def result_csv(csvfile, param, logger):
my_csv_treatment = my_csv.CSVtreatment(csvfile, 'utf_8_sig')
s = my_csv_treatment.read_csv()
if len(s) ==0:
data = []
data.append(KEY_OUTPATH)
data.append(KEY_SEED)
data.append(KEY_PROMPT_JP)
data.append(KEY_PROMPT)
data.append(KEY_INPUTPATH)
data.append(KEY_MAXSIZE)
data.append(KEY_STEP)
data.append(KEY_SCALE)
data.append(KEY_STRENGTH)
data.append(SKEY_MODEL_DIR)
data.append(KEY_MODEL)
data.append(KEY_DEVICE)
data.append(KEY_TIME)
data.append(KEY_LOOP)
my_csv_treatment.write_csv(data)
logger.debug(data)
data = []
data.append(param[KEY_OUTPATH])
data.append(str(param[KEY_SEED]))
data.append(param[KEY_PROMPT_JP])
data.append(param[KEY_PROMPT])
data.append(param[KEY_INPUTPATH])
data.append(str(param[KEY_MAXSIZE]))
data.append(str(param[KEY_STEP]))
data.append(str(param[KEY_SCALE]))
data.append(str(param[KEY_STRENGTH]))
data.append(param[SKEY_MODEL_DIR])
data.append(param[KEY_MODEL])
data.append(param[KEY_DEVICE])
data.append(param[KEY_TIME])
data.append(param[KEY_LOOP])
my_csv_treatment.append_csv(data)
logger.debug(data)
# 結果ログからパラメータを取得
def read_result_csv(csvfile, param, logger, imgfile = '0' ):
bf = False
try:
with open(csvfile, 'r', encoding = 'utf_8_sig') as f:
reader = csv.DictReader(f)
if imgfile == '0': # 最初の行
for dd in reader:
bf = True
break
elif imgfile == '1': # 最後の行
for dd in reader:
bf = True
continue
else: # 出力ファイルの行
for dd in reader:
if dd[KEY_OUTPATH] == imgfile:
bf = True
break
except FileNotFoundError as e:
pass
except csv.Error as e:
logger.debug(e)
if bf:
param[KEY_OUTPATH] = dd[KEY_OUTPATH]
param[KEY_SEED_INPUT] = int(dd[KEY_SEED]) # シード値を固定
param[KEY_PROMPT_JP] = dd[KEY_PROMPT_JP]
param[KEY_PROMPT] = dd[KEY_PROMPT]
param[KEY_INPUTPATH] = dd[KEY_INPUTPATH]
param[KEY_MAXSIZE] = int(dd[KEY_MAXSIZE])
param[KEY_STEP] = int(dd[KEY_STEP])
param[KEY_SCALE] = float(dd[KEY_SCALE])
param[KEY_STRENGTH] = float(dd[KEY_STRENGTH])
param[KEY_MODEL] = dd[KEY_MODEL]
logout_data(dd, logger)
elif os.path.isfile(imgfile):
seed = path2seed(imgfile) # ファイル名に含まれるシード値
param[KEY_OUTPATH] = dd[KEY_OUTPATH] = imgfile
param[KEY_SEED_INPUT] = dd[KEY_SEED] = seed
logout_data(dd, logger)
bf = True
return bf
# 動的パラメータの出力
def logout_data(dd, logger):
logger.debug('---------------------------------------------')
logger.debug(dd[KEY_OUTPATH])
logger.debug(int(dd[KEY_SEED]))
logger.debug(dd[KEY_PROMPT_JP])
logger.debug(dd[KEY_PROMPT])
logger.debug(dd[KEY_INPUTPATH])
logger.debug(int(dd[KEY_MAXSIZE]))
logger.debug(int(dd[KEY_STEP]))
logger.debug(float(dd[KEY_SCALE]))
logger.debug(float(dd[KEY_STRENGTH]))
logger.debug(dd[SKEY_MODEL_DIR])
logger.debug(dd[KEY_MODEL])
logger.debug(dd[KEY_DEVICE])
logger.debug('---------------------------------------------')
# 画像生成
def generate_image(param):
model = param[SKEY_MODEL_DIR] + '/' + param[KEY_MODEL]
image_path = param[KEY_INPUTPATH]
prompt = param[KEY_PROMPT]
max_size = param[KEY_MAXSIZE]
seed = param[KEY_SEED]
num_inference_steps = param[KEY_STEP]
guidance_scale = param[KEY_SCALE]
strength = param[KEY_STRENGTH]
device = param[KEY_DEVICE]
out_path = param[SKEY_RESULT_PATH] + '/' + sdt.make_filename_by_seq(param[SKEY_RESULT_PATH], param[SKEY_RESULT_FILE], seq_digit = 5, ex = seed)
param[KEY_OUTPATH] = out_path
logger.debug(f'model: {model｝')
logger.debug(f'prompt: {prompt｝')
logger.debug(f'image_path: {image_path｝')
logger.debug(f'max_size: {max_size｝')
logger.debug(f'seed: {seed｝')
logger.debug(f'num_inference_steps: {num_inference_steps｝')
logger.debug(f'guidance_scale: {guidance_scale｝')
logger.debug(f'strength: {strength｝')
logger.debug(f'device: {device｝')
logger.debug(f'loop: {param[KEY_LOOP]｝')
logger.debug(f'out_path: {out_path｝')
image = sdt._get_resize_image(image_path, max_size, logger) # 元画像リサイズ
image = sd.image_generation(model, prompt, image, seed, num_inference_steps, guidance_scale, strength, device)
image.save(out_path)
logger.info(f'result_file: {out_path｝')
sd.device_empty_cache(device) # メモリー開放
# ** main関数 **
def main(opt, logger):
# ------------------------------------------
# キャンバスをクリア
def clear_canvas(key, msg, color):
frame = np.zeros((CANVAS_SIZE, CANVAS_SIZE, 3), np.uint8)
frame[:,:,] = 0xf0
return msg_out_canvas(key, frame, msg, color)
def msg_out_canvas(key, frame, msg, color):
x0,y0,x1,y1 = cv2_putText(img=frame, text=msg, org=(CANVAS_SIZE//2, CANVAS_SIZE//2), fontFace=font_face, fontScale=16, color=color, mode=2,areaf=True)
cv2.rectangle(frame,(x0-8, y0), (x1+8, y1), (0xf0,0xf0,0xf0), -1)
img = cv2_putText(img=frame, text=msg, org=(CANVAS_SIZE//2, CANVAS_SIZE//2), fontFace=font_face, fontScale=16, color=color, mode=2)
img = cv2.imencode('.png', frame)[1].tobytes()
window[key].update(img)
return frame
# キャンバスへ画像の表示
def update_canvas(key, imgfile):
if os.path.isfile(imgfile):
frame = cv2.imread(imgfile)
frame = my_imagetool.frame_square(frame, (240, 240, 240))
frame = cv2.resize(frame, dsize = (CANVAS_SIZE, CANVAS_SIZE))
img = cv2.imencode('.png', frame)[1].tobytes()
window[key].update(img)
else:
frame = clear_canvas(key, 'Generate Image', (0,0,0))
return frame
# ウイジェットの更新
def update_widget():
frame = update_canvas(KEY_IMAGE, param[KEY_OUTPATH])
frame_input = update_canvas(KEY_INPUT_IMAGE, param[KEY_INPUTPATH])
window[KEY_OUTPATH].update(param[KEY_OUTPATH])
window[KEY_SEED].update(param[KEY_SEED_INPUT]) # シード値の表示
window[KEY_PROMPT_JP].update(param[KEY_PROMPT_JP])
window[KEY_PROMPT].update(param[KEY_PROMPT])
window[KEY_INPUTPATH].update(param[KEY_INPUTPATH])
window[KEY_MODEL].update(param[KEY_MODEL])
window[KEY_MAXSIZE].update(param[KEY_MAXSIZE])
window[KEY_STEP].update(param[KEY_STEP])
window[KEY_SCALE].update(param[KEY_SCALE])
window[KEY_STRENGTH].update(param[KEY_STRENGTH])
window[KEY_CPU].update(param[KEY_DEVICE] == 'cpu')
window[KEY_GPU].update(param[KEY_DEVICE] == 'cuda')
return frame
# ウイジェットの禁止・許可
def set_enb_dis(disabled):
window[KEY_IMAGE_SEL].update(disabled = disabled)
window[KEY_MODELSEL].update(disabled = disabled)
window[KEY_GENERATE].update(disabled = disabled)
window[KEY_LOOP].update(disabled = disabled)
window[KEY_EXIT].update(disabled = disabled)
# ------------------------------------------
# パラメータ設定
device = sdt._get_device(opt, logger)
result_path = sdt._get_result_path(opt, logger)
result_file = sdt._get_result_file(opt, logger)
image_path = sdt._get_source_image_path(opt, logger)
prompt = sdt._get_prompt(opt, logger)
model_path = sdt._get_model_path(opt, logger)
max_size = opt.max_size
seed = sdt.path2seed(DEF_IMAGE) # 初期ファイル名に含まれるシード値
num_inference_steps = sdt._get_inference_steps(opt, logger)
guidance_scale = sdt._get_guidance_scale(opt, logger)
strength = sdt._get_strength(opt, logger)
param = {｝
param[KEY_OUTPATH] = ''
param[KEY_PROMPT_JP] = opt.prompt
param[KEY_PROMPT] = prompt
param[KEY_INPUTPATH] = image_path
param[KEY_MODEL] = opt.model_path
param[KEY_MAXSIZE] = max_size
param[KEY_SEED_INPUT] = opt.seed
param[KEY_SEED] = seed
param[KEY_STEP] = num_inference_steps
param[KEY_SCALE] = guidance_scale
param[KEY_STRENGTH] = strength
param[KEY_DEVICE] = device
param[KEY_LOOP] = 1
param[KEY_TIME] = sdt.elapsed_time_str(0)
param[SKEY_RESULT_PATH] = result_path
param[SKEY_RESULT_FILE] = result_file
param[SKEY_MODEL_DIR] = opt.model_dir
logout_data(param, logger)
csvfile = param[SKEY_RESULT_PATH]+ '/' + CSV_LOG_FILE # 出力画像ログファイル名
# 出力フォルダ
os.makedirs(result_path, exist_ok = True)
# フォント取得
from my_puttext import get_font, cv2_putText
font_face = get_font()
# ウィンドウのテーマ
sg.theme(DEF_THEME)
canvas_input_img = sg.Image(size = (CANVAS_SIZE, CANVAS_SIZE), key=KEY_INPUT_IMAGE)
canvas_img = sg.Image(size = (CANVAS_SIZE, CANVAS_SIZE), key=KEY_IMAGE)
col_left = [
[canvas_input_img],
[sg.Text("Input File", size=(14, 1)), sg.Text(param[KEY_INPUTPATH], size=(38,1), text_color='#008800', background_color='LightSteelBlue1', key=KEY_INPUTPATH), sg.Button('Image', size=(6, 1), key=KEY_IMAGE_SEL)],
[sg.Text("Model", size=(14, 1)), sg.Text(param[KEY_MODEL], size=(38,1), text_color='#008800', background_color='LightSteelBlue1', key=KEY_MODEL), sg.Button('Model', size=(6, 1), key=KEY_MODELSEL)],
[sg.Text("Prompt input", size=(14, 1)), sg.Multiline(param[KEY_PROMPT_JP], size=(52,4), key=KEY_PROMPT_JP)],
[sg.Text("Prompt", size=(14, 1)), sg.Multiline(param[KEY_PROMPT], size=(52,4), text_color='#008800', background_color='LightSteelBlue1', key=KEY_PROMPT)],
]
col_right = [
[canvas_img],
[sg.Text("Output File", size=(14, 1)), sg.Text(param[KEY_OUTPATH], size=(48,1), key=KEY_OUTPATH)],
[sg.Text("Max size (0= full)", size=(14, 1)), sg.Input(param[KEY_MAXSIZE], size=(10,1), key=KEY_MAXSIZE)],
[sg.Text("Seed (-1=Random)", size=(14, 1)), sg.Input(param[KEY_SEED_INPUT], size=(20,1), key=KEY_SEED_INPUT), sg.Text(param[KEY_SEED], size=(20,1), text_color='#008800', background_color='LightSteelBlue1', key=KEY_SEED)],
[sg.Text("Detail (Steps)", size=(14, 1)), sg.Slider((10, 150), float(param[KEY_STEP]), 1, orientation='h', size=(42, 5), key=KEY_STEP)],
[sg.Text("Guidance Scale", size=(14, 1)), sg.Slider((1, 50), float(param[KEY_SCALE]), 0.1, orientation='h', size=(42, 5), key=KEY_SCALE)],
[sg.Text("Strength", size=(14, 1)), sg.Slider((0, 1), float(param[KEY_STRENGTH]), 0.1, orientation='h', size=(42, 5), key=KEY_STRENGTH)],
[sg.Text("Device", size=(14, 1)), sg.Radio('CPU', group_id='device', default=(device == 'cpu'), key=KEY_CPU), sg.Radio("GPU", group_id='device', default=(device == 'cuda'), key=KEY_GPU)],
[sg.Text("Loop count (1-50)", size=(14, 1)), sg.Input(param[KEY_LOOP], size=(2,1), justification='right', key=KEY_LOOP), sg.Text("", size=(4, 1)), sg.Button('Generate', size=(10, 1), key=KEY_GENERATE), sg.Text("", size=(10, 1)), sg.Button('Exit', size=(10, 1), key=KEY_EXIT)]
]
# ウィンドウのレイアウト
layout = [[sg.Column(col_left, vertical_alignment='top'), sg.Column(col_right, vertical_alignment='top')]]
# ウィンドウオブジェクトの作成
window = sg.Window(title, layout, finalize=True, return_keyboard_events=True)
# ユーザーイベントの定義
canvas_img.bind('<ButtonPress>', '_click_on')
# キャンバス初期化
ss = param[KEY_SEED_INPUT] # 最初のシード入力値
bf = read_result_csv(csvfile, param, logger, '1')
if bf:
frame = update_widget()
param[KEY_SEED] = param[KEY_SEED_INPUT]
param[KEY_SEED_INPUT] = ss
window[KEY_SEED].update(param[KEY_SEED])
window[KEY_SEED_INPUT].update(param[KEY_SEED_INPUT] )
else:
frame = update_canvas(KEY_IMAGE, param[KEY_OUTPATH])
frame_input = update_canvas(KEY_INPUT_IMAGE, param[KEY_INPUTPATH])
param[SKEY_DEF_OUTPATH] = param[KEY_OUTPATH] # 現在の画像パス
new_make_f = False
window[KEY_PROMPT].update(disabled = True)
# イベントのループ
while True:
event, values = window.read(timeout = 30)
# 画像生成
if new_make_f:
# 処理プロセス
logger.info(f'{sdt.CYAN｝** Start {param[KEY_LOOP]｝ **{sdt.NOCOLOR｝')
start_time = time.time()
generate_image(param)
param[KEY_TIME] = sdt.elapsed_time_str(time.time() - start_time)
result_csv(csvfile, param, logger)
frame = update_canvas(KEY_IMAGE, param[KEY_OUTPATH])
param[SKEY_DEF_OUTPATH] = param[KEY_OUTPATH] # 現在の画像パス
window[KEY_OUTPATH].update(param[KEY_OUTPATH])
param[KEY_LOOP] = param[KEY_LOOP] - 1
window[KEY_LOOP].update(param[KEY_LOOP])
if param[KEY_LOOP] < 1:
param[KEY_LOOP] = 1
window[KEY_LOOP].update(param[KEY_LOOP])
set_enb_dis(False)
new_make_f = False
else:
get_paramlist2(window, values, param)
window[KEY_SEED].update(param[KEY_SEED])
logger.info(f'{sdt.CYAN｝** Complete **{sdt.NOCOLOR｝ {param[KEY_TIME]｝')
# 終了
if event == KEY_EXIT or event == sg.WIN_CLOSED:
break
# KEY_IMAGE
if event == KEY_IMAGE + '_click_on':
logger.debug(f'{event｝')
set_enb_dis(True)
def_file = param[SKEY_DEF_OUTPATH]
if os.path.isfile(def_file):
imgfile = my_thumbnail.image_dialog(def_file, 'Image file select', my_thumbnail.DEF_THEME, 10, 4, ret = '', logger = logger)
if os.path.isfile(imgfile):
logger.debug(f'Output select: {imgfile｝')
bf = read_result_csv(csvfile, param, logger, imgfile)
if bf:
frame = update_widget()
param[SKEY_DEF_OUTPATH] = param[KEY_OUTPATH]
set_enb_dis(False)
# 画像選択ボタン
if event == KEY_IMAGE_SEL:
logger.debug(f'{event｝')
set_enb_dis(True)
filename = my_dialog.select_image_file(initdir=os.path.dirname(param[KEY_INPUTPATH]))
if len(filename) > 0:
param[KEY_INPUTPATH] = filename
frame_input = update_canvas(KEY_INPUT_IMAGE, param[KEY_INPUTPATH])
window[KEY_INPUTPATH].update(param[KEY_INPUTPATH] )
set_enb_dis(False)
# Model ボタン
if event == KEY_MODELSEL:
logger.debug(f'{event｝')
set_enb_dis(True)
s0 = param[KEY_MODEL]
id = s0.find('/')
s1 = '/' + s0[:id] if id > 0 else ''
s2 = param[SKEY_MODEL_DIR] + s1
ttl = 'モデルファイルを選択'
types = [("Model file", ".safetensors .pt"), ("model", ".safetensors"), ("pt file", ".pt") ]
filename = my_dialog.file_dialog(ttl, types, s2)
if len(filename) > 0:
s3 = os.path.basename(filename) # ファイル名
s4 = os.path.dirname(filename) # ディレクトリ名
id = s4.find('SD1.5')
if id >= 0 and id == len(s4) - 5: # SD1.5
s5 = s4[:id - 1]
s3 = 'SD1.5/' + s3
else: # SDXL
s5 = s4
param[KEY_MODEL] = s3
s6 = os.path.abspath(param[SKEY_MODEL_DIR])
s7 = s6.replace('\\', '/')
if s7 == s5: # モデルディレクトリは変更不可
window[KEY_MODEL].update(param[KEY_MODEL])
logger.debug(f'{param[SKEY_MODEL_DIR]｝/{param[KEY_MODEL]｝')
else:
my_dialog.warning_dialog(ttl, 'モデルディレクトリは変更できません')
set_enb_dis(False)
# Genarate ボタン
if event == KEY_GENERATE:
logger.debug(f'{event｝')
set_enb_dis(True)
get_paramlist(window, values, param)
frame = msg_out_canvas(KEY_IMAGE, frame, 'Generating ...', (240,0,0))
window[KEY_OUTPATH].update('')
window[KEY_PROMPT].update(param[KEY_PROMPT])
window[KEY_SEED].update(param[KEY_SEED])
new_make_f = True
# ウィンドウ終了処理
window.close()
# main関数エントリーポイント(実行開始)
if __name__ == "__main__":
parser = sdt.parse_args(None, sd.opt_list)
opt = parser.parse_args()
# アプリケーション・ログ設定
module = os.path.basename(__file__)
module_name = os.path.splitext(module)[0]
logger = my_logging.get_module_logger_sel(module_name, int(opt.log))
sdt._get_device(opt)
sdt.display_info(opt, title)
main(opt, logger)
logger.info('\nFinished.\n')
※ 上記ソースコードは表示の都合上、半角コード '}' が 全角 '｝'になっていることに注意
