import random
import os
import uuid
from datetime import datetime
import gradio as gr
import numpy as np
import spaces
import torch
from diffusers import DiffusionPipeline
from PIL import Image
# Create permanent storage directory
SAVE_DIR = "saved_images" # Gradio will handle the persistence
if not os.path.exists(SAVE_DIR):
os.makedirs(SAVE_DIR, exist_ok=True)
device = "cuda" if torch.cuda.is_available() else "cpu"
repo_id = "black-forest-labs/FLUX.1-dev"
adapter_id = "openfree/flux-lora-korea-palace"
pipeline = DiffusionPipeline.from_pretrained(repo_id, torch_dtype=torch.bfloat16)
pipeline.load_lora_weights(adapter_id)
pipeline = pipeline.to(device)
MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 1024
def save_generated_image(image, prompt):
# Generate unique filename with timestamp
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
unique_id = str(uuid.uuid4())[:8]
filename = f"{timestamp}_{unique_id}.png"
filepath = os.path.join(SAVE_DIR, filename)
# Save the image
image.save(filepath)
# Save metadata
metadata_file = os.path.join(SAVE_DIR, "metadata.txt")
with open(metadata_file, "a", encoding="utf-8") as f:
f.write(f"{filename}|{prompt}|{timestamp}\n")
return filepath
def load_generated_images():
if not os.path.exists(SAVE_DIR):
return []
# Load all images from the directory
image_files = [os.path.join(SAVE_DIR, f) for f in os.listdir(SAVE_DIR)
if f.endswith(('.png', '.jpg', '.jpeg', '.webp'))]
# Sort by creation time (newest first)
image_files.sort(key=lambda x: os.path.getctime(x), reverse=True)
return image_files
def load_predefined_images():
# Return empty list since we're not using predefined images
return []
@spaces.GPU(duration=120)
def inference(
prompt: str,
seed: int,
randomize_seed: bool,
width: int,
height: int,
guidance_scale: float,
num_inference_steps: int,
lora_scale: float,
progress: gr.Progress = gr.Progress(track_tqdm=True),
):
if randomize_seed:
seed = random.randint(0, MAX_SEED)
generator = torch.Generator(device=device).manual_seed(seed)
image = pipeline(
prompt=prompt,
guidance_scale=guidance_scale,
num_inference_steps=num_inference_steps,
width=width,
height=height,
generator=generator,
joint_attention_kwargs={"scale": lora_scale},
).images[0]
# Save the generated image
filepath = save_generated_image(image, prompt)
# Return the image, seed, and updated gallery
return image, seed, load_generated_images()
examples = [
"Geunjeongjeon Hall of Gyeongbokgung Palace in spring, with cherry blossoms in full bloom. The majestic throne hall stands proudly against a backdrop of pink petals, its vibrant dancheong colors harmonizing with the spring flowers. Traditional stone markers and carefully manicured royal gardens frame the scene, while Mount Bugaksan rises majestically in the background. [trigger]",
"Summer sunrise at Geunjeongjeon Hall, Gyeongbokgung Palace. The golden morning light illuminates the grand wooden pillars and intricate roof tiles. Royal court musicians in traditional hanbok are preparing for the morning ceremony on the courtyard's stone steps, while the hall's reflection shimmers in the morning dew. [trigger]",
"Autumn twilight at Geunjeongjeon Hall. The royal throne hall is surrounded by maple and ginkgo trees in brilliant red and gold. The traditional blue and red dancheong paintings contrast beautifully with the warm autumn colors, while palace lanterns begin to glow in the approaching dusk. [trigger]",
"Winter scene at Geunjeongjeon Hall, with heavy snow blanketing the palace grounds. The hall's majestic double-tiered roof stands out against the pure white landscape, its dragon carvings dusted with snow. Frozen lotus ponds and snow-covered stone bridges create a serene winter wonderland. [trigger]",
"Geunjeongjeon Hall during a traditional royal ceremony under the full moon. Palace guards in historical uniforms stand at attention as lantern light dances across the ancient wooden structures. The moonlight casts dramatic shadows of the throne hall's curved eaves onto the frost-covered courtyard. [trigger]",
"Rainy season at Geunjeongjeon Hall. Mist shrouds the grand throne hall as summer rain falls softly on the ancient tiles. Water droplets cascade from the ornate roof dragons, while the wet stone steps gleam with reflected light from traditional palace lanterns. The rain creates a mystical atmosphere around the royal court architecture. [trigger]"
]
css = """
footer {
visibility: hidden;
}
"""
with gr.Blocks(theme="Yntec/HaleyCH_Theme_Orange", css=css, analytics_enabled=False) as demo:
gr.HTML(' KOREA PALACE STUDIO
')
gr.HTML('')
with gr.Tabs() as tabs:
with gr.Tab("Generation"):
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",
container=False,
)
run_button = gr.Button("Run", scale=0)
result = gr.Image(label="Result", show_label=False)
with gr.Accordion("Advanced Settings", open=False):
seed = gr.Slider(
label="Seed",
minimum=0,
maximum=MAX_SEED,
step=1,
value=42,
)
randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
with gr.Row():
width = gr.Slider(
label="Width",
minimum=256,
maximum=MAX_IMAGE_SIZE,
step=32,
value=1024,
)
height = gr.Slider(
label="Height",
minimum=256,
maximum=MAX_IMAGE_SIZE,
step=32,
value=768,
)
with gr.Row():
guidance_scale = gr.Slider(
label="Guidance scale",
minimum=0.0,
maximum=10.0,
step=0.1,
value=3.5,
)
num_inference_steps = gr.Slider(
label="Number of inference steps",
minimum=1,
maximum=50,
step=1,
value=30,
)
lora_scale = gr.Slider(
label="LoRA scale",
minimum=0.0,
maximum=1.0,
step=0.1,
value=1.0,
)
gr.Examples(
examples=examples,
inputs=[prompt],
outputs=[result, seed],
)
with gr.Tab("Gallery"):
gallery_header = gr.Markdown("### Generated Images Gallery")
generated_gallery = gr.Gallery(
label="Generated Images",
columns=6,
show_label=False,
value=load_generated_images(),
elem_id="generated_gallery",
height="auto"
)
refresh_btn = gr.Button("🔄 Refresh Gallery")
# Event handlers
def refresh_gallery():
return load_generated_images()
refresh_btn.click(
fn=refresh_gallery,
inputs=None,
outputs=generated_gallery,
)
gr.on(
triggers=[run_button.click, prompt.submit],
fn=inference,
inputs=[
prompt,
seed,
randomize_seed,
width,
height,
guidance_scale,
num_inference_steps,
lora_scale,
],
outputs=[result, seed, generated_gallery],
)
demo.queue()
demo.launch()