app.py

# @title 📸 Image Generation (multimodel + Gradio Web Interface)

# Install Dependencies
!pip install transformers torch diffusers accelerate invisible_watermark safetensors huggingface-hub gradio --quiet

from diffusers import DiffusionPipeline, StableDiffusionPipeline
import torch
from gradio.components import Gallery
from PIL import Image
from IPython.display import display
import os
from datetime import datetime
import gradio as gr

# Function to generate and display images
def generate_and_display_images(model_selection, scenery, style, height, width, num_images=2, n_steps=50, high_noise_frac=0.5, guidance_scale=2.6, negative_prompt="", seed=None):
    if seed is None or seed == '':
        seed = torch.randint(low=0, high=2**32, size=(1,)).item()
    else:
        try:
            seed = int(seed)
        except ValueError:
            return "Invalid seed value. Seed must be an integer."
    torch.manual_seed(seed)

    prompt = f"Scenery: {scenery}; Style: {style}"

    generated_images = []
    if model_selection == "dreamlike-art/dreamlike-photoreal-2.0":
        model = StableDiffusionPipeline.from_pretrained(model_selection, torch_dtype=torch.float16).to("cuda")
        for _ in range(num_images):
            image = model(prompt=prompt, num_inference_steps=n_steps, guidance_scale=guidance_scale, negative_prompt=negative_prompt, height=height, width=width).images[0]
            generated_images.append(image)
    else:
        base = DiffusionPipeline.from_pretrained(model_selection, torch_dtype=torch.float16, use_auth_token=True).to("cuda")
        for _ in range(num_images):
            if "refiner" in model_selection:
                image_latent = base(prompt=prompt, num_inference_steps=n_steps, denoising_end=high_noise_frac, output_type="latent").images
                image = image_latent[0]  # Placeholder for actual refiner step
            else:
                image = base(prompt=prompt, num_inference_steps=n_steps, guidance_scale=guidance_scale, negative_prompt=negative_prompt, height=height, width=width).images[0]
            generated_images.append(image)
    
    # Save images and return file paths for Gradio display
    file_paths = []
    for i, image in enumerate(generated_images):
        timestamp = datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
        filename = f"{seed}_{timestamp}_{i}.jpg"
        image.save(filename)
        file_paths.append(filename)

    return file_paths



# Define Gradio interface
iface = gr.Interface(
    fn=generate_and_display_images,
    inputs=[
        gr.components.Dropdown(choices=["stabilityai/sdxl-turbo", "stabilityai/stable-diffusion-xl-base-1.0", "runwayml/stable-diffusion-v1-5", "dreamlike-art/dreamlike-photoreal-2.0", "Kardbord/stable-diffusion-v1-5-unsafe"], label="Model Selection"),
        gr.components.Textbox(label="Scenery", placeholder="Describe the scenery you want in the image"),
        gr.components.Textbox(label="Style", placeholder="Describe the style of the image (e.g., photorealistic, liminal, dark)"),
        gr.components.Slider(minimum=1, maximum=2048, step=1, value=1024, label="Height"),
        gr.components.Slider(minimum=1, maximum=2048, step=1, value=576, label="Width"),
        gr.components.Number(value=10, label="Number of Images"),
        gr.components.Slider(minimum=0, maximum=60, step=1, value=30, label="Number of Inference Steps"),
        gr.components.Slider(minimum=0.0, maximum=1.0, step=0.01, value=0.16, label="High Noise Fraction"),
        gr.components.Slider(minimum=0.0, maximum=10.0, step=0.1, value=8, label="Guidance Scale"),
        gr.components.Textbox(value="", label="Negative Prompt"),
        gr.components.Textbox(value=None, label="Seed (Optional)")
    ],
    outputs=Gallery(label="Generated Images"),
    examples=[["dreamlike-art/dreamlike-photoreal-2.0", "scenery : melting flesh", "style : (((photorealistic))), liminal, cryptic, cinematic, highly detailed, sharp focus, dark, creepy, weirdcore", 1024, 576, 10, 30, 0.16, 8, "2D || naked || Low Quality || text logos || watermarks || signatures || out of frame || jpeg artifacts || ugly || poorly drawn || extra limbs || extra hands || extra feet || backwards limbs || extra fingers || extra toes || unrealistic, incorrect, bad anatomy || cut off body pieces || strange body positions || impossible body positioning || Mismatched eyes || cross eyed || crooked face || crooked lips || unclear || undefined || mutations || deformities || off center || poor_composition || duplicate faces, blurry, blurred, unclear, deformed anatomy, deformed face, crazy eyes, bad hands, deformed body", None]],
    title="Image Generation Tool",
    description="Generate images using various diffusion models."
)

# Launch the interface
iface.launch(share=True, debug=True)