app.py

import gradio as gr
from diffusers import AutoPipelineForText2Image
import torch
import random

if torch.backends.mps.is_available():
    device = "mps"
elif torch.cuda.is_available():
    device = "cuda"
else:
    device = "cpu"

torch_dtype = torch.float32 if device == "cpu" else torch.float16

model_repo_id = "stabilityai/sdxl-turbo"
pipe = AutoPipelineForText2Image.from_pretrained(model_repo_id, torch_dtype=torch_dtype, variant="fp16")
pipe = pipe.to(device)

def infer(prompt):
    num_inference_steps = 2
    guidance_scale = 0.0
    width = 512
    height = 512
    image = pipe(
        prompt=prompt,
        guidance_scale=guidance_scale,
        num_inference_steps=num_inference_steps,
        width=width,
        height=height,
    ).images[0]

    return image

# Lists for random prompt generation
subjects = [
    "a cute red panda", "a robot", "a steampunk owl", "a magical crystal", 
    "a floating island", "a cybernetic dolphin", "a time-traveling historian",
    "a ghost pianist", "a rainbow dragon", "a space explorer", "a majestic tiger",
    "a giant mushroom", "a neon cityscape", "a cosmic jellyfish", "a dreamy unicorn",
    "a celestial mermaid", "a mechanical spider", "a mystical forest", "a futuristic city",
    "a whimsical fairy", "a mythical phoenix", "a cosmic nebula", "a magical castle",
    "a cyberpunk samurai", "a time-traveling astronaut", "a crystal cavern",
]

contexts = [
    "in an ancient library", "during a neon sunset", "floating in space",
    "deep underwater", "in a floating market", "on top of a giant mushroom",
    "inside a clockwork machine", "in a crystal cave", "during a meteor shower",
    "in a bamboo forest", "in a futuristic cityscape", "on a magical island",
    "in a cyberpunk alleyway", "in a steampunk workshop",
    "in a post-apocalyptic wasteland", "in a dreamy cloud kingdom",
]

styles = [
    "cyberpunk style", "watercolor painting", "digital art", "oil painting",
    "studio ghibli style", "synthwave aesthetics", "detailed pencil sketch",
    "photorealistic", "vaporwave style", "low poly art", "lo-fi aesthetics",
    "grainy film photography", "comic book style", "minimalist design",
]

image_qualities = [
    "highly detailed", "8K resolution", "dramatic lighting",
    "professional photography", "cinematic composition", "hyperrealistic",
    "atmospheric", "stunning", "award-winning", "masterpiece",
    "low quality", "grainy", "blurry", "abstract", "surreal",
]

def generate_not_hot_dog_prompt():
    subject = random.choice(subjects)
    context = random.choice(contexts)
    style = random.choice(styles)
    quality = random.choice(image_qualities)
    return f"{subject} {context}, {style}, {quality}"

hot_dog_prompt = "A cinematic shot of a gourmet hotdog in extreme close-up, sitting in a rustic brioche bun, garnished with perfectly drizzled mustard and ketchup. The hotdog glistens under dramatic studio lighting with a slight overhead angle. Shallow depth of field with professional food photography styling. The background is dark and moody with subtle bokeh effects. Shot with a high-end DSLR camera, f/2.8 aperture, warm color grading reminiscent of a Wes Anderson film. Photorealistic quality, 8K resolution."


def generate_hot_dog():
    return infer(hot_dog_prompt)

def generate_not_hot_dog():
    return infer(generate_not_hot_dog_prompt())

with gr.Blocks(theme="compact") as demo:
    gr.Markdown("# What Do You Want To Create?")
    
    with gr.Row():
        hot_dog_btn = gr.Button("🌭 Hot Dog", variant="primary")
        not_hot_dog_btn = gr.Button("❌ Not Hot Dog", variant="primary")
    
    output_image = gr.Image(label="Generated Image", type="numpy")
    
    hot_dog_btn.click(
        fn=generate_hot_dog,
        inputs=[],
        outputs=output_image
    )
    
    not_hot_dog_btn.click(
        fn=generate_not_hot_dog,
        inputs=[],
        outputs=output_image
    )

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