app.py

import tempfile
import time
from collections.abc import Sequence
from typing import Any, cast
import os
from huggingface_hub import login, hf_hub_download

import gradio as gr
import numpy as np
import pillow_heif
import spaces
import torch
from gradio_image_annotation import image_annotator
from gradio_imageslider import ImageSlider
from PIL import Image
from pymatting.foreground.estimate_foreground_ml import estimate_foreground_ml
from refiners.fluxion.utils import no_grad
from refiners.solutions import BoxSegmenter
from transformers import GroundingDinoForObjectDetection, GroundingDinoProcessor
from diffusers import FluxPipeline
from transformers import pipeline, AutoTokenizer, AutoModelForSeq2SeqLM
import gc

from PIL import Image, ImageDraw, ImageFont
from PIL import Image
from gradio_client import Client, handle_file
import uuid

import gradio as gr
import spaces
import torch
from diffusers import AutoencoderKL, TCDScheduler
from diffusers.models.model_loading_utils import load_state_dict
from gradio_imageslider import ImageSlider
from huggingface_hub import hf_hub_download
from transformers import pipeline

from controlnet_union import ControlNetModel_Union
from pipeline_fill_sd_xl import StableDiffusionXLFillPipeline

def debug_event(event_name, *args):
    """이벤트 디버깅 유틸리티"""
    print(f"Event '{event_name}' triggered at {time.strftime('%H:%M:%S')}")
    print(f"Arguments: {args}")
    return args
    
def clear_memory():
    """메모리 정리 함수"""
    gc.collect()
    try:
        if torch.cuda.is_available():
            with torch.cuda.device(0):  # 명시적으로 device 0 사용
                torch.cuda.empty_cache()
    except:
        pass

# GPU 설정
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")  # 명시적으로 cuda:0 지정

# GPU 설정을 try-except로 감싸기
if torch.cuda.is_available():
    try:
        with torch.cuda.device(0):
            torch.cuda.empty_cache()
            torch.backends.cudnn.benchmark = True
            torch.backends.cuda.matmul.allow_tf32 = True
    except:
        print("Warning: Could not configure CUDA settings")

# 번역 모델 초기화
model_name = "Helsinki-NLP/opus-mt-ko-en"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name).to('cpu')
translator = pipeline("translation", model=model, tokenizer=tokenizer, device=-1)

def translate_to_english(text: str) -> str:
    """한글 텍스트를 영어로 번역"""
    try:
        if any(ord('가') <= ord(char) <= ord('힣') for char in text):
            translated = translator(text, max_length=128)[0]['translation_text']
            print(f"Translated '{text}' to '{translated}'")
            return translated
        return text
    except Exception as e:
        print(f"Translation error: {str(e)}")
        return text

BoundingBox = tuple[int, int, int, int]

pillow_heif.register_heif_opener()
pillow_heif.register_avif_opener()

# HF 토큰 설정
HF_TOKEN = os.getenv("HF_TOKEN")
if HF_TOKEN is None:
    raise ValueError("Please set the HF_TOKEN environment variable")

try:
    login(token=HF_TOKEN)
except Exception as e:
    raise ValueError(f"Failed to login to Hugging Face: {str(e)}")

# 모델 초기화
segmenter = BoxSegmenter(device="cpu")
segmenter.device = device
segmenter.model = segmenter.model.to(device=segmenter.device)

gd_model_path = "IDEA-Research/grounding-dino-base"
gd_processor = GroundingDinoProcessor.from_pretrained(gd_model_path)
gd_model = GroundingDinoForObjectDetection.from_pretrained(gd_model_path, torch_dtype=torch.float32)
gd_model = gd_model.to(device=device)
assert isinstance(gd_model, GroundingDinoForObjectDetection)

# FLUX 파이프라인 초기화
pipe = FluxPipeline.from_pretrained(
    "black-forest-labs/FLUX.1-dev",
    torch_dtype=torch.float16,
    use_auth_token=HF_TOKEN
)
pipe.enable_attention_slicing(slice_size="auto")

# LoRA 가중치 로드
pipe.load_lora_weights(
    hf_hub_download(
        "ByteDance/Hyper-SD",
        "Hyper-FLUX.1-dev-8steps-lora.safetensors",
        use_auth_token=HF_TOKEN
    )
)
pipe.fuse_lora(lora_scale=0.125)

# GPU 설정을 try-except로 감싸기
try:
    if torch.cuda.is_available():
        pipe = pipe.to("cuda:0")  # 명시적으로 cuda:0 지정
except Exception as e:
    print(f"Warning: Could not move pipeline to CUDA: {str(e)}")


#------------------------------- 이미지 인페인팅 ----------------------

client = Client("NabeelShar/BiRefNet_for_text_writing")

MODELS = {
    "RealVisXL V5.0 Lightning": "SG161222/RealVisXL_V5.0_Lightning",
}

config_file = hf_hub_download(
    "xinsir/controlnet-union-sdxl-1.0",
    filename="config_promax.json",
)

config = ControlNetModel_Union.load_config(config_file)
controlnet_model = ControlNetModel_Union.from_config(config)
model_file = hf_hub_download(
    "xinsir/controlnet-union-sdxl-1.0",
    filename="diffusion_pytorch_model_promax.safetensors",
)
state_dict = load_state_dict(model_file)
model, _, _, _, _ = ControlNetModel_Union._load_pretrained_model(
    controlnet_model, state_dict, model_file, "xinsir/controlnet-union-sdxl-1.0"
)
model.to(device="cuda", dtype=torch.float16)

vae = AutoencoderKL.from_pretrained(
    "madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16
).to("cuda")

pipe = StableDiffusionXLFillPipeline.from_pretrained(
    "SG161222/RealVisXL_V5.0_Lightning",
    torch_dtype=torch.float16,
    vae=vae,
    controlnet=model,
    variant="fp16",
).to("cuda")

pipe.scheduler = TCDScheduler.from_config(pipe.scheduler.config)

def translate_if_korean(text):
    # 입력된 텍스트가 한글을 포함하고 있는지 확인
    if any('\u3131' <= char <= '\u318E' or '\uAC00' <= char <= '\uD7A3' for char in text):
        # 한글이 포함되어 있다면 번역
        translated = translator(text)[0]['translation_text']
        print(f"Translated prompt: {translated}")  # 디버깅을 위한 출력
        return translated
    return text

@spaces.GPU
def fill_image(prompt, image, model_selection):
    # 프롬프트 번역
    translated_prompt = translate_if_korean(prompt)
    
    (
        prompt_embeds,
        negative_prompt_embeds,
        pooled_prompt_embeds,
        negative_pooled_prompt_embeds,
    ) = pipe.encode_prompt(translated_prompt, "cuda", True)

    source = image["background"]
    mask = image["layers"][0]

    alpha_channel = mask.split()[3]
    binary_mask = alpha_channel.point(lambda p: p > 0 and 255)
    cnet_image = source.copy()
    cnet_image.paste(0, (0, 0), binary_mask)

    for image in pipe(
        prompt_embeds=prompt_embeds,
        negative_prompt_embeds=negative_prompt_embeds,
        pooled_prompt_embeds=pooled_prompt_embeds,
        negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
        image=cnet_image,
    ):
        yield image, cnet_image

    image = image.convert("RGBA")
    cnet_image.paste(image, (0, 0), binary_mask)

    yield source, cnet_image

def clear_result():
    return gr.update(value=None)

def process_inpainting(image, mask_input, prompt):
    """이미지 인페인팅 처리 함수"""
    try:
        if image is None or mask_input is None or not prompt:
            raise gr.Error("Please provide image, mask, and prompt")
            
        # 프롬프트 번역 (한글인 경우)
        translated_prompt = translate_if_korean(prompt)
        
        # 마스크 처리
        source = image
        if isinstance(mask_input, dict):
            mask = mask_input["layers"][0]
            alpha_channel = mask.split()[3]
            binary_mask = alpha_channel.point(lambda p: p > 0 and 255)
        else:
            raise gr.Error("Invalid mask input")

        # 인페인팅을 위한 이미지 준비
        cnet_image = source.copy()
        cnet_image.paste(0, (0, 0), binary_mask)

        # 프롬프트 임베딩
        (
            prompt_embeds,
            negative_prompt_embeds,
            pooled_prompt_embeds,
            negative_pooled_prompt_embeds,
        ) = pipe.encode_prompt(translated_prompt, "cuda", True)

        # 인페인팅 실행
        result = None
        for image in pipe(
            prompt_embeds=prompt_embeds,
            negative_prompt_embeds=negative_prompt_embeds,
            pooled_prompt_embeds=pooled_prompt_embeds,
            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
            image=cnet_image,
        ):
            result = image

        if result is None:
            raise gr.Error("Inpainting failed")

        # 결과 이미지 처리
        result = result.convert("RGBA")
        cnet_image.paste(result, (0, 0), binary_mask)
        
        return cnet_image

    except Exception as e:
        print(f"Inpainting error: {str(e)}")
        raise gr.Error(f"Inpainting failed: {str(e)}")
    finally:
        clear_memory()
        
#--------------- 이미지 인페인팅 끝 ----------------    

class timer:
    def __init__(self, method_name="timed process"):
        self.method = method_name
    def __enter__(self):
        self.start = time.time()
        print(f"{self.method} starts")
    def __exit__(self, exc_type, exc_val, exc_tb):
        end = time.time()
        print(f"{self.method} took {str(round(end - self.start, 2))}s")

def bbox_union(bboxes: Sequence[list[int]]) -> BoundingBox | None:
    if not bboxes:
        return None
    for bbox in bboxes:
        assert len(bbox) == 4
        assert all(isinstance(x, int) for x in bbox)
    return (
        min(bbox[0] for bbox in bboxes),
        min(bbox[1] for bbox in bboxes),
        max(bbox[2] for bbox in bboxes),
        max(bbox[3] for bbox in bboxes),
    )

def corners_to_pixels_format(bboxes: torch.Tensor, width: int, height: int) -> torch.Tensor:
    x1, y1, x2, y2 = bboxes.round().to(torch.int32).unbind(-1)
    return torch.stack((x1.clamp_(0, width), y1.clamp_(0, height), x2.clamp_(0, width), y2.clamp_(0, height)), dim=-1)

def gd_detect(img: Image.Image, prompt: str) -> BoundingBox | None:
    inputs = gd_processor(images=img, text=f"{prompt}.", return_tensors="pt").to(device=device)
    with no_grad():
        outputs = gd_model(**inputs)
    width, height = img.size
    results: dict[str, Any] = gd_processor.post_process_grounded_object_detection(
        outputs,
        inputs["input_ids"],
        target_sizes=[(height, width)],
    )[0]
    assert "boxes" in results and isinstance(results["boxes"], torch.Tensor)
    bboxes = corners_to_pixels_format(results["boxes"].cpu(), width, height)
    return bbox_union(bboxes.numpy().tolist())

def apply_mask(img: Image.Image, mask_img: Image.Image, defringe: bool = True) -> Image.Image:
    assert img.size == mask_img.size
    img = img.convert("RGB")
    mask_img = mask_img.convert("L")
    if defringe:
        rgb, alpha = np.asarray(img) / 255.0, np.asarray(mask_img) / 255.0
        foreground = cast(np.ndarray[Any, np.dtype[np.uint8]], estimate_foreground_ml(rgb, alpha))
        img = Image.fromarray((foreground * 255).astype("uint8"))
    result = Image.new("RGBA", img.size)
    result.paste(img, (0, 0), mask_img)
    return result


def adjust_size_to_multiple_of_8(width: int, height: int) -> tuple[int, int]:
    """이미지 크기를 8의 배수로 조정하는 함수"""
    new_width = ((width + 7) // 8) * 8
    new_height = ((height + 7) // 8) * 8
    return new_width, new_height

def calculate_dimensions(aspect_ratio: str, base_size: int = 512) -> tuple[int, int]:
    """선택된 비율에 따라 이미지 크기 계산"""
    if aspect_ratio == "1:1":
        return base_size, base_size
    elif aspect_ratio == "16:9":
        return base_size * 16 // 9, base_size
    elif aspect_ratio == "9:16":
        return base_size, base_size * 16 // 9
    elif aspect_ratio == "4:3":
        return base_size * 4 // 3, base_size
    return base_size, base_size

@spaces.GPU(duration=20)  # 40초에서 20초로 감소
def generate_background(prompt: str, aspect_ratio: str) -> Image.Image:
    try:
        width, height = calculate_dimensions(aspect_ratio)
        width, height = adjust_size_to_multiple_of_8(width, height)
        
        max_size = 768
        if width > max_size or height > max_size:
            ratio = max_size / max(width, height)
            width = int(width * ratio)
            height = int(height * ratio)
            width, height = adjust_size_to_multiple_of_8(width, height)

        with timer("Background generation"):
            try:
                with torch.inference_mode():
                    image = pipe(
                        prompt=prompt,
                        width=width,
                        height=height,
                        num_inference_steps=8,
                        guidance_scale=4.0
                    ).images[0]
            except Exception as e:
                print(f"Pipeline error: {str(e)}")
                return Image.new('RGB', (width, height), 'white')
                
        return image
    except Exception as e:
        print(f"Background generation error: {str(e)}")
        return Image.new('RGB', (512, 512), 'white')

def create_position_grid():
    return """
    <div class="position-grid" style="display: grid; grid-template-columns: repeat(3, 1fr); gap: 10px; width: 150px; margin: auto;">
        <button class="position-btn" data-pos="top-left">↖</button>
        <button class="position-btn" data-pos="top-center">↑</button>
        <button class="position-btn" data-pos="top-right">↗</button>
        <button class="position-btn" data-pos="middle-left">←</button>
        <button class="position-btn" data-pos="middle-center">•</button>
        <button class="position-btn" data-pos="middle-right">→</button>
        <button class="position-btn" data-pos="bottom-left">↙</button>
        <button class="position-btn" data-pos="bottom-center" data-default="true">↓</button>
        <button class="position-btn" data-pos="bottom-right">↘</button>
    </div>
    """

def calculate_object_position(position: str, bg_size: tuple[int, int], obj_size: tuple[int, int]) -> tuple[int, int]:
    """오브젝트의 위치 계산"""
    bg_width, bg_height = bg_size
    obj_width, obj_height = obj_size
    
    positions = {
        "top-left": (0, 0),
        "top-center": ((bg_width - obj_width) // 2, 0),
        "top-right": (bg_width - obj_width, 0),
        "middle-left": (0, (bg_height - obj_height) // 2),
        "middle-center": ((bg_width - obj_width) // 2, (bg_height - obj_height) // 2),
        "middle-right": (bg_width - obj_width, (bg_height - obj_height) // 2),
        "bottom-left": (0, bg_height - obj_height),
        "bottom-center": ((bg_width - obj_width) // 2, bg_height - obj_height),
        "bottom-right": (bg_width - obj_width, bg_height - obj_height)
    }
    
    return positions.get(position, positions["bottom-center"])

def resize_object(image: Image.Image, scale_percent: float) -> Image.Image:
    """오브젝트 크기 조정"""
    width = int(image.width * scale_percent / 100)
    height = int(image.height * scale_percent / 100)
    return image.resize((width, height), Image.Resampling.LANCZOS)

def combine_with_background(foreground: Image.Image, background: Image.Image, 
                          position: str = "bottom-center", scale_percent: float = 100) -> Image.Image:
    """전경과 배경 합성 함수"""
    print(f"Combining with position: {position}, scale: {scale_percent}")
    
    result = background.convert('RGBA')
    scaled_foreground = resize_object(foreground, scale_percent)
    
    x, y = calculate_object_position(position, result.size, scaled_foreground.size)
    print(f"Calculated position coordinates: ({x}, {y})")
    
    result.paste(scaled_foreground, (x, y), scaled_foreground)
    return result

@spaces.GPU(duration=30)  # 120초에서 30초로 감소
def _gpu_process(img: Image.Image, prompt: str | BoundingBox | None) -> tuple[Image.Image, BoundingBox | None, list[str]]:
    time_log: list[str] = []
    try:
        if isinstance(prompt, str):
            t0 = time.time()
            bbox = gd_detect(img, prompt)
            time_log.append(f"detect: {time.time() - t0}")
            if not bbox:
                print(time_log[0])
                raise gr.Error("No object detected")
        else:
            bbox = prompt
        t0 = time.time()
        mask = segmenter(img, bbox)
        time_log.append(f"segment: {time.time() - t0}")
        return mask, bbox, time_log
    except Exception as e:
        print(f"GPU process error: {str(e)}")
        raise

def _process(img: Image.Image, prompt: str | BoundingBox | None, bg_prompt: str | None = None, aspect_ratio: str = "1:1") -> tuple[tuple[Image.Image, Image.Image, Image.Image], gr.DownloadButton]:
    try:
        # 입력 이미지 크기 제한
        max_size = 1024
        if img.width > max_size or img.height > max_size:
            ratio = max_size / max(img.width, img.height)
            new_size = (int(img.width * ratio), int(img.height * ratio))
            img = img.resize(new_size, Image.LANCZOS)

        # CUDA 메모리 관리 수정
        try:
            if torch.cuda.is_available():
                current_device = torch.cuda.current_device()
                with torch.cuda.device(current_device):
                    torch.cuda.empty_cache()
        except Exception as e:
            print(f"CUDA memory management failed: {e}")

        with torch.cuda.amp.autocast(enabled=torch.cuda.is_available()):
            mask, bbox, time_log = _gpu_process(img, prompt)
            masked_alpha = apply_mask(img, mask, defringe=True)

        if bg_prompt:
            background = generate_background(bg_prompt, aspect_ratio)
            combined = background
        else:
            combined = Image.alpha_composite(Image.new("RGBA", masked_alpha.size, "white"), masked_alpha)

        clear_memory()

        with tempfile.NamedTemporaryFile(delete=False, suffix=".png") as temp:
            combined.save(temp.name)
            return (img, combined, masked_alpha), gr.DownloadButton(value=temp.name, interactive=True)
    except Exception as e:
        clear_memory()
        print(f"Processing error: {str(e)}")
        raise gr.Error(f"Processing failed: {str(e)}")

def on_change_bbox(prompts: dict[str, Any] | None):
    return gr.update(interactive=prompts is not None)


def on_change_prompt(img: Image.Image | None, prompt: str | None, bg_prompt: str | None = None):
    return gr.update(interactive=bool(img and prompt))


def process_prompt(img: Image.Image, prompt: str, bg_prompt: str | None = None, 
                  aspect_ratio: str = "1:1", position: str = "bottom-center", 
                  scale_percent: float = 100) -> tuple[Image.Image, Image.Image]:
    try:
        if img is None or prompt.strip() == "":
            raise gr.Error("Please provide both image and prompt")
        
        print(f"Processing with position: {position}, scale: {scale_percent}")  # 디버깅용
        
        try:
            prompt = translate_to_english(prompt)
            if bg_prompt:
                bg_prompt = translate_to_english(bg_prompt)
        except Exception as e:
            print(f"Translation error (continuing with original text): {str(e)}")
        
        results, _ = _process(img, prompt, bg_prompt, aspect_ratio)
        
        if bg_prompt:
            try:
                print(f"Using position: {position}")  # 디버깅용
                # 위치 값 검증
                valid_positions = ["top-left", "top-center", "top-right",
                                 "middle-left", "middle-center", "middle-right",
                                 "bottom-left", "bottom-center", "bottom-right"]
                if position not in valid_positions:
                    position = "bottom-center"
                    print(f"Invalid position, using default: {position}")
                
                combined = combine_with_background(
                    foreground=results[2],
                    background=results[1],
                    position=position,
                    scale_percent=scale_percent
                )
                return combined, results[2]
            except Exception as e:
                print(f"Combination error: {str(e)}")
                return results[1], results[2]
        
        return results[1], results[2]  # 기본 반환 추가
    except Exception as e:
        print(f"Error in process_prompt: {str(e)}")
        raise gr.Error(str(e))
    finally:
        clear_memory()

    
def process_bbox(img: Image.Image, box_input: str) -> tuple[Image.Image, Image.Image]:
    try:
        if img is None or box_input.strip() == "":
            raise gr.Error("Please provide both image and bounding box coordinates")
        
        try:
            coords = eval(box_input)
            if not isinstance(coords, list) or len(coords) != 4:
                raise ValueError("Invalid box format")
            bbox = tuple(int(x) for x in coords)
        except:
            raise gr.Error("Invalid box format. Please provide [xmin, ymin, xmax, ymax]")
        
        # Process the image
        results, _ = _process(img, bbox)
        
        # 합성된 이미지와 추출된 이미지만 반환
        return results[1], results[2]
    except Exception as e:
        raise gr.Error(str(e))

# Event handler functions 수정
def update_process_button(img, prompt):
    return gr.update(
        interactive=bool(img and prompt),
        variant="primary" if bool(img and prompt) else "secondary"
    )

def update_box_button(img, box_input):
    try:
        if img and box_input:
            coords = eval(box_input)
            if isinstance(coords, list) and len(coords) == 4:
                return gr.update(interactive=True, variant="primary")
        return gr.update(interactive=False, variant="secondary")
    except:
        return gr.update(interactive=False, variant="secondary")


css = """
/* 기본 레이아웃 */
footer {display: none !important}
body {background: #f5f7fa !important}

/* 메인 타이틀 */
.main-title {
    text-align: center;
    margin: 1.5em auto;
    padding: 2em;
    background: linear-gradient(135deg, #f5f7fa 0%, #c3cfe2 100%);
    border-radius: 15px;
    box-shadow: 0 8px 16px rgba(0,0,0,0.1);
    max-width: 1200px;
}

.main-title h1 {
    color: #2196F3;
    font-size: 3em;
    margin-bottom: 0.5em;
    font-weight: 700;
    text-shadow: 2px 2px 4px rgba(0,0,0,0.1);
}

.main-title p {
    color: #555;
    font-size: 1.4em;
    line-height: 1.6;
    max-width: 800px;
    margin: 0 auto;
}

/* 패널 스타일링 */
.input-panel, .output-panel {
    background: white;
    padding: 2em;
    border-radius: 15px;
    box-shadow: 0 4px 12px rgba(0,0,0,0.05);
    margin-bottom: 1.5em;
    transition: all 0.3s ease;
}

.input-panel:hover, .output-panel:hover {
    box-shadow: 0 6px 16px rgba(0,0,0,0.1);
}

/* 컨트롤 패널 */
.controls-panel {
    background: #f8f9fa;
    padding: 1.5em;
    border-radius: 12px;
    margin: 1.5em 0;
    border: 1px solid #e9ecef;
}

/* 이미지 디스플레이 */
.image-display {
    min-height: 512px;
    display: flex;
    align-items: center;
    justify-content: center;
    background: #fafafa;
    border-radius: 12px;
    margin: 1.5em 0;
    border: 2px dashed #e0e0e0;
}

/* 버튼 스타일링 */
.position-btn {
    padding: 12px;
    border: 2px solid #ddd;
    border-radius: 8px;
    background: white;
    cursor: pointer;
    transition: all 0.2s ease;
    width: 48px;
    height: 48px;
    display: flex;
    align-items: center;
    justify-content: center;
    font-size: 1.2em;
    margin: 4px;
}

.position-btn:hover {
    background: #e3f2fd;
    transform: translateY(-2px);
    box-shadow: 0 4px 8px rgba(0,0,0,0.1);
}

.position-btn.selected {
    background-color: #2196F3;
    color: white;
    border-color: #1976D2;
    box-shadow: 0 4px 12px rgba(33,150,243,0.3);
}

/* 그리드 레이아웃 */
.position-grid {
    display: grid;
    grid-template-columns: repeat(3, 1fr);
    gap: 10px;
    margin: 1.5em 0;
    padding: 10px;
    background: #f5f5f5;
    border-radius: 12px;
}

/* 입력 필드 스타일링 */
input[type="text"], textarea {
    border: 2px solid #e0e0e0;
    border-radius: 8px;
    padding: 12px;
    font-size: 1.1em;
    transition: all 0.3s ease;
}

input[type="text"]:focus, textarea:focus {
    border-color: #2196F3;
    box-shadow: 0 0 0 3px rgba(33,150,243,0.2);
}

/* 슬라이더 스타일링 */
.slider-container {
    margin: 1.5em 0;
}

.slider {
    height: 6px;
    background: #e0e0e0;
    border-radius: 3px;
}

.slider-handle {
    width: 20px;
    height: 20px;
    background: #2196F3;
    border: 2px solid white;
    box-shadow: 0 2px 4px rgba(0,0,0,0.2);
}

/* 상태 메시지 */
.status-message {
    padding: 10px;
    border-radius: 8px;
    margin: 10px 0;
    font-size: 0.9em;
    transition: all 0.3s ease;
}

.status-success {
    background: #e8f5e9;
    color: #2e7d32;
    border: 1px solid #a5d6a7;
}

.status-error {
    background: #ffebee;
    color: #c62828;
    border: 1px solid #ef9a9a;
}

/* 반응형 디자인 */
@media (max-width: 768px) {
    .main-title h1 {
        font-size: 2em;
    }
    
    .main-title p {
        font-size: 1.1em;
    }
    
    .input-panel, .output-panel {
        padding: 1em;
    }
    
    .position-btn {
        width: 40px;
        height: 40px;
        font-size: 1em;
    }
}

/* 애니메이션 효과 */
@keyframes fadeIn {
    from {opacity: 0; transform: translateY(10px);}
    to {opacity: 1; transform: translateY(0);}
}

.fade-in {
    animation: fadeIn 0.3s ease-out;
}
"""


def add_text_with_stroke(draw, text, x, y, font, text_color, stroke_width):
    """Helper function to draw text with stroke"""
    # Draw the stroke/outline
    for adj_x in range(-stroke_width, stroke_width + 1):
        for adj_y in range(-stroke_width, stroke_width + 1):
            draw.text((x + adj_x, y + adj_y), text, font=font, fill=text_color)

def remove_background(image):
    # Save the image to a specific location
    filename = f"image_{uuid.uuid4()}.png"  # Generates a universally unique identifier (UUID) for the filename
    image.save(filename)
    # Call gradio client for background removal
    result = client.predict(images=handle_file(filename), api_name="/image")
    return Image.open(result[0])

def superimpose(image_with_text, overlay_image):
    # Open image as RGBA to handle transparency
    overlay_image = overlay_image.convert("RGBA")
    # Paste overlay on the background
    image_with_text.paste(overlay_image, (0, 0), overlay_image)
    # Save the final image
    # image_with_text.save("output_image.png")
    return image_with_text

def add_text_to_image(
    input_image,
    text,
    font_size,
    color,
    opacity,
    x_position,
    y_position,
    thickness,
    text_position_type,
    font_choice
):
    try:
        if input_image is None or text.strip() == "":
            return input_image

        # PIL Image 객체로 변환
        if not isinstance(input_image, Image.Image):
            if isinstance(input_image, np.ndarray):
                image = Image.fromarray(input_image)
            else:
                raise ValueError("Unsupported image type")
        else:
            image = input_image.copy()

        # 이미지를 RGBA 모드로 변환
        if image.mode != 'RGBA':
            image = image.convert('RGBA')

        # 폰트 설정
        font_files = {
            "Default": "DejaVuSans.ttf",
            "Korean Regular": "ko-Regular.ttf"
        }
        
        try:
            font_file = font_files.get(font_choice, "DejaVuSans.ttf")
            font = ImageFont.truetype(font_file, int(font_size))
        except Exception as e:
            print(f"Font loading error ({font_choice}): {str(e)}")
            font = ImageFont.load_default()
        
        # 색상 설정
        color_map = {
            'White': (255, 255, 255),
            'Black': (0, 0, 0),
            'Red': (255, 0, 0),
            'Green': (0, 255, 0),
            'Blue': (0, 0, 255),
            'Yellow': (255, 255, 0),
            'Purple': (128, 0, 128)
        }
        rgb_color = color_map.get(color, (255, 255, 255))
        
        # 임시 Draw 객체 생성하여 텍스트 크기 계산
        temp_draw = ImageDraw.Draw(image)
        text_bbox = temp_draw.textbbox((0, 0), text, font=font)
        text_width = text_bbox[2] - text_bbox[0]
        text_height = text_bbox[3] - text_bbox[1]

        # 위치 계산
        actual_x = int((image.width - text_width) * (x_position / 100))
        actual_y = int((image.height - text_height) * (y_position / 100))

        # 텍스트 색상 설정
        text_color = (*rgb_color, int(opacity))
        
        if text_position_type == "Text Behind Image":
            try:
                # 원본 이미지에서 전경 객체만 추출
                foreground = remove_background(image)
                
                # 배경 이미지 생성 (원본 이미지 복사)
                background = image.copy()
                
                # 텍스트를 그릴 임시 레이어 생성
                text_layer = Image.new('RGBA', image.size, (255, 255, 255, 0))
                draw_text = ImageDraw.Draw(text_layer)
                
                # 텍스트 그리기
                add_text_with_stroke(
                    draw_text,
                    text,
                    actual_x,
                    actual_y,
                    font,
                    text_color,
                    int(thickness)
                )
                
                # 배경에 텍스트 합성
                background = Image.alpha_composite(background, text_layer)
                
                # 텍스트가 있는 배경 위에 전경 객체 합성
                output_image = Image.alpha_composite(background, foreground)
            except Exception as e:
                print(f"Error in Text Behind Image processing: {str(e)}")
                return input_image
        else:
            # 텍스트 오버레이 생성
            txt_overlay = Image.new('RGBA', image.size, (255, 255, 255, 0))
            draw = ImageDraw.Draw(txt_overlay)
            
            # 텍스트를 이미지 위에 그리기
            add_text_with_stroke(
                draw,
                text,
                actual_x,
                actual_y,
                font,
                text_color,
                int(thickness)
            )
            output_image = Image.alpha_composite(image, txt_overlay)
        
        # RGB로 변환
        output_image = output_image.convert('RGB')
        
        return output_image

    except Exception as e:
        print(f"Error in add_text_to_image: {str(e)}")
        return input_image

    
def update_position(new_position):
    """위치 업데이트 함수"""
    print(f"Position updated to: {new_position}")
    return new_position

def update_position_and_ui(pos):
    """위치 업데이트 및 UI 반영"""
    updates = {btn: gr.update(value="selected" if pos_val == pos else "") 
              for btn, pos_val in position_mapping.items()}
    updates['position'] = pos
    return [pos] + [updates[btn] for btn in position_mapping.keys()]
    
def process_inpainting_with_feedback(image, mask, prompt):
    """인페인팅 처리 및 피드백"""
    try:
        result = process_inpainting(image, mask, prompt)
        return result, update_ui_state("inpainting", "Inpainting completed successfully!")
    except Exception as e:
        return None, update_ui_state("inpainting", f"Error: {str(e)}", is_error=True)
        
def update_controls(bg_prompt):
    """배경 프롬프트 입력 여부에 따라 컨트롤 표시 업데이트"""
    is_visible = bool(bg_prompt)
    return [
        gr.update(visible=is_visible),  # aspect_ratio
        gr.update(visible=is_visible),  # object_controls
    ]

with gr.Blocks(theme=gr.themes.Soft(), css=css) as demo:
    position = gr.State(value="bottom-center")
    processing_status = gr.State(value="idle")
    
    gr.HTML("""
        <div class="main-title">
            <h1>🎨 GiniGen Canvas-o3</h1>
            <p>Remove background of specified objects, generate new backgrounds, and insert text over or behind images with prompts.</p>
        </div>
    """)
    
    status_message = gr.HTML(
        value='<div class="status-message"></div>',
        visible=False
    )

    
    with gr.Row(equal_height=True):
        # 왼쪽 패널 (입력)
        with gr.Column(scale=1):
            with gr.Group(elem_classes="input-panel"):
                with gr.Tabs():
                    # 첫 번째 탭: 이미지 업로드 및 인페인팅
                    with gr.Tab("Image Upload & Inpainting"):
                        input_image = gr.Image(
                            type="pil",
                            label="Upload Image",
                            interactive=True,
                            height=400,
                            elem_classes="fade-in"
                        )

                        with gr.Group():
                            inpaint_prompt = gr.Textbox(
                                label="Inpainting Prompt",
                                placeholder="Describe what you want to add in the masked area..."
                            )
                            mask_input = image_annotator(
                                label="Draw mask for inpainting",
                                height=400
                            )
                            inpaint_btn = gr.Button("Apply Inpainting", variant="primary")

                    # 두 번째 탭: 배경 제거 및 생성
                    with gr.Tab("Background Removal"):
                        text_prompt = gr.Textbox(
                            label="Object to Extract",
                            placeholder="Enter what you want to extract...",
                            interactive=True,
                            elem_classes="fade-in"
                        )

                        with gr.Row():
                            bg_prompt = gr.Textbox(
                                label="Background Prompt (optional)",
                                placeholder="Describe the background...",
                                interactive=True,
                                scale=3
                            )
                            aspect_ratio = gr.Dropdown(
                                choices=["1:1", "16:9", "9:16", "4:3"],
                                value="1:1",
                                label="Aspect Ratio",
                                interactive=True,
                                visible=True,
                                scale=1
                            )

                        with gr.Group(elem_classes="controls-panel", visible=False) as object_controls:
                            with gr.Column(scale=1):
                                with gr.Row():
                                    btn_top_left = gr.Button("↖", elem_classes="position-btn")
                                    btn_top_center = gr.Button("↑", elem_classes="position-btn")
                                    btn_top_right = gr.Button("↗", elem_classes="position-btn")
                                with gr.Row():
                                    btn_middle_left = gr.Button("←", elem_classes="position-btn")
                                    btn_middle_center = gr.Button("•", elem_classes="position-btn")
                                    btn_middle_right = gr.Button("→", elem_classes="position-btn")
                                with gr.Row():
                                    btn_bottom_left = gr.Button("↙", elem_classes="position-btn")
                                    btn_bottom_center = gr.Button("↓", elem_classes="position-btn", value="selected")
                                    btn_bottom_right = gr.Button("↘", elem_classes="position-btn")
                            with gr.Column(scale=1):
                                scale_slider = gr.Slider(
                                    minimum=10,
                                    maximum=200,
                                    value=50,
                                    step=5,
                                    label="Object Size (%)"
                                )

                        process_btn = gr.Button(
                            "Process",
                            variant="primary",
                            interactive=False,
                            size="lg"
                        )

        # 오른쪽 패널 (출력)
        with gr.Column(scale=1):
            with gr.Group(elem_classes="output-panel"):
                with gr.Tab("Result"):
                    combined_image = gr.Image(
                        label="Combined Result",
                        show_download_button=True,
                        type="pil",
                        height=400
                    )
                    
                    with gr.Accordion("Text Insertion Options", open=False):
                        with gr.Group():
                            with gr.Row():
                                text_input = gr.Textbox(
                                    label="Text Content",
                                    placeholder="Enter text to add..."
                                )
                                text_position_type = gr.Radio(
                                    choices=["Text Over Image", "Text Behind Image"],
                                    value="Text Over Image",
                                    label="Text Position"
                                )
                            
                            with gr.Row():
                                with gr.Column(scale=1):
                                    font_choice = gr.Dropdown(
                                        choices=["Default", "Korean Regular"],
                                        value="Default",
                                        label="Font Selection",
                                        interactive=True
                                    )
                                    font_size = gr.Slider(
                                        minimum=10,
                                        maximum=200,
                                        value=40,
                                        step=5,
                                        label="Font Size"
                                    )
                                    color_dropdown = gr.Dropdown(
                                        choices=["White", "Black", "Red", "Green", "Blue", "Yellow", "Purple"],
                                        value="White",
                                        label="Text Color"
                                    )
                                    thickness = gr.Slider(
                                        minimum=0,
                                        maximum=10,
                                        value=1,
                                        step=1,
                                        label="Text Thickness"
                                    )
                                with gr.Column(scale=1):
                                    opacity_slider = gr.Slider(
                                        minimum=0,
                                        maximum=255,
                                        value=255,
                                        step=1,
                                        label="Opacity"
                                    )
                                    x_position = gr.Slider(
                                        minimum=0,
                                        maximum=100,
                                        value=50,
                                        step=1,
                                        label="Left(0%)~Right(100%)"
                                    )
                                    y_position = gr.Slider(
                                        minimum=0,
                                        maximum=100,
                                        value=50,
                                        step=1,
                                        label="High(0%)~Low(100%)"
                                    )
                            add_text_btn = gr.Button("Apply Text", variant="primary")

                    extracted_image = gr.Image(
                        label="Extracted Object",
                        show_download_button=True,
                        type="pil",
                        height=200
                    )

    # CSS 스타일
    gr.HTML("""
        <style>
        .position-btn.selected {
            background-color: #2196F3 !important;
            color: white !important;
        }
        </style>
    """)

    # 이벤트 바인딩
    position_mapping = {
        btn_top_left: "top-left",
        btn_top_center: "top-center",
        btn_top_right: "top-right",
        btn_middle_left: "middle-left",
        btn_middle_center: "middle-center",
        btn_middle_right: "middle-right",
        btn_bottom_left: "bottom-left",
        btn_bottom_center: "bottom-center",
        btn_bottom_right: "bottom-right"
    }

    def update_ui_state(component_id, value, is_error=False):
        """UI 상태 업데이트 유틸리티"""
        class_name = "status-error" if is_error else "status-success"
        return gr.update(
            value=f'<div class="status-message {class_name}">{value}</div>',
            visible=True
        )

# 위치 버튼 이벤트 바인딩
    for btn, pos in position_mapping.items():
        btn.click(
            fn=lambda p=pos: update_position_and_ui(p),
            outputs=[position] + list(position_mapping.keys())
        )

# 인페인팅 버튼 이벤트
    inpaint_btn.click(
        fn=process_inpainting_with_feedback,
        inputs=[input_image, mask_input, inpaint_prompt],
        outputs=[input_image, status_message]
    )

# 프로세스 버튼 이벤트
    process_btn.click(
        fn=process_prompt,
        inputs=[
            input_image,
            text_prompt,
            bg_prompt,
            aspect_ratio,
            position,
            scale_slider
        ],
        outputs=[combined_image, extracted_image]
    )


    
    for btn, pos in position_mapping.items():
        btn.click(
            fn=lambda pos=pos: update_position(pos),
            outputs=position
        )


    bg_prompt.change(
        fn=update_controls,
        inputs=bg_prompt,
        outputs=[aspect_ratio, object_controls],
        queue=False
    )
    
    input_image.change(
        fn=update_process_button,
        inputs=[input_image, text_prompt],
        outputs=process_btn,
        queue=False
    )
    
    text_prompt.change(
        fn=update_process_button,
        inputs=[input_image, text_prompt],
        outputs=process_btn,
        queue=False
    )



    add_text_btn.click(
        fn=add_text_to_image,
        inputs=[
            combined_image,
            text_input,
            font_size,
            color_dropdown,
            opacity_slider,
            x_position,
            y_position,
            thickness,
            text_position_type,
            font_choice
        ],
        outputs=combined_image,
        api_name="add_text"
    )

demo.queue(max_size=5)
demo.launch(
    server_name="0.0.0.0",
    server_port=7860,
    share=False,
    max_threads=2
)