app.py

import gradio as gr
import numpy as np
import random
from datetime import datetime

import torch
from diffusers import DiffusionPipeline
from optimum.intel.openvino import OVStableDiffusionPipeline

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

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

# Chọn mô hình từ dropdown
model_choices = {
    "SD‑Turbo (stabilityai/sd-turbo)": "stabilityai/sd-turbo",
    "Stable Diffusion 1.5 (runwayml/stable-diffusion-1.5)": "runwayml/stable-diffusion-1.5",
    "OpenVINO version (HARRY07979/sd-v1-5-openvino)": "HARRY07979/sd-v1-5-openvino",
}

# Biến toàn cục để lưu model đang dùng
current_model_id = None
pipe = None

# ---------------------------------------------------------
# Hàm load mô hình
def load_pipeline(model_id):
    print(f"[INFO] Loading model: {model_id}")
    if "openvino" in model_id.lower():
        # Mô hình OpenVINO dùng OVStableDiffusionPipeline
        pipe = OVStableDiffusionPipeline.from_pretrained(model_id)
        pipe.reshape(batch_size=1, height=512, width=512, num_images_per_prompt=1)
        pipe.compile()
    else:
        if torch.cuda.is_available():
            torch_dtype = torch.float16
        else:
            torch_dtype = torch.float32
        pipe = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch_dtype)
        pipe = pipe.to(device)
    return pipe

# ---------------------------------------------------------
# Hàm infer
def infer(
    prompt,
    negative_prompt,
    seed,
    randomize_seed,
    width,
    height,
    guidance_scale,
    num_inference_steps,
    model_selector,
):
    global pipe, current_model_id

    selected_model_id = model_choices[model_selector]

    # Nếu đổi mô hình → load lại
    if selected_model_id != current_model_id or pipe is None:
        pipe = load_pipeline(selected_model_id)
        current_model_id = selected_model_id

    if randomize_seed:
        seed = random.randint(0, MAX_SEED)

    # Thời gian bắt đầu
    t0 = datetime.now()

    # Gọi pipeline theo loại
    if "openvino" in selected_model_id.lower():
        image = pipe(
            prompt=prompt,
            negative_prompt=negative_prompt,
            guidance_scale=guidance_scale,
            num_inference_steps=num_inference_steps,
        ).images[0]
    else:
        generator = torch.Generator().manual_seed(seed)
        image = pipe(
            prompt=prompt,
            negative_prompt=negative_prompt,
            guidance_scale=guidance_scale,
            num_inference_steps=num_inference_steps,
            width=width,
            height=height,
            generator=generator,
        ).images[0]

    # Thời gian kết thúc
    t1 = datetime.now()
    delta = t1 - t0
    total_seconds = delta.total_seconds()
    days = delta.days
    hours, rem = divmod(delta.seconds, 3600)
    minutes, seconds = divmod(rem, 60)
    microsecs = delta.microseconds

    print(f"Start time: {t0.isoformat(sep=' ')}")
    print(f"End time  : {t1.isoformat(sep=' ')}")
    print(f"Elapsed   : {days}d {hours}h {minutes}m {seconds}s {microsecs}µs")
    print(f"Total time: {total_seconds:.3f} seconds")

    return image, seed

# ---------------------------------------------------------
css = """
#col-container {
    margin: 0 auto;
    max-width: 640px;
}
"""

with gr.Blocks(css=css) as demo:
    with gr.Column(elem_id="col-container"):
        gr.Markdown("# Text-to-Image Generator (Supports SD-Turbo / SD 1.5 / OpenVINO)")

        with gr.Row():
            prompt = gr.Text(
                label="Prompt",
                show_label=False,
                max_lines=1,
                placeholder="Enter your prompt here...",
                container=False,
            )
            run_button = gr.Button("Generate", scale=0, variant="primary")

        result = gr.Image(label="Result", show_label=False)

        with gr.Accordion("Advanced Settings", open=False):
            negative_prompt = gr.Text(
                label="Negative prompt",
                max_lines=1,
                placeholder="Enter a negative prompt",
            )
            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():
                width = gr.Slider(
                    label="Width", minimum=256, maximum=MAX_IMAGE_SIZE, step=32, value=512
                )
                height = gr.Slider(
                    label="Height", minimum=256, maximum=MAX_IMAGE_SIZE, step=32, value=512
                )

            with gr.Row():
                guidance_scale = gr.Slider(
                    label="Guidance scale", minimum=0.0, maximum=20.0, step=0.1, value=7.5
                )
                num_inference_steps = gr.Slider(
                    label="Inference steps", minimum=1, maximum=100, step=1, value=25
                )

            model_selector = gr.Dropdown(
                label="Select Model",
                choices=list(model_choices.keys()),
                value="SD‑Turbo (stabilityai/sd-turbo)",
            )

        gr.Examples(
            examples=[
                "Astronaut in a jungle, detailed, 8k",
                "A cyberpunk dragon flying through neon city",
                "A fantasy landscape with floating islands",
            ],
            inputs=[prompt],
        )

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

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