Spaces:
Running
on
Zero
Running
on
Zero
import gradio as gr | |
from diffusers import StableDiffusionPipeline, EulerAncestralDiscreteScheduler, DDIMScheduler, PNDMScheduler | |
import torch | |
from PIL import ImageEnhance, Image | |
import numpy as np | |
# Load Stable Diffusion pipeline | |
model_id = "CompVis/stable-diffusion-v1-4" | |
default_scheduler = EulerAncestralDiscreteScheduler.from_pretrained(model_id, subfolder="scheduler") | |
pipe = StableDiffusionPipeline.from_pretrained(model_id, scheduler=default_scheduler, torch_dtype=torch.float16) | |
device = "cuda" if torch.cuda.is_available() else "cpu" | |
pipe = pipe.to(device) | |
# Scheduler options | |
schedulers = { | |
"Artistic & Imaginative (Euler Ancestral) - Recommended for creative scenes, moderate speed": EulerAncestralDiscreteScheduler, | |
"Photo-Realistic (PNDM) - Best for realistic details, moderate speed": PNDMScheduler, | |
"High-Definition & Fast (DDIM) - Good quality with fastest speed": DDIMScheduler, | |
} | |
# Main image generation function with dynamic scheduling and size option | |
def generate_image(prompt, use_categories, genre, style, theme, lighting, scheduler_choice, quality, size): | |
# Check if additional categories should be added to the prompt | |
if use_categories: | |
prompt_text = ( | |
f"{prompt.strip()} in a {genre.lower()} wallpaper style, " | |
f"with {style.lower()} visuals, focusing on a {theme.lower()} theme " | |
f"and {lighting.lower()} lighting." | |
) | |
else: | |
prompt_text = prompt.strip() | |
# Set the scheduler based on user choice | |
scheduler = schedulers[scheduler_choice].from_pretrained(model_id, subfolder="scheduler") | |
pipe.scheduler = scheduler | |
# Set output size based on selection | |
image_size = (512, 512) if size == "Profile Picture" else (1024, 768) | |
# Generate image with specified quality and size | |
with torch.no_grad(): | |
image = pipe(prompt_text, num_inference_steps=quality, guidance_scale=7.5).images[0] | |
image = image.resize(image_size) # Resize image to fit selected dimensions | |
return np.array(image) # Return as NumPy array for Gradio | |
# Post-processing function for brightness and contrast | |
def adjust_brightness_contrast(image, brightness, contrast): | |
image = Image.fromarray(image.astype('uint8'), 'RGB') | |
image = ImageEnhance.Brightness(image).enhance(brightness) | |
image = ImageEnhance.Contrast(image).enhance(contrast) | |
return np.array(image) | |
# Warning function to show a message if the user selects a high value for quality | |
def show_warning(quality): | |
if quality > 80: | |
return "⚠️ High Quality: This setting may slow down generation and might not provide additional visual improvement. Consider using 50-80 steps for best results." | |
return "" | |
# Build Gradio Interface | |
with gr.Blocks() as demo: | |
gr.Markdown("# ✨ AI-Powered Wallpaper/Profile Picture Generator\n🖼️ A tool to generate and fine-tune AI-created wallpapers and profile pictures with adjustable styles and effects.") | |
gr.Markdown("⚠️ **Live effects and advanced prompt engineering coming soon! Disclaimer**: Results may not always be accurate or perfectly aligned with your prompt. Experiment with prompt adjustments and settings to get the best results.") | |
# Image Generation Section | |
with gr.Tab("Image Generator"): | |
gr.Markdown("## Generate an Image") | |
with gr.Row(): | |
with gr.Column(): | |
custom_prompt = gr.Textbox(label="Custom Prompt", placeholder="Describe your image (e.g., 'A forest at sunset')") | |
# Toggle for using additional categories | |
use_categories = gr.Checkbox(label="Enable Advanced Settings (Genre, Style, Theme, Lighting)", value=False) | |
# Additional categories, hidden by default and shown only if use_categories is checked | |
with gr.Accordion("Advanced Settings", open=False): | |
genre = gr.Dropdown(["Futuristic", "Nature", "Abstract", "Fantasy", "Sci-Fi", "Cyberpunk"], label="Genre") | |
style = gr.Dropdown(["Realistic", "Surreal", "Digital Art", "Cartoon", "Photorealistic"], label="Style") | |
theme = gr.Dropdown(["Landscape", "Portrait", "Abstract Patterns", "Architecture"], label="Theme") | |
lighting = gr.Dropdown(["Warm", "Cool", "Cinematic", "Soft", "Neon"], label="Lighting") | |
quality = gr.Slider(20, 150, value=80, step=10, label="Image Quality", info="Higher values yield more detail but take longer to generate.") | |
warning_message = gr.Markdown("") | |
# Scheduler selection with default option | |
scheduler_choice = gr.Dropdown( | |
[ | |
"Artistic & Imaginative (Euler Ancestral) - Recommended for creative scenes, moderate speed", | |
"Photo-Realistic (PNDM) - Best for realistic details, moderate speed", | |
"High-Definition & Fast (DDIM) - Good quality with fastest speed" | |
], | |
label="Artistic Style & Speed", | |
value="Artistic & Imaginative (Euler Ancestral) - Recommended for creative scenes, moderate speed" | |
) | |
size = gr.Dropdown(["Profile Picture", "Wallpaper"], label="Image Size", value="Profile Picture") | |
generate_button = gr.Button("Generate Image") | |
with gr.Column(): | |
generated_image = gr.Image(label="Generated Image", interactive=False) | |
# Display warning message for high-quality settings | |
quality.change(show_warning, inputs=[quality], outputs=warning_message) | |
# Bind the generate function to the generate button | |
generate_button.click( | |
fn=generate_image, | |
inputs=[custom_prompt, use_categories, genre, style, theme, lighting, scheduler_choice, quality, size], | |
outputs=generated_image | |
) | |
# Post-Processing Section | |
with gr.Tab("Edit Generated Image"): | |
gr.Markdown("## Adjust Brightness & Contrast") | |
with gr.Row(): | |
with gr.Column(): | |
brightness_slider = gr.Slider(0.5, 2.0, value=1.0, label="Brightness") | |
contrast_slider = gr.Slider(0.5, 2.0, value=1.0, label="Contrast") | |
apply_adjustments = gr.Button("Apply Adjustments") | |
with gr.Column(): | |
output_image = gr.Image(label="Adjusted Image", interactive=False) | |
# Bind the brightness and contrast adjustment function to the Apply Adjustments button | |
apply_adjustments.click( | |
fn=adjust_brightness_contrast, | |
inputs=[generated_image, brightness_slider, contrast_slider], | |
outputs=output_image | |
) | |
# Launch with a public shareable link | |
demo.launch(share=True) |