app.py

import gradio as gr
import spaces
import os
import io
from PIL import Image
import base64
from scripts.process_utils import initialize, process_image_as_base64, image_to_base64
from scripts.anime import init_model
from datetime import datetime
from pytz import timezone
from scripts.survey import handle_form_submission, handle_visit_choice, handle_proceed, localize, script, generate_image, send_feedback

# 初期化
initialize(_use_local=False, use_gpu=True, use_dotenv=True)

@spaces.GPU
def process_image(input_image, mode, weight1=None, weight2=None):
    tokyo_time = datetime.now(timezone('Asia/Tokyo')).strftime("%Y-%m-%d %H:%M:%S")  # 日本時間のタイムスタンプ
    print(f"[{tokyo_time}] Processing image with mode={mode}, weight1={weight1}, weight2={weight2}")

    # feedback用のファイル名
    tokyo_time = datetime.now(timezone('Asia/Tokyo')).strftime("%Y%m%d_%H%M%S")
    filename = f"{tokyo_time}_mode={mode}_weight1={weight1}_weight2={weight2}.png"

    # 既存の画像処理ロジック
    if mode == "original":
        sotai_image, sketch_image = process_image_as_base64(input_image, mode, None, None)
    elif mode == "refine":
        sotai_image, sketch_image = process_image_as_base64(input_image, mode, weight1, weight2)

    return sotai_image, sketch_image, None, filename

def mix_images(sotai_image_data, sketch_image_data, opacity1, opacity2):
    sotai_image = Image.open(io.BytesIO(base64.b64decode(sotai_image_data))).convert('RGBA')
    sketch_image = Image.open(io.BytesIO(base64.b64decode(sketch_image_data))).convert('RGBA')

    if sotai_image.size != sketch_image.size:
        sketch_image = sketch_image.resize(sotai_image.size, Image.Resampling.LANCZOS)

    mixed_image = Image.new('RGBA', sotai_image.size, (255, 255, 255, 255))

    sotai_alpha = sotai_image.getchannel('A').point(lambda x: int(x * opacity1))
    sketch_alpha = sketch_image.getchannel('A').point(lambda x: int(x * opacity2))

    mixed_image.paste(sketch_image, (0, 0), mask=sketch_alpha)
    mixed_image.paste(sotai_image, (0, 0), mask=sotai_alpha)

    return mixed_image

def send_mixed_feedback(sotai_image_data, sketch_image_data, filename):
    mixed_image = mix_images(sotai_image_data, sketch_image_data, 0.5, 0.5)
    return send_feedback(mixed_image, filename)

with gr.Blocks() as demo:
    form_visible_flag = gr.Textbox(value="false", elem_id="form_flag", visible=False)
    # title
    gr.HTML("<h1>Image2Body demo</h1>")
    # description with translations and additional notes
    gr.HTML("""
        <p>Upload an image and select processing options to generate body and sketch images.</p>
        <p>まだstandingタグのついた女性キャラクターの1000枚の画像しか学習していないため、他のポーズは上手くできないことをご了承ください。</p>
        <p>さらなる情報は<a href="https://x.com/Yeq6X" target="_blank">@Yeq6X</a>までお問い合わせください。</p>
        <p>Note: Currently, the model has been trained on only 1000 images of female characters with the 'standing' tag, so other poses may not be processed accurately.</p>
        <p>For more information, please contact <a href="https://x.com/Yeq6X" target="_blank">@Yeq6X</a>.</p>
        <p>注意：目前模型仅使用带有“standing”标签的1000张女性角色图像进行训练，因此其他姿势可能无法准确处理。</p>
        <p>如需更多信息，请联系<a href="https://x.com/Yeq6X" target="_blank">@Yeq6X</a>。</p>
    """)

    # 訪問回数の選択
    with gr.Column(visible=False) as visit_section:
        # 言語選択セクション
        with gr.Row():
            language_choice = gr.Radio(
                choices=["en", "ja", "zh"],
                label="Select Language / 言語を選択 / 选择语言",
                value="en"
            )

        localized = localize("en")
        welcome_message = gr.HTML(localized["welcome_message"])
        visit_choice = gr.Radio(choices=localized["visit_choices"], label="")

    # 初回訪問のアンケートセクション
    with gr.Column(visible=False) as survey_section:
        # フォームセクション
        form_section = gr.HTML(localize("en")["form_html"])

    # 2回目以降の進むボタンセクション
    with gr.Column(visible=False) as proceed_section:
        # gr.HTML("<h2>再訪ありがとうございます！</h2>")
        # proceed_button = gr.Button("進む")
        proceed_message = gr.HTML(localize("en")["returning_message"])
        proceed_button = gr.Button(localize("en")["proceed_button"], variant="primary")

    # 言語選択変更時の更新
    def update_language(language):
        localized = localize(language)
        return (
            gr.update(value=localized["welcome_message"]),
            gr.update(choices=localized["visit_choices"]),
            gr.update(value=localized["returning_message"]),
            gr.update(value=localized["proceed_button"]),
            gr.update(value=localized["form_html"])
        )

    language_choice.change(
        update_language,
        inputs=[language_choice],
        outputs=[welcome_message, visit_choice, proceed_message, proceed_button, form_section]
    )

    # フォーム送信時の画面切り替え
    def handle_submit():
        return gr.update(visible=False), gr.update(visible=True)

    submit_flag = gr.Textbox(visible=False, value="false")
    submit_flag.change(
        handle_submit,
        inputs=[],
        outputs=[form_section]
    )

    # メイン画面セクション
    with gr.Column(visible=True) as main_section:
        # interface
        submit = None
        with gr.Row():
            with gr.Column() as input_col:
                input_image = gr.Image(type="pil", label="Input Image", height=512)
                with gr.Tab("original"):
                    original_mode = gr.Text("original", label="Mode", visible=False)
                    original_submit = gr.Button("Submit", variant="primary")
                with gr.Tab("refine"):
                    refine_input = [
                        gr.Text("refine", label="Mode", visible=False),
                        gr.Slider(0, 2, value=0.6, step=0.05, label="Weight 1 (Sketch)"),
                        gr.Slider(0, 1, value=0.05, step=0.025, label="Weight 2 (Body)")
                    ]
                    refine_submit = gr.Button("Submit", variant="primary")
                gr.Examples(
                    examples=[f"images/sample{i}.png" for i in [1, 2, 4, 5, 6, 7, 10, 16, 18, 19]],
                    inputs=[input_image]
                )
            with gr.Column() as output_col:
                sotai_image_data = gr.Text(label="Sotai Image data", visible=False)
                sketch_image_data = gr.Text(label="Sketch Image data", visible=False)
                mixed_image = gr.Image(label="Output Image", elem_id="output_image")
                opacity_slider1 = gr.Slider(0, 1, value=0.5, step=0.05, label="Opacity (Sotai)")
                opacity_slider2 = gr.Slider(0, 1, value=0.5, step=0.05, label="Opacity (Sketch)")
                send_filename = gr.Textbox(label="Feedback", visible=False)
                gr.HTML("<h3>Send Feedback Image/画像を送信</h3>")
                gr.HTML("<p>Images are used only for developer review and will not be shared.</p>")
                gr.HTML("<p>画像は開発者が確認するためだけに使用され、公開されません。</p>")
                send_feedback_button = gr.Button("Contribute as Feedback to Developer/開発者へのフィードバックとして協力する")
                feed_back_result = gr.Textbox(label="Feedback Result")


        original_submit.click(
            process_image,
            inputs=[input_image, original_mode],
            outputs=[sotai_image_data, sketch_image_data, mixed_image, send_filename]
        )
        refine_submit.click(
            process_image,
            inputs=[input_image, refine_input[0], refine_input[1], refine_input[2]],
            outputs=[sotai_image_data, sketch_image_data, mixed_image, send_filename]
        )
        sotai_image_data.change(
            mix_images,
            inputs=[sotai_image_data, sketch_image_data, opacity_slider1, opacity_slider2],
            outputs=mixed_image
        )
        opacity_slider1.change(
            mix_images,
            inputs=[sotai_image_data, sketch_image_data, opacity_slider1, opacity_slider2],
            outputs=mixed_image
        )
        opacity_slider2.change(
            mix_images,
            inputs=[sotai_image_data, sketch_image_data, opacity_slider1, opacity_slider2],
            outputs=mixed_image
        )

    # フラグ変更時に画面切り替え
    form_visible_flag.change(
        handle_form_submission,
        inputs=[form_visible_flag],
        outputs=[survey_section, main_section]
    )

    # 選択肢に応じてセクションを切り替え
    visit_choice.change(
        handle_visit_choice,
        inputs=[visit_choice, language_choice],
        outputs=[visit_section, survey_section, proceed_section]
    )

    # 進むボタン押下時の画面遷移
    proceed_button.click(
        handle_proceed,
        inputs=[],
        outputs=[proceed_section, main_section]
    )
    
    send_feedback_button.click(
        send_mixed_feedback,
        inputs=[sotai_image_data, sketch_image_data, send_filename],
        outputs=[feed_back_result],
    )

    # JavaScriptの読み込み
    demo.load(js=script)

demo.launch()