app.py

import spaces
import gradio as gr
import torch
from PIL import Image
from diffusers import DiffusionPipeline
import random

torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
torch.backends.cuda.matmul.allow_tf32 = True

# Initialize the base model and specific LoRA
base_model = "black-forest-labs/FLUX.1-dev"
pipe = DiffusionPipeline.from_pretrained(base_model, torch_dtype=torch.bfloat16)

lora_repo = "kaytoo2022/jguan_35-flux-2"
trigger_word = "" # Leave trigger_word blank if not used.
pipe.load_lora_weights(lora_repo, adapter_name='jguan')

# ghibsky
lora_repo_2 = "aleksa-codes/flux-ghibsky-illustration"
pipe.load_lora_weights(lora_repo_2, adapter_name='lora_2')

pipe.set_adapters(["jguan", "lora_2"], adapter_weights=[0.85, 0.0])

pipe.to("cuda")

MAX_SEED = 2**32-1

@spaces.GPU()
def run_lora(prompt, cfg_scale, steps, randomize_seed, seed, width, height, lora_scale, lora_scale_2, progress=gr.Progress(track_tqdm=True)):
 # Set random seed for reproducibility
 if randomize_seed:
 seed = random.randint(0, MAX_SEED)
 generator = torch.Generator(device="cuda").manual_seed(seed)

 pipe.set_adapters(["jguan", "lora_2"], adapter_weights=[lora_scale, lora_scale_2])

 # Update progress bar (0% saat mulai)
 progress(0, "Starting image generation...")

 # Generate image with progress updates
 for i in range(1, steps + 1):
 # Simulate the processing step (in a real scenario, you would integrate this with your image generation process)
 if i % (steps // 10) == 0: # Update every 10% of the steps
 progress(i / steps * 100, f"Processing step {i} of {steps}...")

 # Generate image using the pipeline
 image = pipe(
 prompt=f"{prompt} {trigger_word}",
 num_inference_steps=steps,
 guidance_scale=cfg_scale,
 width=width,
 height=height,
 generator=generator,
 # joint_attention_kwargs={"scale": lora_scale},
 ).images[0]

 # Final update (100%)
 progress(100, "Completed!")

 yield image, seed

# Example cached image and settings
example_image_path = "bella_space.jpeg" # Replace with the actual path to the example image
example_prompt = """A portrait picture of [Just1nGu4n] in an astronaut outfit. Planets are visible in the background"""
example_cfg_scale = 3.2
example_steps = 32
example_width = 1152
example_height = 896
example_seed = 3981632454
example_lora_scale = 0.85
example_lora_scale_2 = 0.0

def load_example():
 # Load example image from file
 # example_image = Image.open(example_image_path)
 return example_prompt, example_cfg_scale, example_steps, True, example_seed, example_width, example_height, example_lora_scale, example_lora_scale_2, None

with gr.Blocks() as app:
 gr.Markdown("# Flux Lora Image Generator")
 with gr.Row():
 with gr.Column(scale=3):
 prompt = gr.TextArea(label="Prompt", placeholder="Type a prompt", lines=5)
 generate_button = gr.Button("Generate")
 cfg_scale = gr.Slider(label="CFG Scale", minimum=1, maximum=20, step=0.5, value=example_cfg_scale)
 steps = gr.Slider(label="Steps", minimum=1, maximum=100, step=1, value=example_steps)
 width = gr.Slider(label="Width", minimum=256, maximum=1536, step=64, value=example_width)
 height = gr.Slider(label="Height", minimum=256, maximum=1536, step=64, value=example_height)
 randomize_seed = gr.Checkbox(True, label="Randomize seed")
 seed = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=example_seed)
 lora_scale = gr.Slider(label="LoRA Scale", minimum=0, maximum=1, step=0.01, value=example_lora_scale)
 lora_scale_2 = gr.Slider(label="LoRA Scale (GhibSky)", minimum=0, maximum=1, step=0.01, value=example_lora_scale_2)
 with gr.Column(scale=1):
 result = gr.Image(label="Generated Image")
 gr.Markdown("Generate images using Flux and a text prompt.\nUse `[Just1nGu4n]` in the prompt to activate the LoRA adapter.\n[[non-commercial license, Flux.1 Dev](https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md)]")

 # Automatically load example data and image when the interface is launched
 app.load(load_example, inputs=[], outputs=[prompt, cfg_scale, steps, randomize_seed, seed, width, height, lora_scale, lora_scale_2, result])
 
 generate_button.click(
 run_lora,
 inputs=[prompt, cfg_scale, steps, randomize_seed, seed, width, height, lora_scale, lora_scale_2],
 outputs=[result, seed]
 )

app.queue()
app.launch()