import gradio as gr
import numpy as np
import spaces
import torch
import random
from peft import PeftModel
from diffusers import FluxControlPipeline, FluxTransformer2DModel
from image_gen_aux import DepthPreprocessor

MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 2048

# Initialize models without moving to CUDA yet - following working version
pipe = FluxControlPipeline.from_pretrained(
    "black-forest-labs/FLUX.1-Depth-dev", 
    torch_dtype=torch.bfloat16
)
pipe.enable_attention_slicing()  # Keep this as it's helpful
processor = DepthPreprocessor.from_pretrained("LiheYoung/depth-anything-large-hf")

@spaces.GPU
def load_lora(lora_path):
    if not lora_path.strip():
        return "Please provide a valid LoRA path"
    try:
        # Move to GPU within the wrapped function
        pipe.to("cuda")
        
        # Unload any existing LoRA weights first
        try:
            pipe.unload_lora_weights()
        except:
            pass
            
        # Load new LoRA weights
        pipe.load_lora_weights(lora_path)
        return f"Successfully loaded LoRA weights from {lora_path}"
    except Exception as e:
        return f"Error loading LoRA weights: {str(e)}"

@spaces.GPU
def unload_lora():
    try:
        pipe.to("cuda")
        pipe.unload_lora_weights()
        return "Successfully unloaded LoRA weights"
    except Exception as e:
        return f"Error unloading LoRA weights: {str(e)}"

@spaces.GPU
def infer(control_image, prompt, seed=42, randomize_seed=False, width=1024, height=1024, 
          guidance_scale=3.5, num_inference_steps=28, progress=gr.Progress(track_tqdm=True)):
    
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    
    try:
        # Move pipeline to GPU within the wrapped function
        pipe.to("cuda")
        
        # Process control image
        control_image = processor(control_image)[0].convert("RGB")
        
        # Generate image
        image = pipe(
            prompt=prompt,
            control_image=control_image,
            height=height,
            width=width,
            num_inference_steps=num_inference_steps,
            guidance_scale=guidance_scale,
            generator=torch.Generator("cuda").manual_seed(seed),
        ).images[0]
        
        return image, seed
    except Exception as e:
        return None, f"Error during inference: {str(e)}"

css = """
@keyframes gradientMove {
    0% { background-position: 0% 50%; }
    50% { background-position: 100% 50%; }
    100% { background-position: 0% 50%; }
}
body {
    background: black !important;
    margin: 0;
    min-height: 100vh;
}
body::before {
    content: '';
    position: fixed;
    top: 0;
    left: 0;
    right: 0;
    bottom: 0;
    z-index: -1;
    background: 
        linear-gradient(125deg, rgba(255,105,180,0.3), rgba(0,0,0,0.5)),
        url('data:image/svg+xml,<svg viewBox="0 0 200 200" xmlns="http://www.w3.org/2000/svg"><filter id="noise"><feTurbulence type="fractalNoise" baseFrequency="0.005" numOctaves="3" /><feColorMatrix type="saturate" values="0"/></filter><rect width="100%" height="100%" filter="url(%23noise)"/></svg>');
    filter: blur(70px);
    animation: gradientMove 15s ease infinite;
    background-size: 400% 400%;
    opacity: 0.8;
}
:root {
    --hot-pink: #FF69B4;
    --light-pink: #FFB6C6;
    --dark-pink: #FF1493;
}
#col-container {
    margin: 0 auto;
    max-width: 1200px;
    padding: 2rem;
    background: rgba(0, 0, 0, 0.85);
    border-radius: 15px;
    box-shadow: 0 0 20px rgba(255, 105, 180, 0.3);
    border: 2px solid var(--hot-pink);
    position: relative;
    z-index: 1;
}
.gr-box {
    background: var(--hot-pink) !important;
    border: 2px solid black !important;
    border-radius: 8px !important;
    box-shadow: 0 4px 6px rgba(0, 0, 0, 0.2) !important;
    transition: all 0.3s ease !important;
}
.gr-box:hover {
    box-shadow: 0 0 15px rgba(255, 255, 255, 0.3) !important;
}
.gr-button {
    background: var(--hot-pink) !important;
    border: 2px solid black !important;
    color: black !important;
    font-weight: 600 !important;
    transition: all 0.3s ease !important;
}
.gr-button:hover {
    background: var(--dark-pink) !important;
    box-shadow: 0 0 15px rgba(255, 255, 255, 0.5);
    transform: translateY(-2px);
}
.gr-input, .gr-input-label {
    background: var(--hot-pink) !important;
    border: 2px solid black !important;
    border-radius: 8px !important;
    color: black !important;
    transition: all 0.3s ease !important;
}
.gr-input::placeholder {
    color: rgba(0, 0, 0, 0.6) !important;
}
.gr-input:focus {
    box-shadow: 0 0 15px rgba(255, 255, 255, 0.3) !important;
}
.gr-form {
    gap: 1.5rem !important;
}
.gr-slider {
    accent-color: var(--hot-pink) !important;
}
.gr-slider-value {
    color: white !important;
}
.gr-checkbox {
    accent-color: var(--hot-pink) !important;
}
.gr-panel {
    background: var(--hot-pink) !important;
    border: 2px solid black !important;
    box-shadow: 0 4px 6px rgba(0, 0, 0, 0.2) !important;
}
.gr-accordion {
    border: 2px solid black !important;
    background: var(--hot-pink) !important;
    border-radius: 10px !important;
    margin-top: 1.5rem !important;
}
label, .gr-box label, .gr-accordion-title {
    color: black !important;
    font-weight: 600 !important;
}
.markdown {
    color: white !important;
}
.markdown a {
    color: var(--hot-pink) !important;
    text-decoration: none !important;
    transition: color 0.3s ease !important;
}
.markdown a:hover {
    color: var(--light-pink) !important;
}
.upload-box {
    border: 2px dashed var(--hot-pink) !important;
    background: rgba(0, 0, 0, 0.3) !important;
    transition: all 0.3s ease !important;
}
.upload-box:hover {
    border-color: var(--light-pink) !important;
    box-shadow: 0 0 15px rgba(255, 105, 180, 0.2) !important;
}
.generating {
    box-shadow: 0 0 20px rgba(255, 255, 255, 0.8) !important;
}
.progress-bar {
    background: var(--hot-pink) !important;
}
"""

with gr.Blocks(css=css) as demo:
    with gr.Column(elem_id="col-container"):
        gr.Markdown("""# FLUX.1 Depth [dev] with LoRA Support
(note: clone this repo and run on free gpu, this required hf subscription) 12B param rectified flow transformer structural conditioning tuned, guidance-distilled from [FLUX.1 [pro]](https://blackforestlabs.ai/)  
[[non-commercial license](https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md)] [[blog](https://blackforestlabs.ai/announcing-black-forest-labs/)] [[model](https://huggingface.co/black-forest-labs/FLUX.1-dev)]
        """)

        with gr.Row():
            lora_path = gr.Textbox(
                label="HuggingFace LoRA Path",
                placeholder="e.g., Borcherding/FLUX.1-dev-LoRA-AutumnSpringTrees",
                scale=3
            )
            load_lora_btn = gr.Button("Load LoRA", scale=1)
            unload_lora_btn = gr.Button("Unload LoRA", scale=1)
        
        lora_status = gr.Textbox(label="LoRA Status", interactive=False)

        with gr.Row():
            prompt = gr.Text(
                label="Prompt",
                show_label=True,
                max_lines=1,
                placeholder="Enter your prompt",
                container=True,
            )
            run_button = gr.Button("Run", scale=0)

        with gr.Row(equal_height=True):
            with gr.Column(scale=1):
                control_image = gr.Image(
                    label="Control Image",
                    type="pil",
                    elem_id="image-upload"
                )
            with gr.Column(scale=1):
                result = gr.Image(
                    label="Generated Result",
                    elem_id="result-image"
                )
        
        with gr.Accordion("Advanced Settings", open=False):
            with gr.Row():
                with gr.Column(scale=1):
                    seed = gr.Slider(
                        label="Seed",
                        minimum=0,
                        maximum=MAX_SEED,
                        step=1,
                        value=0,
                    )
                    randomize_seed = gr.Checkbox(
                        label="Randomize seed",
                        value=True
                    )
            
            with gr.Row():
                with gr.Column(scale=1):
                    width = gr.Slider(
                        label="Width",
                        minimum=256,
                        maximum=MAX_IMAGE_SIZE,
                        step=32,
                        value=1024,
                    )
                    height = gr.Slider(
                        label="Height",
                        minimum=256,
                        maximum=MAX_IMAGE_SIZE,
                        step=32,
                        value=1024,
                    )
                with gr.Column(scale=1):
                    guidance_scale = gr.Slider(
                        label="Guidance Scale",
                        minimum=1,
                        maximum=30,
                        step=0.5,
                        value=10,
                    )
                    num_inference_steps = gr.Slider(
                        label="Number of inference steps",
                        minimum=1,
                        maximum=50,
                        step=1,
                        value=28,
                    )

    load_lora_btn.click(
        fn=load_lora,
        inputs=[lora_path],
        outputs=[lora_status]
    )
    
    unload_lora_btn.click(
        fn=unload_lora,
        inputs=[],
        outputs=[lora_status]
    )

    gr.on(
        triggers=[run_button.click, prompt.submit],
        fn=infer,
        inputs=[control_image, prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps],
        outputs=[result, seed]
    )

demo.launch()