import gradio as gr
import torch
import numpy as np
import random
from diffusers import DiffusionPipeline

device = "cuda" if torch.cuda.is_available() else "cpu"
torch_dtype = torch.float16 if device == "cuda" else torch.float32

MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 1024

# 預設可選模型
available_models = [
    "digiplay/AM-mix1",
    "digiplay/pan04",
    "digiplay/2K"
]

def load_model(selected_model_id, custom_model_id):
    model_id = custom_model_id.strip() if custom_model_id.strip() else selected_model_id
    try:
        pipe = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch_dtype).to(device)
        return pipe, model_id, f"✅ Model '{model_id}' loaded successfully!"
    except Exception as e:
        return None, "", f"❌ Failed to load model: {e}"

def generate_image(pipe, prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps):
    if pipe is None:
        raise ValueError("No model loaded. Please load a model first.")
    
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    
    generator = torch.Generator().manual_seed(seed)
    image = pipe(
        prompt=prompt,
        negative_prompt=negative_prompt,
        width=width,
        height=height,
        guidance_scale=guidance_scale,
        num_inference_steps=num_inference_steps,
        generator=generator,
    ).images[0]

    return image, seed

with gr.Blocks(css="#container { max-width: 700px; margin: auto; }") as demo:
    gr.Markdown("## Text-to-Image Generator with Model Selector")

    pipe_state = gr.State(None)
    model_id_state = gr.State("")

    with gr.Column(elem_id="container"):
        gr.Markdown("### 1. Choose or Enter Model")
        with gr.Row():
            selected_model = gr.Dropdown(label="Choose a model", choices=available_models, value=available_models[0])
            custom_model = gr.Textbox(label="Or enter custom model ID", placeholder="e.g. runwayml/stable-diffusion-v1-5")

        load_button = gr.Button("Load Model")
        load_status = gr.Textbox(label="Model Load Status", interactive=False)

        load_button.click(
            fn=load_model,
            inputs=[selected_model, custom_model],
            outputs=[pipe_state, model_id_state, load_status]
        )

        gr.Markdown("### 2. Generate Image")
        prompt = gr.Textbox(label="Prompt", placeholder="e.g. A futuristic city at night")
        negative_prompt = gr.Textbox(label="Negative Prompt", placeholder="(optional)", value="", visible=True)
        with gr.Row():
            width = gr.Slider(256, MAX_IMAGE_SIZE, step=32, value=512, label="Width")
            height = gr.Slider(256, MAX_IMAGE_SIZE, step=32, value=512, label="Height")
        with gr.Row():
            guidance_scale = gr.Slider(0.0, 10.0, step=0.1, value=7.5, label="Guidance Scale")
            num_inference_steps = gr.Slider(1, 50, step=1, value=25, label="Inference Steps")
        with gr.Row():
            seed = gr.Slider(0, MAX_SEED, step=1, value=0, label="Seed")
            randomize_seed = gr.Checkbox(label="Randomize Seed", value=True)

        generate_button = gr.Button("Generate Image")
        output_image = gr.Image(label="Result")
        final_seed = gr.Number(label="Used Seed", precision=0)

        generate_button.click(
            fn=generate_image,
            inputs=[pipe_state, prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps],
            outputs=[output_image, final_seed]
        )

if __name__ == "__main__":
    demo.launch()