app.py

import os, gc
import gradio as gr
import numpy as np
import random
from transformers import CLIPTokenizer, CLIPFeatureExtractor
import spaces
from diffusers import DiffusionPipeline, FlowMatchEulerDiscreteScheduler
import torch
torch.cuda.empty_cache()

device = "cuda" if torch.cuda.is_available() else "cpu"
model_repo_id = "tensorart/stable-diffusion-3.5-large-TurboX"

if torch.cuda.is_available():
    torch_dtype = torch.float16
else:
    torch_dtype = torch.bfloat16

tokenizer = CLIPTokenizer.from_pretrained(
    "openai/clip-vit-base-patch32",  # or clip-vit-large if you prefer
    use_fast=True
)
feature_extractor = CLIPFeatureExtractor.from_pretrained(
    "openai/clip-vit-base-patch32"
)

# 3) Dispatch & load in FP16 with offloading
pipe = DiffusionPipeline.from_pretrained(
    model_repo_id,
    scheduler=FlowMatchEulerDiscreteScheduler.from_pretrained(
        model_repo_id,
        subfolder="scheduler",
        shift=5,
        use_safetensors=True
    ),
    tokenizer=tokenizer,
    feature_extractor=feature_extractor,
    torch_dtype=torch.bfloat16,   # load weights in half-precision
    use_safetensors=True
)

# 4) Memory savings hooks (all on your single GPU + CPU offload)
pipe.enable_attention_slicing()                     # slice big attention maps
pipe.vae.enable_slicing()                          # slice VAE decode
pipe.enable_xformers_memory_efficient_attention()   # if xformers is installed
pipe.enable_model_cpu_offload()                     # offload idle submodules to CPU

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

@spaces.GPU(duration=65)
def infer(
    prompt,
    negative_prompt="",
    seed=42,
    randomize_seed=False,
    width=1024,
    height=1024,
    guidance_scale=1.5,
    num_inference_steps=8,
    progress=gr.Progress(track_tqdm=True),
):
    full_prompt = "cartoon styled korean" + prompt

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

    generator = torch.Generator().manual_seed(seed)

    image = pipe(
        prompt=full_prompt,
        negative_prompt=negative_prompt,
        guidance_scale=guidance_scale,
        num_inference_steps=num_inference_steps,
        width=width,
        height=height,
        generator=generator,
    ).images[0]

    return image, seed

css = """
body {
    background: linear-gradient(135deg, #f9e2e6 0%, #e8f3fc 50%, #e2f9f2 100%);
    background-attachment: fixed;
    min-height: 100vh;
}

#col-container {
    margin: 0 auto;
    max-width: 640px;
    background-color: rgba(255, 255, 255, 0.85);
    border-radius: 16px;
    box-shadow: 0 8px 16px rgba(0, 0, 0, 0.1);
    padding: 24px;
    backdrop-filter: blur(10px);
}

.gradio-container {
    background: transparent !important;
}

.gr-button-primary {
    background: linear-gradient(90deg, #6b9dfc, #8c6bfc) !important;
    border: none !important;
    transition: all 0.3s ease;
}

.gr-button-primary:hover {
    transform: translateY(-2px);
    box-shadow: 0 5px 15px rgba(108, 99, 255, 0.3);
}

.gr-form {
    border-radius: 12px;
    background-color: rgba(255, 255, 255, 0.7);
}

.gr-accordion {
    border-radius: 12px;
    overflow: hidden;
}

h1 {
    background: linear-gradient(90deg, #6b9dfc, #8c6bfc);
    -webkit-background-clip: text;
    -webkit-text-fill-color: transparent;
    font-weight: 800;
}
"""

with gr.Blocks(theme="apriel", css=css) as demo:
    with gr.Column(elem_id="col-container"):
        with gr.Row():
            prompt = gr.Text(
                label="Prompt",
                show_label=False,
                max_lines=1,
                placeholder="Enter your prompt copied from the previous website",
                container=False,
            )

            run_button = gr.Button("Run", 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=512,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=1024, 
                )

                height = gr.Slider(
                    label="Height",
                    minimum=512,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=1024,
                )

            with gr.Row():
                guidance_scale = gr.Slider(
                    label="Guidance scale",
                    minimum=0.0,
                    maximum=7.5,
                    step=0.1,
                    value=1.5,
                )

                num_inference_steps = gr.Slider(
                    label="Number of inference steps",
                    minimum=1,
                    maximum=50,
                    step=1,
                    value=8, 
                )

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

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