Create app.py

app.py

@@ -0,0 +1,914 @@
+import spaces
+import argparse
+import os
+import shutil
+import cv2
+import gradio as gr
+import numpy as np
+import torch
+from facexlib.utils.face_restoration_helper import FaceRestoreHelper
+import huggingface_hub
+from huggingface_hub import hf_hub_download
+from PIL import Image
+from torchvision.transforms.functional import normalize
+from gradio_client import Client
+import logging
+import time
+
+from dreamo.dreamo_pipeline import DreamOPipeline
+from dreamo.utils import img2tensor, resize_numpy_image_area, tensor2img, resize_numpy_image_long
+from tools import BEN2
+
+parser = argparse.ArgumentParser()
+parser.add_argument('--port', type=int, default=8080)
+parser.add_argument('--no_turbo', action='store_true')
+args = parser.parse_args()
+
+huggingface_hub.login(os.getenv('HF_TOKEN'))
+
+# Text-to-Image API URL
+TEXT2IMG_API_URL = "http://211.233.58.201:7896"
+
+# 로깅 설정
+logging.basicConfig(
+    level=logging.DEBUG,
+    format='%(asctime)s - %(levelname)s - %(message)s')
+
+try:
+    shutil.rmtree('gradio_cached_examples')
+except FileNotFoundError:
+    print("cache folder not exist")
+
+class Generator:
+    def __init__(self):
+        device = torch.device('cuda')
+        # preprocessing models
+        # background remove model: BEN2
+        self.bg_rm_model = BEN2.BEN_Base().to(device).eval()
+        hf_hub_download(repo_id='PramaLLC/BEN2', filename='BEN2_Base.pth', local_dir='models')
+        self.bg_rm_model.loadcheckpoints('models/BEN2_Base.pth')
+        # face crop and align tool: facexlib
+        self.face_helper = FaceRestoreHelper(
+            upscale_factor=1,
+            face_size=512,
+            crop_ratio=(1, 1),
+            det_model='retinaface_resnet50',
+            save_ext='png',
+            device=device,
+        )
+
+        # load dreamo
+        model_root = 'black-forest-labs/FLUX.1-dev'
+        dreamo_pipeline = DreamOPipeline.from_pretrained(model_root, torch_dtype=torch.bfloat16)
+        dreamo_pipeline.load_dreamo_model(device, use_turbo=not args.no_turbo)
+        self.dreamo_pipeline = dreamo_pipeline.to(device)
+
+    @torch.no_grad()
+    def get_align_face(self, img):
+        # the face preprocessing code is same as PuLID
+        self.face_helper.clean_all()
+        image_bgr = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
+        self.face_helper.read_image(image_bgr)
+        self.face_helper.get_face_landmarks_5(only_center_face=True)
+        self.face_helper.align_warp_face()
+        if len(self.face_helper.cropped_faces) == 0:
+            return None
+        align_face = self.face_helper.cropped_faces[0]
+
+        input = img2tensor(align_face, bgr2rgb=True).unsqueeze(0) / 255.0
+        input = input.to(torch.device("cuda"))
+        parsing_out = self.face_helper.face_parse(normalize(input, [0.485, 0.456, 0.406], [0.229, 0.224, 0.225]))[0]
+        parsing_out = parsing_out.argmax(dim=1, keepdim=True)
+        bg_label = [0, 16, 18, 7, 8, 9, 14, 15]
+        bg = sum(parsing_out == i for i in bg_label).bool()
+        white_image = torch.ones_like(input)
+        # only keep the face features
+        face_features_image = torch.where(bg, white_image, input)
+        face_features_image = tensor2img(face_features_image, rgb2bgr=False)
+
+        return face_features_image
+
+
+generator = Generator()
+
+
+@spaces.GPU
+@torch.inference_mode()
+def generate_image(
+    ref_image1,
+    ref_image2,
+    ref_task1,
+    ref_task2,
+    prompt,
+    seed,
+    width=1024,
+    height=1024,
+    ref_res=512,
+    num_steps=12,
+    guidance=3.5,
+    true_cfg=1,
+    cfg_start_step=0,
+    cfg_end_step=0,
+    neg_prompt='',
+    neg_guidance=3.5,
+    first_step_guidance=0,
+):
+    print(prompt)
+    ref_conds = []
+    debug_images = []
+
+    ref_images = [ref_image1, ref_image2]
+    ref_tasks = [ref_task1, ref_task2]
+
+    for idx, (ref_image, ref_task) in enumerate(zip(ref_images, ref_tasks)):
+        if ref_image is not None:
+            if ref_task == "id":
+                ref_image = resize_numpy_image_long(ref_image, 1024)
+                ref_image = generator.get_align_face(ref_image)
+            elif ref_task != "style":
+                ref_image = generator.bg_rm_model.inference(Image.fromarray(ref_image))
+            if ref_task != "id":
+                ref_image = resize_numpy_image_area(np.array(ref_image), ref_res * ref_res)
+            debug_images.append(ref_image)
+            ref_image = img2tensor(ref_image, bgr2rgb=False).unsqueeze(0) / 255.0
+            ref_image = 2 * ref_image - 1.0
+            ref_conds.append(
+                {
+                    'img': ref_image,
+                    'task': ref_task,
+                    'idx': idx + 1,
+                }
+            )
+
+    seed = int(seed)
+    if seed == -1:
+        seed = torch.Generator(device="cpu").seed()
+
+    image = generator.dreamo_pipeline(
+        prompt=prompt,
+        width=width,
+        height=height,
+        num_inference_steps=num_steps,
+        guidance_scale=guidance,
+        ref_conds=ref_conds,
+        generator=torch.Generator(device="cpu").manual_seed(seed),
+        true_cfg_scale=true_cfg,
+        true_cfg_start_step=cfg_start_step,
+        true_cfg_end_step=cfg_end_step,
+        negative_prompt=neg_prompt,
+        neg_guidance_scale=neg_guidance,
+        first_step_guidance_scale=first_step_guidance if first_step_guidance > 0 else guidance,
+    ).images[0]
+
+    return image, debug_images, seed
+
+
+# Video generation functions
+import requests
+import random
+import tempfile
+import subprocess
+from gradio_client import Client, handle_file
+
+REMOTE_ENDPOINT = os.getenv("H100_URL") 
+
+client = Client(REMOTE_ENDPOINT)
+
+def run_process_video_api(image_path: str, prompt: str, video_length: float = 2.0):
+    seed_val = random.randint(0, 9999999)
+    negative_prompt = ""
+    use_teacache = True
+
+    result = client.predict(
+        input_image=handle_file(image_path),
+        prompt=prompt,
+        n_prompt=negative_prompt,
+        seed=seed_val,
+        use_teacache=use_teacache,
+        video_length=video_length,
+        api_name="/process",
+    )
+    video_dict, preview_dict, md_text, html_text = result
+    video_path = video_dict.get("video") if isinstance(video_dict, dict) else None
+    return video_path
+
+def add_watermark_to_video(input_video_path: str, watermark_text="Ginigen.com") -> str:
+    if not os.path.exists(input_video_path):
+        raise FileNotFoundError(f"Input video not found: {input_video_path}")
+
+    base, ext = os.path.splitext(input_video_path)
+    watermarked_path = base + "_wm" + ext
+    cmd = [
+        "ffmpeg", "-y",
+        "-i", input_video_path,
+        "-vf", f"drawtext=fontsize=20:fontcolor=white:text='{watermark_text}':x=w-tw-10:y=h-th-10:box=1:[email protected]:boxborderw=5",
+        "-codec:a", "copy",
+        watermarked_path
+    ]
+    try:
+        subprocess.run(cmd, check=True)
+    except Exception as e:
+        print(f"[WARN] FFmpeg watermark failed: {e}")
+        return input_video_path
+
+    return watermarked_path
+
+def generate_video_from_image(image_array: np.ndarray):
+    if image_array is None:
+        raise gr.Error("이미지가 없습니다.")
+
+    with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as fp:
+        temp_img_path = fp.name
+        Image.fromarray(image_array).save(temp_img_path, format="PNG")
+
+    default_video_prompt = "Generate a video with smooth and natural movement. Objects should have visible motion while maintaining fluid transitions."
+    result_video_path = run_process_video_api(
+        image_path=temp_img_path,
+        prompt=default_video_prompt,
+        video_length=2.0,
+    )
+    if result_video_path is None:
+        raise gr.Error("영상 API 호출 실패 또는 결과 없음")
+
+    final_video = add_watermark_to_video(result_video_path, watermark_text="Ginigen.com")
+    return final_video
+
+
+# Text-to-Image functions
+def test_text2img_api_connection() -> str:
+    """Text-to-Image API 서버 연결 테스트"""
+    try:
+        client = Client(TEXT2IMG_API_URL)
+        return "API 연결 성공: 정상 작동 중"
+    except Exception as e:
+        logging.error(f"API connection test failed: {e}")
+        return f"API 연결 실패: {e}"
+
+def generate_text_to_image(prompt: str, width: int, height: int, guidance: float, inference_steps: int, seed: int) -> tuple:
+    """텍스트를 이미지로 생성하는 함수"""
+    if not prompt:
+        return None, "오류: 프롬프트를 입력해주세요"
+    
+    try:
+        client = Client(TEXT2IMG_API_URL)
+        result = client.predict(
+            prompt=prompt,
+            width=int(width),
+            height=int(height),
+            guidance=float(guidance),
+            inference_steps=int(inference_steps),
+            seed=int(seed),
+            do_img2img=False,
+            init_image=None,
+            image2image_strength=0.8,
+            resize_img=True,
+            api_name="/generate_image"
+        )
+        return result[0], f"사용된 시드: {result[1]}"
+    except Exception as e:
+        logging.error(f"Image generation failed: {str(e)}")
+        return None, f"오류: {str(e)}"
+
+# Image size presets
+IMAGE_PRESETS = {
+    "커스텀": {"width": 1024, "height": 1024, "label": "커스텀 크기"},
+    "1:1 정사각형": {"width": 1024, "height": 1024, "label": "1:1 (정사각형)"},
+    "4:3 표준": {"width": 1024, "height": 768, "label": "4:3 (표준)"},
+    "16:9 와이드스크린": {"width": 1024, "height": 576, "label": "16:9 (와이드스크린)"},
+    "9:16 세로형": {"width": 576, "height": 1024, "label": "9:16 (세로형)"},
+    "6:19 특수 세로형": {"width": 324, "height": 1024, "label": "6:19 (특수 세로형)"},
+    "Instagram 정사각형": {"width": 1080, "height": 1080, "label": "Instagram 정사각형 (1:1)"},
+    "Instagram 스토리": {"width": 1080, "height": 1920, "label": "Instagram 스토리 (9:16)"},
+    "Instagram 가로형": {"width": 1080, "height": 566, "label": "Instagram 가로형 (1.91:1)"},
+    "Facebook 커버": {"width": 820, "height": 312, "label": "Facebook 커버 (2.63:1)"},
+    "Twitter 헤더": {"width": 1500, "height": 500, "label": "Twitter 헤더 (3:1)"},
+    "YouTube 썸네일": {"width": 1280, "height": 720, "label": "YouTube 썸네일 (16:9)"},
+    "LinkedIn 배너": {"width": 1584, "height": 396, "label": "LinkedIn 배너 (4:1)"},
+}
+
+def update_dimensions(preset):
+    """선택된 프리셋에 따라 width, height 업데이트"""
+    if preset in IMAGE_PRESETS:
+        return IMAGE_PRESETS[preset]["width"], IMAGE_PRESETS[preset]["height"]
+    return 1024, 1024
+
+
+# Custom CSS
+_CUSTOM_CSS_ = """
+:root {
+    --primary-color: #f8c3cd;
+    --secondary-color: #b3e5fc;
+    --background-color: #f5f5f7;
+    --card-background: #ffffff;
+    --text-color: #424242;
+    --accent-color: #ffb6c1;
+    --success-color: #c8e6c9;
+    --warning-color: #fff9c4;
+    --shadow-color: rgba(0, 0, 0, 0.1);
+    --border-radius: 12px;
+}
+
+body {
+    background-color: var(--background-color) !important;
+    font-family: 'Inter', -apple-system, BlinkMacSystemFont, sans-serif !important;
+}
+
+.gradio-container {
+    max-width: 1200px !important;
+    margin: 0 auto !important;
+}
+
+h1 {
+    color: #9c27b0 !important;
+    font-weight: 800 !important;
+    text-shadow: 2px 2px 4px rgba(156, 39, 176, 0.2) !important;
+    letter-spacing: -0.5px !important;
+}
+
+.panel-box {
+    border-radius: var(--border-radius) !important;
+    box-shadow: 0 8px 16px var(--shadow-color) !important;
+    background-color: var(--card-background) !important;
+    border: none !important;
+    overflow: hidden !important;
+    padding: 20px !important;
+    margin-bottom: 20px !important;
+}
+
+button.gr-button {
+    background: linear-gradient(135deg, var(--primary-color), #e1bee7) !important;
+    border-radius: var(--border-radius) !important;
+    color: #4a148c !important;
+    font-weight: 600 !important;
+    border: none !important;
+    padding: 10px 20px !important;
+    transition: all 0.3s ease !important;
+    box-shadow: 0 4px 6px rgba(0, 0, 0, 0.1) !important;
+}
+
+button.gr-button:hover {
+    transform: translateY(-2px) !important;
+    box-shadow: 0 6px 10px rgba(0, 0, 0, 0.15) !important;
+    background: linear-gradient(135deg, #e1bee7, var(--primary-color)) !important;
+}
+
+input, select, textarea, .gr-input {
+    border-radius: 8px !important;
+    border: 2px solid #e0e0e0 !important;
+    padding: 10px 15px !important;
+    transition: all 0.3s ease !important;
+    background-color: #fafafa !important;
+}
+
+input:focus, select:focus, textarea:focus, .gr-input:focus {
+    border-color: var(--primary-color) !important;
+    box-shadow: 0 0 0 3px rgba(248, 195, 205, 0.3) !important;
+}
+
+.gr-form input[type=range] {
+    appearance: none !important;
+    width: 100% !important;
+    height: 6px !important;
+    background: #e0e0e0 !important;
+    border-radius: 5px !important;
+    outline: none !important;
+}
+
+.gr-form input[type=range]::-webkit-slider-thumb {
+    appearance: none !important;
+    width: 16px !important;
+    height: 16px !important;
+    background: var(--primary-color) !important;
+    border-radius: 50% !important;
+    cursor: pointer !important;
+    border: 2px solid white !important;
+    box-shadow: 0 2px 4px rgba(0, 0, 0, 0.1) !important;
+}
+
+.gr-form select {
+    background-color: white !important;
+    border: 2px solid #e0e0e0 !important;
+    border-radius: 8px !important;
+    padding: 10px 15px !important;
+}
+
+.gr-image-input {
+    border: 2px dashed #b39ddb !important;
+    border-radius: var(--border-radius) !important;
+    background-color: #f3e5f5 !important;
+    padding: 20px !important;
+    display: flex !important;
+    flex-direction: column !important;
+    align-items: center !important;
+    justify-content: center !important;
+    transition: all 0.3s ease !important;
+}
+
+.gr-image-input:hover {
+    background-color: #ede7f6 !important;
+    border-color: #9575cd !important;
+}
+
+body::before {
+    content: "" !important;
+    position: fixed !important;
+    top: 0 !important;
+    left: 0 !important;
+    width: 100% !important;
+    height: 100% !important;
+    background: 
+        radial-gradient(circle at 10% 20%, rgba(248, 195, 205, 0.1) 0%, rgba(245, 245, 247, 0) 20%),
+        radial-gradient(circle at 80% 70%, rgba(179, 229, 252, 0.1) 0%, rgba(245, 245, 247, 0) 20%) !important;
+    pointer-events: none !important;
+    z-index: -1 !important;
+}
+
+.gr-gallery {
+    grid-gap: 15px !important;
+}
+
+.gr-gallery-item {
+    border-radius: var(--border-radius) !important;
+    overflow: hidden !important;
+    box-shadow: 0 4px 8px var(--shadow-color) !important;
+    transition: transform 0.3s ease !important;
+}
+
+.gr-gallery-item:hover {
+    transform: scale(1.02) !important;
+}
+
+.gr-form label {
+    font-weight: 600 !important;
+    color: #673ab7 !important;
+    margin-bottom: 5px !important;
+}
+
+.gr-padded {
+    padding: 20px !important;
+}
+
+.gr-compact {
+    gap: 15px !important;
+}
+
+.gr-form > div {
+    margin-bottom: 16px !important;
+}
+
+.gr-form h3 {
+    color: #7b1fa2 !important;
+    margin-top: 5px !important;
+    margin-bottom: 15px !important;
+    border-bottom: 2px solid #e1bee7 !important;
+    padding-bottom: 8px !important;
+}
+
+#examples-panel {
+    background-color: #f3e5f5 !important;
+    border-radius: var(--border-radius) !important;
+    padding: 15px !important;
+    box-shadow: 0 4px 8px rgba(0, 0, 0, 0.05) !important;
+}
+
+#examples-panel h2 {
+    color: #7b1fa2 !important;
+    font-size: 1.5rem !important;
+    margin-bottom: 15px !important;
+}
+
+.gr-accordion {
+    border: 1px solid #e0e0e0 !important;
+    border-radius: var(--border-radius) !important;
+    overflow: hidden !important;
+}
+
+.gr-accordion summary {
+    padding: 12px 16px !important;
+    background-color: #f9f9f9 !important;
+    cursor: pointer !important;
+    font-weight: 600 !important;
+    color: #673ab7 !important;
+}
+
+#generate-btn, #text2img-generate-btn {
+    background: linear-gradient(135deg, #ff9a9e, #fad0c4) !important;
+    font-size: 1.1rem !important;
+    padding: 12px 24px !important;
+    margin-top: 10px !important;
+    margin-bottom: 15px !important;
+    width: 100% !important;
+}
+
+#generate-btn:hover, #text2img-generate-btn:hover {
+    background: linear-gradient(135deg, #fad0c4, #ff9a9e) !important;
+}
+
+/* Tab styling */
+.gr-tabs {
+    border: none !important;
+    margin-top: 20px !important;
+}
+
+.gr-tab {
+    background-color: #f3e5f5 !important;
+    border: none !important;
+    padding: 12px 24px !important;
+    font-weight: 600 !important;
+    color: #673ab7 !important;
+    transition: all 0.3s ease !important;
+}
+
+.gr-tab.selected {
+    background: linear-gradient(135deg, var(--primary-color), #e1bee7) !important;
+    color: white !important;
+}
+
+.gr-tab:hover {
+    background-color: #ede7f6 !important;
+}
+"""
+
+_HEADER_ = '''
+<div style="text-align: center; max-width: 850px; margin: 0 auto; padding: 25px 0;">
+    <div style="background: linear-gradient(135deg, #f8c3cd, #e1bee7, #b3e5fc); color: white; padding: 15px; border-radius: 15px; box-shadow: 0 10px 20px rgba(0,0,0,0.1); margin-bottom: 20px;">
+        <h1 style="font-size: 3rem; font-weight: 800; margin: 0; color: white; text-shadow: 2px 2px 4px rgba(0,0,0,0.2);">✨ DreamO Video ✨</h1>
+        <p style="font-size: 1.2rem; margin: 10px 0 0;">Create customized images with advanced AI</p>
+    </div>
+
+    <div style="background: white; padding: 15px; border-radius: 12px; box-shadow: 0 5px 15px rgba(0,0,0,0.05);">
+        <p style="font-size: 1rem; margin: 0;">In the current demo version, due to ZeroGPU limitations, video generation is restricted to 2 seconds only. (The full version supports generation of up to 60 seconds)</p>
+    </div>
+
+</div>
+
+<div style="background: #fff9c4; padding: 15px; border-radius: 12px; margin-bottom: 20px; border-left: 5px solid #ffd54f; box-shadow: 0 5px 15px rgba(0,0,0,0.05);">
+    <h3 style="margin-top: 0; color: #ff6f00;">🚩 Update Notes:</h3>
+    <ul style="margin-bottom: 0; padding-left: 20px;">
+        <li><b>2025.05.11:</b> We have updated the model to mitigate over-saturation and plastic-face issues. The new version shows consistent improvements over the previous release.</li>
+        <li><b>2025.05.13:</b> 'DreamO Video' Integration version Release</li>
+        <li><b>2025.05.28:</b> Added 'Text-to-Image' tab with multiple aspect ratios and SNS presets</li>
+    </ul>
+</div>
+'''
+
+_CITE_ = r"""
+<div style="background: white; padding: 20px; border-radius: 12px; margin-top: 20px; box-shadow: 0 5px 15px rgba(0,0,0,0.05);">
+    <p style="margin: 0; font-size: 1.1rem;">If DreamO is helpful, please help to ⭐ the <a href='https://discord.gg/openfreeai' target='_blank' style="color: #9c27b0; font-weight: 600;">community</a>. Thanks!</p>
+    <hr style="border: none; height: 1px; background-color: #e0e0e0; margin: 15px 0;">
+    <h4 style="margin: 0 0 10px; color: #7b1fa2;">📧 Contact</h4>
+    <p style="margin: 0;">If you have any questions or feedback, feel free to open a discussion or contact <b>[email protected]</b></p>
+</div>
+"""
+
+def create_demo():
+    with gr.Blocks(css=_CUSTOM_CSS_) as demo:
+        gr.HTML(_HEADER_)
+
+        with gr.Tabs():
+            # DreamO Tab
+            with gr.Tab("DreamO (참조 이미지 기반)"):
+                with gr.Row():
+                    with gr.Column(scale=6):
+                        with gr.Group(elem_id="input-panel", elem_classes="panel-box"):
+                            gr.Markdown("### 📸 Reference Images")
+                            with gr.Row():
+                                with gr.Column():
+                                    ref_image1 = gr.Image(label="Reference Image 1", type="numpy", height=256, elem_id="ref-image-1")
+                                    ref_task1 = gr.Dropdown(choices=["ip", "id", "style"], value="ip", label="Task for Reference Image 1", elem_id="ref-task-1")
+                                
+                                with gr.Column():
+                                    ref_image2 = gr.Image(label="Reference Image 2", type="numpy", height=256, elem_id="ref-image-2")
+                                    ref_task2 = gr.Dropdown(choices=["ip", "id", "style"], value="ip", label="Task for Reference Image 2", elem_id="ref-task-2")
+                            
+                            gr.Markdown("### ✏️ Generation Parameters")
+                            prompt = gr.Textbox(label="Prompt", value="a person playing guitar in the street", elem_id="prompt-input")
+                            
+                            with gr.Row():
+                                width = gr.Slider(768, 1024, 1024, step=16, label="Width", elem_id="width-slider")
+                                height = gr.Slider(768, 1024, 1024, step=16, label="Height", elem_id="height-slider")
+                            
+                            with gr.Row():
+                                num_steps = gr.Slider(8, 30, 12, step=1, label="Number of Steps", elem_id="steps-slider")
+                                guidance = gr.Slider(1.0, 10.0, 3.5, step=0.1, label="Guidance Scale", elem_id="guidance-slider")
+                            
+                            seed = gr.Textbox(label="Seed (-1 for random)", value="-1", elem_id="seed-input")
+                            
+                            with gr.Accordion("Advanced Options", open=False):
+                                ref_res = gr.Slider(512, 1024, 512, step=16, label="Resolution for Reference Image")
+                                neg_prompt = gr.Textbox(label="Negative Prompt", value="")
+                                neg_guidance = gr.Slider(1.0, 10.0, 3.5, step=0.1, label="Negative Guidance")
+                                
+                                with gr.Row():
+                                    true_cfg = gr.Slider(1, 5, 1, step=0.1, label="True CFG")
+                                    first_step_guidance = gr.Slider(0, 10, 0, step=0.1, label="First Step Guidance")
+                                
+                                with gr.Row():
+                                    cfg_start_step = gr.Slider(0, 30, 0, step=1, label="CFG Start Step")
+                                    cfg_end_step = gr.Slider(0, 30, 0, step=1, label="CFG End Step")
+                            
+                            generate_btn = gr.Button("✨ Generate Image", elem_id="generate-btn")
+                            gr.HTML(_CITE_)
+
+                    with gr.Column(scale=6):
+                        with gr.Group(elem_id="output-panel", elem_classes="panel-box"):
+                            gr.Markdown("### 🖼️ Generated Result")
+                            output_image = gr.Image(label="Generated Image", elem_id="output-image", format='png')
+                            seed_output = gr.Textbox(label="Used Seed", elem_id="seed-output")
+                            
+                            generate_video_btn = gr.Button("🎬 Generate Video from Image")
+                            output_video = gr.Video(label="Generated Video", elem_id="video-output")
+
+                            gr.Markdown("### 🔍 Preprocessing")
+                            debug_image = gr.Gallery(
+                                label="Preprocessing Results (including face crop and background removal)",
+                                elem_id="debug-gallery",
+                            )
+
+                with gr.Group(elem_id="examples-panel", elem_classes="panel-box"):
+                    gr.Markdown("## 📚 Examples")
+                    example_inps = [
+                        [
+                            'example_inputs/choi.jpg',
+                            None,
+                            'ip',
+                            'ip',
+                            'a woman sitting on the cloud, playing guitar',
+                            1206523688721442817,
+                        ],
+                        [
+                            'example_inputs/choi.jpg',
+                            None,
+                            'id',
+                            'ip',
+                            'a woman holding a sign saying "TOP", on the mountain',
+                            10441727852953907380,
+                        ],
+                        [
+                            'example_inputs/perfume.png',
+                            None,
+                            'ip',
+                            'ip',
+                            'a perfume under spotlight',
+                            116150031980664704,
+                        ],
+                        [
+                            'example_inputs/choi.jpg',
+                            None,
+                            'id',
+                            'ip',
+                            'portrait, in alps',
+                            5443415087540486371,
+                        ],
+                        [
+                            'example_inputs/mickey.png',
+                            None,
+                            'style',
+                            'ip',
+                            'generate a same style image. A rooster wearing overalls.',
+                            6245580464677124951,
+                        ],
+                        [
+                            'example_inputs/mountain.png',
+                            None,
+                            'style',
+                            'ip',
+                            'generate a same style image. A pavilion by the river, and the distant mountains are endless',
+                            5248066378927500767,
+                        ],
+                        [
+                            'example_inputs/shirt.png',
+                            'example_inputs/skirt.jpeg',
+                            'ip',
+                            'ip',
+                            'A girl is wearing a short-sleeved shirt and a short skirt on the beach.',
+                            9514069256241143615,
+                        ],
+                        [
+                            'example_inputs/woman2.png',
+                            'example_inputs/dress.png',
+                            'id',
+                            'ip',
+                            'the woman wearing a dress, In the banquet hall',
+                            7698454872441022867,
+                        ],
+                        [
+                            'example_inputs/dog1.png',
+                            'example_inputs/dog2.png',
+                            'ip',
+                            'ip',
+                            'two dogs in the jungle',
+                            6187006025405083344,
+                        ],
+                    ]
+                    gr.Examples(
+                        examples=example_inps,
+                        inputs=[ref_image1, ref_image2, ref_task1, ref_task2, prompt, seed],
+                        label='Examples by category: IP task (rows 1-4), ID task (row 5), Style task (rows 6-7), Try-On task (rows 8-9)',
+                        cache_examples='lazy',
+                        outputs=[output_image, debug_image, seed_output],
+                        fn=generate_image,
+                    )
+
+                # Event handlers for DreamO tab
+                generate_btn.click(
+                    fn=generate_image,
+                    inputs=[
+                        ref_image1,
+                        ref_image2,
+                        ref_task1,
+                        ref_task2,
+                        prompt,
+                        seed,
+                        width,
+                        height,
+                        ref_res,
+                        num_steps,
+                        guidance,
+                        true_cfg,
+                        cfg_start_step,
+                        cfg_end_step,
+                        neg_prompt,
+                        neg_guidance,
+                        first_step_guidance,
+                    ],
+                    outputs=[output_image, debug_image, seed_output],
+                )
+
+                def on_click_generate_video(img):
+                    if img is None:
+                        raise gr.Error("먼저 이미지를 생성해주세요.")
+                    video_path = generate_video_from_image(img)
+                    return video_path
+
+                generate_video_btn.click(
+                    fn=on_click_generate_video,
+                    inputs=[output_image],
+                    outputs=[output_video],
+                )
+
+            # Text-to-Image Tab
+            with gr.Tab("텍스트 to 이미지"):
+                with gr.Row():
+                    with gr.Column(scale=6):
+                        with gr.Group(elem_id="text2img-input-panel", elem_classes="panel-box"):
+                            gr.Markdown("### 📝 텍스트로 이미지 생성")
+                            
+                            # API 상태 표시
+                            text2img_status = gr.Textbox(
+                                label="API 상태",
+                                value="API 연결 확인 중...",
+                                interactive=False
+                            )
+                            
+                            # 프롬프트 입력
+                            text2img_prompt = gr.Textbox(
+                                label="프롬프트",
+                                placeholder="생성하고 싶은 이미지를 설명하세요...",
+                                lines=3
+                            )
+                            
+                            # 이미지 크기 프리셋
+                            size_preset = gr.Dropdown(
+                                choices=list(IMAGE_PRESETS.keys()),
+                                value="1:1 정사각형",
+                                label="이미지 크기 프리셋",
+                                interactive=True
+                            )
+                            
+                            with gr.Row():
+                                text2img_width = gr.Slider(
+                                    minimum=256,
+                                    maximum=2048,
+                                    value=1024,
+                                    step=64,
+                                    label="너비"
+                                )
+                                
+                                text2img_height = gr.Slider(
+                                    minimum=256,
+                                    maximum=2048,
+                                    value=1024,
+                                    step=64,
+                                    label="높이"
+                                )
+                            
+                            with gr.Row():
+                                text2img_guidance = gr.Slider(
+                                    minimum=1.0,
+                                    maximum=20.0,
+                                    value=3.5,
+                                    step=0.1,
+                                    label="가이던스 스케일"
+                                )
+                                
+                                text2img_steps = gr.Slider(
+                                    minimum=1,
+                                    maximum=50,
+                                    value=30,
+                                    step=1,
+                                    label="인퍼런스 스텝"
+                                )
+                            
+                            text2img_seed = gr.Number(
+                                label="시드 (-1: 랜덤)",
+                                value=-1,
+                                precision=0
+                            )
+                            
+                            text2img_generate_btn = gr.Button("✨ 이미지 생성", elem_id="text2img-generate-btn")
+                            
+                            # 생성 상태 표시
+                            text2img_generation_status = gr.Textbox(
+                                label="생성 상태",
+                                value="",
+                                interactive=False,
+                                visible=False
+                            )
+
+                    with gr.Column(scale=6):
+                        with gr.Group(elem_id="text2img-output-panel", elem_classes="panel-box"):
+                            gr.Markdown("### 🖼️ 생성 결과")
+                            text2img_output = gr.Image(label="생성된 이미지", format='png')
+                            text2img_used_seed = gr.Textbox(label="사용된 시드")
+                            
+                            # 비디오 생성 버튼
+                            text2img_video_btn = gr.Button("🎬 이미지를 비디오로 변환")
+                            text2img_video = gr.Video(label="생성된 비디오")
+
+                # Text-to-Image 탭 예제
+                with gr.Group(elem_id="text2img-examples-panel", elem_classes="panel-box"):
+                    gr.Markdown("## 📚 텍스트 to 이미지 예제")
+                    text2img_examples = [
+                        ["A serene Japanese garden with cherry blossoms", "1:1 정사각형", 3.5, 30, 42],
+                        ["Futuristic cityscape at sunset, cyberpunk style", "16:9 와이드스크린", 4.0, 35, 123],
+                        ["Portrait of a mysterious woman with flowing hair", "Instagram 스토리", 3.0, 25, 789],
+                        ["Epic fantasy dragon breathing fire", "YouTube 썸네일", 5.0, 40, 456],
+                        ["Minimalist logo design for tech company", "LinkedIn 배너", 3.5, 30, 321],
+                    ]
+                    gr.Examples(
+                        examples=text2img_examples,
+                        inputs=[text2img_prompt, size_preset, text2img_guidance, text2img_steps, text2img_seed],
+                        label='예제 프롬프트와 설정',
+                        cache_examples=False,
+                    )
+
+                # Event handlers for Text-to-Image tab
+                size_preset.change(
+                    fn=update_dimensions,
+                    inputs=[size_preset],
+                    outputs=[text2img_width, text2img_height]
+                )
+
+                def on_text2img_generate_click():
+                    text2img_generation_status.visible = True
+                    text2img_generation_status.value = "이미지 생성 중... 잠시만 기다려주세요"
+                    return text2img_generation_status
+
+                def on_text2img_generate_complete():
+                    text2img_generation_status.value = "이미지 생성 완료!"
+                    return text2img_generation_status
+
+                text2img_generate_btn.click(
+                    fn=on_text2img_generate_click,
+                    outputs=[text2img_generation_status]
+                ).then(
+                    fn=generate_text_to_image,
+                    inputs=[text2img_prompt, text2img_width, text2img_height, text2img_guidance, text2img_steps, text2img_seed],
+                    outputs=[text2img_output, text2img_used_seed]
+                ).then(
+                    fn=on_text2img_generate_complete,
+                    outputs=[text2img_generation_status]
+                )
+
+                def on_text2img_video_click(img):
+                    if img is None:
+                        raise gr.Error("먼저 이미지를 생성해주세요.")
+                    video_path = generate_video_from_image(img)
+                    return video_path
+
+                text2img_video_btn.click(
+                    fn=on_text2img_video_click,
+                    inputs=[text2img_output],
+                    outputs=[text2img_video],
+                )
+
+                # API 상태 확인
+                def check_text2img_api_status():
+                    return test_text2img_api_connection()
+
+                demo.load(
+                    fn=check_text2img_api_status,
+                    outputs=[text2img_status]
+                )
+
+    return demo
+
+
+if __name__ == '__main__':
+    demo = create_demo()
+    demo.launch(
+        server_name="0.0.0.0",
+        share=True,
+        ssr_mode=False
+    )