app.py

import gradio as gr
from diffusers import StableDiffusionImg2ImgPipeline
import torch
from PIL import Image

# Load pretrained stable-diffusion model
pipe = StableDiffusionImg2ImgPipeline.from_pretrained(
    "runwayml/stable-diffusion-v1-5", 
    torch_dtype=torch.float16, 
    revision="fp16", 
    use_auth_token=True
).to("cuda" if torch.cuda.is_available() else "cpu")

def generate_thumbnail(prompt, input_image):
    # Resize image to 512x512
    image = input_image.resize((512, 512)).convert("RGB")
    
    # Generate image
    output = pipe(prompt=prompt, image=image, strength=0.75, guidance_scale=7.5)
    return output.images[0]

# Gradio UI
demo = gr.Interface(
    fn=generate_thumbnail,
    inputs=[
        gr.Textbox(label="Enter your prompt (e.g., NOOB vs PRO Minecraft style)"),
        gr.Image(type="pil", label="Upload Background Image"),
    ],
    outputs=gr.Image(label="Generated Thumbnail"),
    title="AI Thumbnail Generator",
    description="Upload an image and enter your text to create a Minecraft-style YouTube thumbnail"
)

demo.launch()